Briefing · 2026-06-29

Your briefing

86 ranked ·

Today's dispatch

Filed · 86 ranked

All Read Saved
  1. 94 score Twitter/X · Must read · 3 min Three firms—Samsung, SK Hynix, and Micron—make almost all the world’s DRAM; a… Three firms—Samsung, SK Hynix, and Micron—make almost all the world’s DRAM; a gigabyte of HBM for AI uses roughly 3–4× the wafer capacity of a gigabyte of DDR5, so HBM disproportionately drains the shared wafer pool.
  2. 94 score Twitter/X · Must read · 5 min NVIDIA DGX Spark hardware NVIDIA DGX Spark hardware: GB10 (Grace Blackwell) SoC, 128 GB unified CPU+GPU memory, ConnectX‑7 200GbE QSFP ports, 4 TB encrypted NVMe M.2, 1 PFLOPS FP4 performance; MSRP listed at $4,699/unit.
  3. 92 score Twitter/X · Must read · 1 min TSMC and Winbond will collaborate on next-generation Wafer-on-Wafer (WoW) 3D… TSMC and Winbond will collaborate on next-generation Wafer-on-Wafer (WoW) 3D stacking: Winbond will supply DRAM wafers and TSMC will stack them with its logic wafers to produce AI chips.
  4. 92 score Twitter/X · Must read · 1 min Oracle's SEC filing attributes mass layoffs to AI Oracle's SEC filing attributes mass layoffs to AI: 21,000 jobs cut (13% of workforce) with a $1.8 billion restructuring charge.
  5. 92 score Twitter/X · Must read · 3 min DeepSeek open-sourced DSpark/DeepSpec on 2026-06-28 (post by @heyshrutimishra)… DeepSeek open-sourced DSpark/DeepSpec on 2026-06-28 (post by @heyshrutimishra), releasing the V4 Pro DSpark checkpoint on Hugging Face and training code on GitHub under an MIT license; the system is already running in production on millions of users and available as a free download.
  6. 88 score Twitter/X · Worth reading · 2 min As of reports on June 26, 2026, industry inventories of high‑purity CO2 have… As of reports on June 26, 2026, industry inventories of high‑purity CO2 have fallen below the typical combined one‑month buffer (manufacturers and suppliers each normally hold ~2 weeks), prompting intensified procurement by chipmakers.
  7. 80 score Twitter/X · Worth reading · 1 min La IA está generando una demanda de energía que las redes tradicionales no pueden… La IA está generando una demanda de energía que las redes tradicionales no pueden satisfacer; empresas como xAI y OpenAI planean centros de datos que consumirán cientos de megavatios cada uno solo para entrenar modelos.
  8. 78 score Twitter/X · Worth reading · 1 min Gavin S. Gavin S. Baker said on All‑In at 1:04:06: "DRAM is the most important bottleneck," and Elon is focusing the Terafab on memory (DRAM), not lasers, capacitors, power semis, NAND, or HDDs.
  9. 72 score Twitter/X · Worth reading · 1 min Humanoid production hit 15,000 units Humanoid production hit 15,000 units: two years to produce the first 1,000 and three months to produce the last 5,000, indicating a steep adoption/production ramp.
  10. 62 score Twitter/X · Worth reading · 1 min España tiene uno de los mejores recursos solares y eólicos de Europa pero, según… España tiene uno de los mejores recursos solares y eólicos de Europa pero, según @martinvars (publicado 2026-06-28), paga una de las electricidades más caras del continente.
  11. 62 score Twitter/X · Worth reading · 1 min Micron (MU) could become more important than Nvidia for AI inference because… Micron (MU) could become more important than Nvidia for AI inference because inference ROI depends more on memory than on additional GPUs, per @jukan05 (2026-06-28).
  12. 62 score Twitter/X · Worth reading · 2 min Rubycon (Japan's third-largest aluminum capacitor maker) issued a price-increase… Rubycon (Japan's third-largest aluminum capacitor maker) issued a price-increase notice effective August 1, 2026 covering aluminum electrolytic capacitors, solid aluminum capacitors, and film capacitors and warned a second increase may follow; the company cited high crude oil, rising metal prices (aluminum, copper, tin), supplier hikes across raw/auxiliary materials and logistics.
  13. 60 score Twitter/X · Worth reading · 1 min Gary Marcus (2026-06-28) wrote “hard to see how anyone makes much money from all… Gary Marcus (2026-06-28) wrote “hard to see how anyone makes much money from all this in the long run,” likening the AI sector to the airline industry with “very small margins and big expenses” and calling the “pouring trillions in” probably unwise.
  14. 58 score Twitter/X · Worth reading · 1 min Friend runs an AI SaaS at $70k MRR with ~30% profit margin (~$21k/month) after… Friend runs an AI SaaS at $70k MRR with ~30% profit margin (~$21k/month) after paying for infrastructure, AI agents, and Google/Meta ads.
  15. 58 score Twitter/X · Worth reading · 1 min John Loeber (2026-06-28) asserts all future programs will be formally verifiable… John Loeber (2026-06-28) asserts all future programs will be formally verifiable, claiming formal verification is necessary for cybersecurity and to enable codebases 100x+ larger as AI drives software scale.
  16. 58 score Twitter/X · Worth reading · 1 min Defaulting engineers to open-weight models (GLM 5.2 and Kimi 2.7) via an LLM… Defaulting engineers to open-weight models (GLM 5.2 and Kimi 2.7) via an LLM gateway lowered cost pressure—91% of employees never hit usage caps, so the team moved to cheaper defaults instead of lowering caps or adding friction.
  17. 48 score Twitter/X · Quick skim · 1 min El 28 de junio de 2026 @martinvars afirma que España volverá a registrar más… El 28 de junio de 2026 @martinvars afirma que España volverá a registrar más muertes que nacimientos en 2026 y que la población activa que sostiene a los jubilados se está achicando.
  18. 48 score Twitter/X · Quick skim · 1 min @bcherny (2026-06-28) identifies five archetypes on the Claude Code team… @bcherny (2026-06-28) identifies five archetypes on the Claude Code team: Prototyper (many new ideas, most don't ship), Builder (turns prototypes into production), Sweeper (cleans UI/code, unships, optimizes), Grower (iterates to improve PMF), and Maintainer (owns mature system reliability/efficiency).
  19. 46 score Twitter/X · Quick skim · 1 min Anton Osika (@antonosika) says his friend Kristian Rønn (@Kristian_Ronn) is… Anton Osika (@antonosika) says his friend Kristian Rønn (@Kristian_Ronn) is building cryptographic verification to confirm superintelligence is not misused, enabling both open-source projects and commercial labs to prove compliance without exposing model secrets or private data.
  20. 46 score Twitter/X · Quick skim · 1 min Black Spark claims it infiltrated the Alabuga special economic zone in Tatarstan… Black Spark claims it infiltrated the Alabuga special economic zone in Tatarstan for several months, exfiltrated databases on personnel involved in assembling Geran-2 and Geran-3 drones, and hacked Alabuga’s website (archived copy dated 27 June 2026).
  21. 42 score Twitter/X · Quick skim · 1 min Levie (2026-06-28) claims 'mythos level' cybersecurity models will soon be open… Levie (2026-06-28) claims 'mythos level' cybersecurity models will soon be open and available to anyone, which will enable alternative tech stacks that shift economic value and control away from the US tech stack.
  22. 42 score Twitter/X · Quick skim · 1 min Ford rehired, newly hired, or promoted 350 experienced engineers after AI systems… Ford rehired, newly hired, or promoted 350 experienced engineers after AI systems and adjusted design requirements failed to maintain product quality, according to Ford's VP of hardware engineering.
  23. 42 score Twitter/X · Quick skim · 2 min On June 23, 2026 the KOSPI plunged 9.99% (910 points) to 8,203.84 after Samsung… On June 23, 2026 the KOSPI plunged 9.99% (910 points) to 8,203.84 after Samsung (-12.3%) and SK Hynix (-12.5%) triggered a 20‑minute circuit breaker; the index had closed at an all‑time high of 9,114.55 the previous day after tripling from ~3,000 a year earlier on AI‑chip strength.
  24. 40 score Twitter/X · Quick skim · 1 min Brad began at O'Reilly in July 1996 as a 17-year-old parts specialist in Wagoner… Brad began at O'Reilly in July 1996 as a 17-year-old parts specialist in Wagoner, Oklahoma, worked every store role before age 20 (counterperson, night manager, assistant store manager), and became the company's fourth CEO in 2024 after 28 years.
  25. 39 score Twitter/X · Quick skim · 1 min On 2026-06-28 @cramforce warned there is a short-time window where the defensive… On 2026-06-28 @cramforce warned there is a short-time window where the defensive play (making source code available) is superior to the offensive play (black-box pen testing).
  26. 35 score Twitter/X · Quick skim · 1 min In June 2026 the 'Mag 7' hyperscalers faced 'brutal' pressure, but @jvisserlabs… In June 2026 the 'Mag 7' hyperscalers faced 'brutal' pressure, but @jvisserlabs asserts this is rotation, not collapse: capital is rotating into healthcare, small caps, equal-weight, AI infrastructure receivers, memory, biotech, and agentic commerce.
  27. 35 score Twitter/X · Quick skim · 1 min The author claims Anthropic's CEO called for regulation to shift legal and… The author claims Anthropic's CEO called for regulation to shift legal and political responsibility away from the company if 'Mythos 5' later ends up in China's hands and causes damage.
  28. 35 score Twitter/X · Quick skim · 1 min Josh Rauh (quoting Chamath Palihapitiya) claims California’s Property Seizure Act… Josh Rauh (quoting Chamath Palihapitiya) claims California’s Property Seizure Act — labeled the "Billionaire Tax" — would require handing over 5% in cash of all tangible and intangible property every year, resulting in confiscation of assets, destroyed 401(k)s, and lost pensions.
  29. 35 score Twitter/X · Quick skim · 1 min Five-level AI maturity framework Five-level AI maturity framework: Level 1 — AI as a thought partner (bouncing strategy off ChatGPT); Level 2 — AI as an assistant (drafting, summarizing); Level 3 — AI as a team (running creatives, handling first drafts, building reports); Level 4 — AI as the system (entire workflows rebuilt); Level 5 — agents managing agents with compounding feedback loops.
  30. 35 score Twitter/X · Quick skim · 1 min On 2026-06-28 @kimmonismus flagged claims that “a new model from Zhipu AI” is at… On 2026-06-28 @kimmonismus flagged claims that “a new model from Zhipu AI” is at least as strong as Fable5 for cybersecurity, but their search found only a Wall Street Journal article naming GLM-5.2 as a recently released model.
  31. 35 score Twitter/X · Quick skim · 1 min Argentina closed its first sustained fiscal surplus in more than a decade… Argentina closed its first sustained fiscal surplus in more than a decade, according to the post dated 2026-06-28, a result the author credits to Milei's government.
  32. 35 score Twitter/X · Quick skim · 1 min At age 5 in Villa Gesell, Argentina, the author got lost on the beach; a very… At age 5 in Villa Gesell, Argentina, the author got lost on the beach; a very tall man put him on his shoulders, a crowd formed and clapped, and his parents appeared instantly—described as a spontaneous local system to reunite lost children.
  33. 35 score Twitter/X · Quick skim · 1 min @martinvars (publicado 2026-06-28) afirma que la principal aportación de la IA en… @martinvars (publicado 2026-06-28) afirma que la principal aportación de la IA en medicina es resolver el problema de acceso: "millones de personas" viven a horas del especialista más cercano y esperan "meses" por consultas necesarias.
  34. 35 score Twitter/X · Quick skim · 1 min A clear glass disc spun at 3,000 RPM (50 rev/s) in front of a lens prevents… A clear glass disc spun at 3,000 RPM (50 rev/s) in front of a lens prevents raindrops: drops near the edge experience roughly 500 g and are torn into mist by shear; only a tiny drop at the axis (zero tangential speed) can remain and is too small to register on camera.
  35. 35 score Twitter/X · Quick skim · 1 min @martinvars, fundador con 30 años de experiencia (publicación 2026-06-28), afirma… @martinvars, fundador con 30 años de experiencia (publicación 2026-06-28), afirma que la IA da una ventaja enorme al fundador que entiende su negocio en detalle.
  36. 35 score Twitter/X · Quick skim · 1 min On 2026-06-28 @deanwball argued the most important legal questions in AI now… On 2026-06-28 @deanwball argued the most important legal questions in AI now center on the First Amendment, asserting creation, distribution, and use of 'frontier AI' (large language models) should be treated as protected expression.
  37. 35 score Twitter/X · Quick skim · 1 min @deanwball (2026-06-28) says the First Amendment ("1A") will place real limits on… @deanwball (2026-06-28) says the First Amendment ("1A") will place real limits on state intervention in AI, but argues those limits do not make AI regulation impossible; he warns 1A could "bite" if policymakers try to regulate alignment in high levels of detail.
  38. 35 score Twitter/X · Quick skim · 1 min Austria publicly invited Anthropic to 'strategically establish and participate'… Austria publicly invited Anthropic to 'strategically establish and participate' in the EU, promising 'legal certainty, market access, capital and a set of values' (quote from a letter shared by Andrew Curran in the 2026-06-28 post).
  39. 35 score Twitter/X · Quick skim · 1 min Dean W. Dean W. Ball (tweeted 2026-06-28) says the most important legal questions in AI are First Amendment issues: what fact patterns demonstrate that creation, distribution, and use of 'frontier AI' qualify as protected expression.
  40. 35 score Twitter/X · Quick skim · 1 min African elephants carry about 2,000 olfactory receptor genes versus 396 in humans… African elephants carry about 2,000 olfactory receptor genes versus 396 in humans and ~800 in bloodhounds, meaning an elephant's trunk has roughly five times the scent hardware of a human and twice that of a dog.
  41. 35 score Twitter/X · Quick skim · 2 min Laurel built a 200+ row spreadsheet that breaks every function into Categories… Laurel built a 200+ row spreadsheet that breaks every function into Categories and Tasks (Product splits into five: Feature Work, Strategy & Roadmap, Customer Research, Product Analytics, Voice of Customer) and adds a one-line description for each task plus an explicit Negative Examples column that defines wrong versions of the work (e.g., Competitive & Market Analysis negative: ongoing tracking of competitor release notes).
  42. 35 score Twitter/X · Quick skim · 1 min Used Qwen3.6-27B-MTP-GGUF on four 64×64 images Used Qwen3.6-27B-MTP-GGUF on four 64×64 images: three processed plain and one with MTP, encapsulated by the claim “An image is worth 16x16 words.”
  43. 35 score Twitter/X · Quick skim · 1 min Ramp's product-design workflow now begins in an LLM, includes user calls and… Ramp's product-design workflow now begins in an LLM, includes user calls and AI-generated prototypes, then moves into Figma and finally iterates against live behavior—relegating Figma to 'step 4 of 5'.
  44. 35 score Twitter/X · Quick skim · 1 min Scott Wu claims AI will so transform work that future generations won't even… Scott Wu claims AI will so transform work that future generations won't even recognize today's jobs as 'work', contrasting modern office tasks — 'pushing buttons' and 'sitting in a room and talking' (meetings) — with ancestral manual labor like farming and making clothes by hand.
  45. 35 score Twitter/X · Quick skim · 1 min @ivanfioravanti says they want to work at SpaceX specifically to try, play with… @ivanfioravanti says they want to work at SpaceX specifically to try, play with, and interact with the latest models on SpaceX hardware.
  46. 35 score Twitter/X · Quick skim · 1 min Dan McAteer predicts GPT-5.6 will ship and Fable will return the week of June 28… Dan McAteer predicts GPT-5.6 will ship and Fable will return the week of June 28, 2026, saying Axios reports the government may lift export controls on Fable “as soon as this week” and is unlikely to favor Anthropic over OpenAI.
  47. 35 score Twitter/X · Quick skim · 1 min Austria (and the UK earlier) has formally invited Anthropic to establish a… Austria (and the UK earlier) has formally invited Anthropic to establish a presence in the EU; a quoted letter (Andrew Curran) urges joint exploration to 'strategically establish and participate of Anthropic within the European Union' and praises Anthropic's safety-first values.
  48. 35 score Twitter/X · Quick skim · 2 min On 2026-06-28, OpenAI previewed GPT-5.6 in three tiers—Sol (frontier), Terra, and… On 2026-06-28, OpenAI previewed GPT-5.6 in three tiers—Sol (frontier), Terra, and Luna—and at the U.S. government's request all three will begin as limited previews to approved partners rather than immediate public releases; OpenAI positions Sol as its most capable model and strongest for cybersecurity.
  49. 35 score Twitter/X · Quick skim · 2 min Corey Ganim (post 2026-06-28) proposes five 'Second Brain as a Service'… Corey Ganim (post 2026-06-28) proposes five 'Second Brain as a Service' verticals: Law firm case intake, Tax firm client-doc, B2B sales call, Restaurant franchise SOP, and Tutoring center curriculum — each stores specific artifacts (e.g., intake notes and jurisdiction rules; client checklists and deadline rules; call transcripts and objections) to automate prep and follow-ups.
  50. 35 score Twitter/X · Quick skim · 1 min Chamath (@chamath) posted on 2026-06-28 that California’s Property Seizure… Chamath (@chamath) posted on 2026-06-28 that California’s Property Seizure Act—marketed as “The Billionaire Tax”—is now on the ballot and links a 4-minute Hoover Institution video (piped.video/x6k4W5Qzg8U) explaining it.
  51. 35 score Twitter/X · Quick skim · 1 min Shruti (@heyshrutimishra) says her keyboard usage is down to 20% and she now… Shruti (@heyshrutimishra) says her keyboard usage is down to 20% and she now dictates almost everything, claiming she moves through work 5x faster.
  52. 35 score Twitter/X · Quick skim · 1 min On 2026-06-28 @jasonc_nc argued a suggested plan would On 2026-06-28 @jasonc_nc argued a suggested plan would: use state power to bankrupt current owners; deploy public funds to selected new owners; have a government agency provide below-cost management; ignore how to fund lost property-tax revenue; and thereby halt new construction while expanding the population needing subsidies.
  53. 35 score Twitter/X · Quick skim · 1 min @bscholl (published 2026-06-28) asserts that a common business evil is using… @bscholl (published 2026-06-28) asserts that a common business evil is using government power under false pretenses to hamstring competition.
  54. 35 score Twitter/X · Quick skim · 1 min Jason C (@jasonc_nc) posted on 2026-06-28 that "virtually every great… Jason C (@jasonc_nc) posted on 2026-06-28 that "virtually every great neighborhood" was created before the American Planning Association’s predecessors existed and claims that as planning became more controlling the quality of cities "steadily got worse."
  55. 35 score Twitter/X · Quick skim · 1 min GLM-5.2, a Chinese open-weight large language model released by 2026-06-28, is… GLM-5.2, a Chinese open-weight large language model released by 2026-06-28, is claimed to match the performance of currently available models from OpenAI and Anthropic.
  56. 35 score Twitter/X · Quick skim · 2 min @levelsio has coded almost exclusively on his VPS with Claude Code for almost a… @levelsio has coded almost exclusively on his VPS with Claude Code for almost a year, live-editing his production server; it only 'messed up' twice in 12 months, causing about 10 seconds of downtime each time.
  57. 35 score Twitter/X · Quick skim · 1 min Jason (@jasonc_nc) (posted 2026-06-28) argues fire departments should rebrand as… Jason (@jasonc_nc) (posted 2026-06-28) argues fire departments should rebrand as 'emergency services departments' with a tiny ("really tiny") slice dedicated to fire to adopt a broader operational perspective.
  58. 35 score Twitter/X · Quick skim · 1 min Sarvesh Shrivastava (@bloggersarvesh) claims on 2026-06-28 that using Claude… Sarvesh Shrivastava (@bloggersarvesh) claims on 2026-06-28 that using Claude (Claude Cowork) for $20/month can 'run like a $10k SEO team' if users stop treating it like a beginner.
  59. 35 score Twitter/X · Quick skim · 1 min Created a dynamic, auto-updating chart (levels.io/bitcoin-4-year-cyc…) that… Created a dynamic, auto-updating chart (levels.io/bitcoin-4-year-cyc…) that visualizes Bitcoin's repeating 4-year cycles and keeps adding new cycles over time; the chart marks October 31, 2026 as the start of the next cycle and theoretically the cycle bottom.
  60. 35 score Twitter/X · Quick skim · 1 min Codex usage increased 6x since February (2026), now exceeding 5 million weekly… Codex usage increased 6x since February (2026), now exceeding 5 million weekly active users; nearly 100% of OpenAI employees reportedly use the Codex desktop app regularly.
  61. 35 score Twitter/X · Quick skim · 3 min Chubby♨️ (@kimmonismus) asserts that Dario Amodei’s alleged “fearmongering” was… Chubby♨️ (@kimmonismus) asserts that Dario Amodei’s alleged “fearmongering” was not the reason the U.S. embargoed Fable 5 and GPT-5.6 and calls that attribution mistaken.
  62. 35 score Twitter/X · Quick skim · 1 min FreeLLMAPI (GitHub repo by tashfeenahmed) aggregates 11 free LLM… FreeLLMAPI (GitHub repo by tashfeenahmed) aggregates 11 free LLM providers—Google, Groq, Mistral, OpenRouter, GitHub Models, Cohere, Cloudflare, HuggingFace, Z AI, Ollama, and Kimi—and claims the combined free tiers provide about 1.7 billion tokens per month.
  63. 35 score Twitter/X · Quick skim · 1 min In 1955 Sony (only months old) showed the TR-55 radio; Bulova offered to buy… In 1955 Sony (only months old) showed the TR-55 radio; Bulova offered to buy 100,000 units on the condition Sony's logo be removed — cofounder Akio Morita refused, overruling partner Masaru Ibuka and the Tokyo board who wanted to accept the order.
  64. 35 score Twitter/X · Quick skim · 3 min Elon Musk turns 55 on 2026-06-28; the author frames his lifetime of projects as… Elon Musk turns 55 on 2026-06-28; the author frames his lifetime of projects as driven by humanity-first motives rather than money and says his plans ‘span decades’.
  65. 35 score Twitter/X · Quick skim · 1 min swyx asked what the AIEWF "Stress Curve" looked like, defining it explicitly as… swyx asked what the AIEWF "Stress Curve" looked like, defining it explicitly as the Gini coefficient over time from 0 (ticket launch) to sold out.
  66. 35 score Twitter/X · Quick skim · 1 min @0xd1namit (published 2026-06-28) claims users demand features Polymarket will… @0xd1namit (published 2026-06-28) claims users demand features Polymarket will never add—examples include copytrading, trader tracking, and extensive analytics.
  67. 35 score Twitter/X · Quick skim · 1 min Liebig's 1835 method uses ammoniated silver nitrate (Ag⁺ suspended in ammonia)… Liebig's 1835 method uses ammoniated silver nitrate (Ag⁺ suspended in ammonia) plus a reducing sugar (e.g., glucose); the sugar's aldehyde donates electrons so each Ag⁺ gains one electron and precipitates as metallic silver while the sugar oxidizes to gluconic acid.
  68. 35 score Twitter/X · Quick skim · 1 min The 2026 World's Fair is "completely sold out" (post published 2026-06-28) and is… The 2026 World's Fair is "completely sold out" (post published 2026-06-28) and is billed as "the largest AI industry expo on earth"; the Leadership track for CTOs & VP AI's, tomorrow's workshops, and ALL late-bird tickets are sold out.
  69. 35 score Twitter/X · Quick skim · 2 min Yishan (@yishan) re-ran the paper authors' open-sourced evaluation framework… Yishan (@yishan) re-ran the paper authors' open-sourced evaluation framework (github.com/aiden-ygu/health-...) and published his reproduction and results at github.com/ywong137/health-a..., successfully reproducing tests on the public datasets the paper included.
  70. 35 score Twitter/X · Quick skim · 1 min On 2026-06-28, Fox News Media issued an apology after Kevin O'Leary retracted… On 2026-06-28, Fox News Media issued an apology after Kevin O'Leary retracted claims about opponents of his planned Utah data center and stated the network is aware of no evidence they were funded by or acting in coordination with the Chinese Communist Party.
  71. 35 score Twitter/X · Quick skim · 2 min cryptopunk7213: Over the next 6 months, the most valuable capabilities will be… cryptopunk7213: Over the next 6 months, the most valuable capabilities will be per-task model choice, multi-model routing/harnesses, caching to reduce token spend, and AI-cloud management of memory/databases; the bottleneck is shifting from LLM capacity to aggregation.
  72. 35 score Twitter/X · Quick skim · 1 min Daniel Mac (@daniel_mac8) predicts GPT-5.6 will ship and Fable will return the… Daniel Mac (@daniel_mac8) predicts GPT-5.6 will ship and Fable will return the week of 2026-06-28; Axios reportedly says the U.S. government may lift export controls on Fable as soon as that week and is unlikely to favor Anthropic over OpenAI.
  73. 35 score Twitter/X · Quick skim · 1 min ODNI is drafting an unclassified report, mandated by the FY2026 Intelligence… ODNI is drafting an unclassified report, mandated by the FY2026 Intelligence Authorization Act and due before December 2026, intended to disclose personal wealth, offshore accounts, and global real estate of Xi Jinping and the 25-member Politburo.
  74. 35 score Twitter/X · Quick skim · 1 min Paris Fire Brigade (serving a metro of 10M) posts average fire response of 12… Paris Fire Brigade (serving a metro of 10M) posts average fire response of 12 minutes and EMS 7 minutes; Houston reports 90th‑percentile times of 13 minutes for fire and 15 minutes for EMS.
  75. 35 score Twitter/X · Quick skim · 1 min Arvind Krishna (IBM CEO) said in a podcast that “Year one of enterprise AI is a… Arvind Krishna (IBM CEO) said in a podcast that “Year one of enterprise AI is a net loss,” causing 80% of companies to currently “look like they have failed,” but “Year two is precisely when the 10x return kicks in.”
  76. 35 score Twitter/X · Quick skim · 1 min Andrew Ambrosino (@ajambrosino) leads the Codex desktop app team at OpenAI; Codex… Andrew Ambrosino (@ajambrosino) leads the Codex desktop app team at OpenAI; Codex usage has grown 6x since February to over 5 million weekly active users, and nearly 100% of OpenAI employees use the Codex app regularly (not just engineers).
  77. 35 score Twitter/X · Quick skim · 1 min Andrew Ambrosino leads the Codex desktop app team at OpenAI and states his… Andrew Ambrosino leads the Codex desktop app team at OpenAI and states his mission is to build “the best desktop app that has ever existed, full stop.”
  78. 35 score Twitter/X · Quick skim · 1 min On 2026-06-28 @kimmonismus asserted that “the open internet was the training… On 2026-06-28 @kimmonismus asserted that “the open internet was the training ground for modern AI,” citing “billions of texts, forum posts, code repositories, Wikipedia articles, scientific papers, blogs, and discussions” as the collective data foundation for current models.
  79. 35 score Twitter/X · Quick skim · 1 min Aakash Gupta: $60,000 for a private cross‑country flight buys 60 first‑class… Aakash Gupta: $60,000 for a private cross‑country flight buys 60 first‑class coast‑to‑coast tickets; on a 5‑hour flight the total time difference is ~45 minutes, implying about $1,333 saved per minute by flying private.
  80. 35 score Twitter/X · Quick skim · 2 min Mixture of Agents (MoA) runs several LLMs in parallel and an aggregator… Mixture of Agents (MoA) runs several LLMs in parallel and an aggregator synthesizes their answers; Together AI's MoA scored 65.1% on AlpacaEval 2.0 vs GPT-4o's 57.5% using only open models, and the exam used to claim superiority was not released and was authored by the lab.
  81. 16 score Twitter/X · Less notable · 1 min @ianmiles (2026-06-28) asserts the British public has voted AGAINST more… @ianmiles (2026-06-28) asserts the British public has voted AGAINST more migration "election after election, for decades," claiming politicians repeatedly ignore this mandate.
  82. 16 score Twitter/X · Less notable · 1 min The viral claim that “9 million women were burned as witches” is false; modern… The viral claim that “9 million women were burned as witches” is false; modern historians estimate about 40,000–60,000 executions across three centuries in Europe.
  83. 12 score Twitter/X · Less notable · 1 min On 2026-06-28 @bswud (citing @vpostrel) argued modernism is more popular in… On 2026-06-28 @bswud (citing @vpostrel) argued modernism is more popular in California and Chicago than in Europe or the US East Coast.
  84. 12 score Twitter/X · Less notable · 1 min At age 40, studios offered Meryl Streep three "witch" roles in a single year; by… At age 40, studios offered Meryl Streep three "witch" roles in a single year; by then she already held two Oscars (Kramer vs. Kramer at 30, Sophie's Choice at 33), signaling industry typecasting of women over 40 as crones or villains by 1989.
  85. 12 score Twitter/X · Less notable · 1 min Elon Musk (tweet, 2026-06-28) asserts critics who claim “millions” died cannot… Elon Musk (tweet, 2026-06-28) asserts critics who claim “millions” died cannot cite a single named person who allegedly died.
  86. 7 score Twitter/X · Less notable · 2 min In 1972 off California an orca bit a surfer, took one taste, released him, and… In 1972 off California an orca bit a surfer, took one taste, released him, and the man survived after receiving over 100 stitches—used by the author as direct evidence that orcas reject unfamiliar prey.
Twitter/X · 3 min Signal

Three firms—Samsung, SK Hynix, and Micron—make almost all the world’s DRAM; a…

Three firms—Samsung, SK Hynix, and Micron—make almost all the world’s DRAM; a gigabyte of HBM for AI uses roughly 3–4× the wafer capacity of a gigabyte of DDR5, so HBM disproportionately drains the shared wafer pool.

DRAM supply dynamics in 2026 are being reshaped by AI demand: three producers (Samsung, SK Hynix, Micron) share a finite wafer pool and HBM—used in AI servers—consumes roughly 3–4× the wafer capacity per GB of standard DDR5. That imbalance has pushed HBM to ~30% of DRAM revenue while only supplying ~8% of bits, contributed to consumer/laptop RAM doubling in price this year, and left SK Hynix with essentially no sellable HBM capacity through 2026. Micron has reallocated wafers (walking away from Crucial) and locked multi‑year, prepaid contracts covering ~50% of revenue at floor prices above prior cycle margins, which the author argues breaks the historical commodity cycle. With HBM4 increasing stacking (12→16 dies) and buyers fronting cash, memory looks more like a constrained toll road for AI capex than a tradable commodity, a thesis the market may not yet have fully priced.

HBM already generates about 30% of DRAM revenue while shipping only ~8% of the bits, and that wafer diversion helped laptop/consumer RAM roughly double in price in 2026 as server/AI demand surged.
Open reader
J/K nav · S save · ? shortcuts
Must read

Start here. These are the items with the strongest reader value today.

5 items
2 Twitter/X 2026-06-28 5 min read
Open

NVIDIA DGX Spark hardware

Why it matters

NVIDIA DGX Spark hardware: GB10 (Grace Blackwell) SoC, 128 GB unified CPU+GPU memory, ConnectX‑7 200GbE QSFP ports, 4 TB encrypted NVMe M.2, 1 PFLOPS FP4 performance; MSRP listed at $4,699/unit.

  • 2x DGX Spark is the "sweet spot" (tested by @MiaAI_lab): connect two units with one QSFP 200GbE cable, prefer manual static IPs (example Node1 192.168.100.10/24, Node2 192.168.100.11/24), enable passwordless SSH, run NCCL tests and ensure identical driver versions.
  • Model performance benchmarks: single Spark — Qwen 3.6 35B NVFP4: 256k context, ~110 tok/s (Alibaba, Apr 2026); 2x Sparks — DeepSeek V4 Flash DSpark: 1M context, 40–45 tok/s single session, ~79 tok/s with 3 concurrent sessions (NVFP4); 4x Sparks — GLM 5.2 NVFP4: ~20 tok/s at 1M context.
  • vLLM/docker production tips and critical flags: example run commands provided for single and distributed serve (add --tensor-parallel-size 2 for 2x); use --gpu-memory-utilization 0.85, --max-num-seqs 4, set --max-model-len per model (131072/262144/1048576), consider --kv-cache-dtype fp8, DO NOT add --quantization for NVFP4, cache safetensors first (10–15 min), warm up after boot (~25s), and pin image digests.

NVIDIA DGX Spark is presented as a practical, low‑cost (listed $4,699) way to run models that exceed conventional GPU VRAM limits: the unit uses a GB10 (Grace Blackwell) SoC with 128 GB unified memory, ConnectX‑7 200GbE QSFP, 4 TB encrypted NVMe and ~1 PFLOPS FP4. The author and community testers (@MiaAI_lab) recommend a 2x Spark cluster as the "sweet spot": direct 200GbE QSFP link, static IP netplan examples, passwordless SSH, NCCL verification and matching drivers. Benchmarks show Qwen 3.6 35B NVFP4 at 256k context ≈110 tok/s (Apr 2026) on 1x Spark, DeepSeek V4 Flash DSpark hitting 1M context and 40–45 tok/s (≈79 tok/s with 3 concurrent sessions) on 2x, and GLM 5.2 NVFP4 ~20 tok/s at 1M context on 4x. The post includes vLLM Docker commands for single and distributed serving and operational guidance (--gpu-memory-utilization 0.85, --max-num-seqs 4, cache safetensors, warm up, pin digests) plus notes on scaling to 4–8 Sparks with 200GbE switches (e.g., MikroTik CRS504).

By @ivanfioravanti
3 Twitter/X 2026-06-28 1 min read
Open

TSMC and Winbond will collaborate on next-generation Wafer-on-Wafer (WoW) 3D…

Why it matters

TSMC and Winbond will collaborate on next-generation Wafer-on-Wafer (WoW) 3D stacking: Winbond will supply DRAM wafers and TSMC will stack them with its logic wafers to produce AI chips.

  • TSMC, which historically relied on Samsung, SK Hynix, and Micron, is diversifying to cultivate a Taiwanese memory supply chain to strengthen AI-chip supply stability; Winbond expects the partnership to springboard it into AI server and high-performance computing supply chains.
  • Report by @jukan05 on 2026-06-28 frames the deal as a response to 'deepening global memory shortages.'

TSMC and Winbond will jointly develop Wafer-on-Wafer (WoW) 3D stacking: Winbond supplying DRAM wafers and TSMC stacking them with logic wafers for AI chips. The move, reported 2026-06-28, is positioned as TSMC’s strategy to diversify beyond Samsung, SK Hynix, and Micron and build a Taiwanese memory supply chain while enabling Winbond’s push into AI server and HPC markets.

By @jukan05
4 Twitter/X 2026-06-28 1 min read
Open

Oracle's SEC filing attributes mass layoffs to AI

Why it matters

Oracle's SEC filing attributes mass layoffs to AI: 21,000 jobs cut (13% of workforce) with a $1.8 billion restructuring charge.

  • Internal pilots replaced teams of 47 database administrators with 3 senior architects plus AI automation; the AI reportedly catches ~94% of issues and reduced solution engineering from 6 weeks to 6 hours.
  • Oracle will redirect savings into a $50 billion FY2026 AI infrastructure buildout (data centers, GPUs, cloud capacity); by early March 2026 ~90,000 tech jobs had been cut industry-wide, with over 70% citing AI.

Oracle became the first S&P 500 company to explicitly blame AI in its SEC filing: cutting 21,000 roles (13%) and taking a $1.8B charge after pilots that replaced 47 DBAs with 3 senior architects plus AI (catching ~94% of issues and shrinking 6‑week tasks to 6 hours). Oracle says savings will fund a $50B FY2026 AI infrastructure buildout; the post frames this as a replicable Fortune 500 playbook amid ~90,000 tech layoffs by March 2026, >70% citing AI.

By @heyshrutimishra
5 Twitter/X 2026-06-28 3 min read
Open

DeepSeek open-sourced DSpark/DeepSpec on 2026-06-28 (post by @heyshrutimishra)…

Why it matters

DeepSeek open-sourced DSpark/DeepSpec on 2026-06-28 (post by @heyshrutimishra), releasing the V4 Pro DSpark checkpoint on Hugging Face and training code on GitHub under an MIT license; the system is already running in production on millions of users and available as a free download.

  • DSpark combines a parallel backbone with a lightweight Markov head to preserve token-to-token coherence while generating in parallel, improving accepted token length by 26–31% versus Eagle3 across Qwen3 model families and by 16–18% versus DFlash.
  • To cut verification waste under load, DSpark adds a confidence-prediction head that estimates rejection probability and dynamically trims verification length per request based on server load, switching verification aggressiveness as GPUs idle or saturate.
  • Measured production impact: per-user generation speed increased 60–85% on DeepSeek-V4 Flash and 57–78% on V4 Pro versus the MTP-1 baseline; aggregate throughput under high-concurrency/strict latency improved 51–400% depending on server load. Paper co-authored by founder Liang Wenfeng and Peking University researchers; framework already works with Qwen3 and Gemma model families.

DeepSeek released DSpark (and the DeepSpec training/eval stack) on 2026-06-28, open-sourcing the code and checkpoints under an MIT license and claiming production deployment across millions of users. DSpark addresses the token-by-token latency of large LLMs by combining a parallel drafter backbone with a lightweight Markov head so drafts are both fast and token-coherent, yielding 26–31% better accepted-token lengths versus Eagle3 (Qwen3 families) and 16–18% versus DFlash. It also adds a confidence-prediction head that estimates rejection probability and dynamically trims verification length based on server load to avoid wasted GPU verification. DeepSeek reports per-user generation speed gains of 60–85% on V4 Flash and 57–78% on V4 Pro (vs MTP-1), with aggregate throughput improvements of 51–400% under high concurrency. The paper lists founder Liang Wenfeng and Peking University collaborators; checkpoints are on Hugging Face and training code on GitHub, and the framework already supports Qwen3 and Gemma models.

By @heyshrutimishra
Worth reading

Useful context and follow-up reading when you have more time.

11 items
1 Twitter/X 2026-06-28 2 min read
Open

As of reports on June 26, 2026, industry inventories of high‑purity CO2 have…

Why it matters

As of reports on June 26, 2026, industry inventories of high‑purity CO2 have fallen below the typical combined one‑month buffer (manufacturers and suppliers each normally hold ~2 weeks), prompting intensified procurement by chipmakers.

  • Samsung Electronics consumes ~1,800–2,000 tons/month and SK Hynix ~600–700 tons/month; liquefied CO2 prices have risen ~20% since the start of 2026 and suppliers say they cannot ramp output short‑term due to reduced feedstock from lower refinery/petrochemical utilization tied to US/Iran/Middle East crude instability.
  • High‑purity CO2 is used in large volumes for supercritical wafer cleaning in advanced nodes (to dissolve residues and penetrate fine patterns); major domestic suppliers named are Taekyung Chemical (market leader), Sundo Chemical, Dongkwang Chemical, and SK Air Plus, and the supply crunch is expected to likely continue through year‑end.

High‑purity CO2 procurement for semiconductor supercritical cleaning is under strain after June 26, 2026 reports showed inventories falling below the usual combined one‑month buffer. Reduced CO2 feedstock from lower oil refinery and petrochemical utilization—linked to US/Iran/Middle East crude instability—has driven ~20% price rises and constrained supplies that cannot be quickly ramped up, risking continued shortages through year‑end.

By @jukan05
2 Twitter/X 2026-06-28 1 min read
Open

La IA está generando una demanda de energía que las redes tradicionales no pueden…

Why it matters

La IA está generando una demanda de energía que las redes tradicionales no pueden satisfacer; empresas como xAI y OpenAI planean centros de datos que consumirán cientos de megavatios cada uno solo para entrenar modelos.

  • Un solo clúster de entrenamiento puede requerir potencia equivalente a la de una ciudad de 100,000 habitantes.
  • En lugar de depender de subsidios estatales, las compañías están cerrando acuerdos privados con productores de gas natural licuado (GNL) y con reactores nucleares modulares pequeños, creando capacidad y empleos bien pagados en estados como Texas y Nevada; el autor sostiene que el mercado responde más rápido que comités gubernamentales y que el progreso en IA depende de libertad para innovar en infraestructura energética sin interferencias políticas.

La inteligencia artificial está disparando la demanda de energía y obliga a empresas como xAI y OpenAI a planear centros de datos de cientos de megavatios para entrenar modelos; un clúster puede consumir tanta potencia como una ciudad de 100,000 habitantes. Las compañías están cerrando acuerdos privados con productores de GNL y reactores nucleares modulares, generando capacidad y empleos en Texas y Nevada, según el autor.

By @martinvars
3 Twitter/X 2026-06-28 1 min read
Open

Gavin S.

Why it matters

Gavin S. Baker said on All‑In at 1:04:06: "DRAM is the most important bottleneck," and Elon is focusing the Terafab on memory (DRAM), not lasers, capacitors, power semis, NAND, or HDDs.

  • Jukan (@jukan05) says this validates DRAM’s strategic importance and challenges the view that DRAM is a commodity easily replaceable by Chinese products (tweet published 2026-06-28).
  • Jukan argues Micron (MU) could become more important than Nvidia because inference ROI is driven more by adding memory—GPUs are often underutilized in inference due to DRAM bottlenecks—so 'inference is memory.'

Jukan (@jukan05) cites Gavin S. Baker’s All‑In comment at 1:04:06 that "DRAM is the most important bottleneck" and that Elon is focusing the Terafab on memory, not lasers, capacitors, power semis, NAND or HDDs. Jukan says this validates DRAM’s strategic value, rejects the idea it’s easily replaced by Chinese products, and argues Micron (MU) could exceed Nvidia in importance because inference ROI depends on memory, not GPUs.

By @jukan05
4 Twitter/X 2026-06-28 1 min read
Open

Humanoid production hit 15,000 units

Why it matters

Humanoid production hit 15,000 units: two years to produce the first 1,000 and three months to produce the last 5,000, indicating a steep adoption/production ramp.

  • AgiBot G2 at Longcheer in Nanchang ran an 8-hour shift, self-corrected a conveyor position deviation, the fleet ran 10 hours straight, and sorted >3,000 units at a claimed 100% success rate while communicating with factory software and flagging defective parts.
  • Shruti says these are live, unstaged factory deployments next to human workers, compares the ramp to EVs and smartphones, and warns the U.S. is underrating the pace (author returned from China observing the ecosystem).

Shruti (@heyshrutimishra) reports humanoid production has reached 15,000, with two years to build the first 1,000 and only three months for the most recent 5,000, signaling a steep adoption curve. At Longcheer in Nanchang, AgiBot's G2 ran an 8‑hour shift, self-corrected a conveyor deviation, ran 10 hours total, and sorted >3,000 units at 100% success while interfacing with factory software; she warns the U.S. is underrating the pace.

By @heyshrutimishra
5 Twitter/X 2026-06-28 1 min read
Open

España tiene uno de los mejores recursos solares y eólicos de Europa pero, según…

Why it matters

España tiene uno de los mejores recursos solares y eólicos de Europa pero, según @martinvars (publicado 2026-06-28), paga una de las electricidades más caras del continente.

  • El autor atribuye los precios altos a una política que programa el cierre de reactores nucleares que funcionan perfectamente, carga el sistema con impuestos y peajes, y deja la red dependiente del clima; menciona el apagón reciente como prueba.
  • Consecuencia económica: la energía cara empuja a la industria española a mirar a Estados Unidos, donde ‘el megavatio cuesta una fracción’; el precio de la energía es una decisión de gobierno que, afirma, puede revertirse.

España dispone de recursos solares y eólicos entre los mejores de Europa pero sufre algunas de las tarifas eléctricas más altas, según @martinvars (28 junio 2026). Culpa a decisiones de política —cierres programados de centrales nucleares operativas, impuestos y peajes elevados y una red dependiente del clima—; el apagón reciente lo evidenció. Afirma que la energía cara es resultado de una decisión de gobierno y puede cambiarse.

By @martinvars
6 Twitter/X 2026-06-28 1 min read
Open

Micron (MU) could become more important than Nvidia for AI inference because…

Why it matters

Micron (MU) could become more important than Nvidia for AI inference because inference ROI depends more on memory than on additional GPUs, per @jukan05 (2026-06-28).

  • Inference workloads often leave GPUs underutilized due to memory bottlenecks, so adding memory (not more Nvidia GPUs) yields greater inference performance/value.
  • A reply from @TicTocTick predicts MU will crash from "now 1200" to 700 soon (recalling a past level of 80) and bluntly states "RAM is NOT GPU."

Micron (MU) is positioned as potentially more critical than Nvidia for AI inference: @jukan05 (2026-06-28) argues inference ROI hinges on memory capacity, not GPU count, because memory bottlenecks leave GPUs idle. A reply from @TicTocTick predicts MU falling from "now 1200" to 700 and repeats the slogan "RAM is NOT GPU."

By @jukan05
7 Twitter/X 2026-06-28 2 min read
Open

Rubycon (Japan's third-largest aluminum capacitor maker) issued a price-increase…

Why it matters

Rubycon (Japan's third-largest aluminum capacitor maker) issued a price-increase notice effective August 1, 2026 covering aluminum electrolytic capacitors, solid aluminum capacitors, and film capacitors and warned a second increase may follow; the company cited high crude oil, rising metal prices (aluminum, copper, tin), supplier hikes across raw/auxiliary materials and logistics.

  • Jiang Hai, the leading Chinese aluminum capacitor maker, sent a late-June 2026 price-notice raising prices for aluminum capacitors, film capacitors, and supercapacitors, attributing the move to steep rises in aluminum foil, chemical raw materials, carbon powder and electricity costs.
  • With Rubycon's move, Japan's top three makers—Nippon Chemicon, Nichicon, and Rubycon—have effectively all raised prices; the simultaneous action amid the AI infrastructure boom is notable because these Japanese firms ship significant volumes for AI servers.
  • Upstream cost drivers named include lead pins, aluminum cases, metal materials, petroleum-linked inputs and logistics; the author argues these conditions give Taiwanese capacitor makers favorable leverage to pass on cost pressure to customers.

Aluminum-capacitor makers in China and Japan have launched a coordinated round of price increases as H2 2026 begins. Rubycon — Japan's third-largest maker — announced a price rise effective August 1, 2026 for electrolytic, solid-aluminum and film capacitors and warned of a possible second hike, citing sustained high crude oil, surging metals (aluminum, copper, tin) and supplier-wide cost increases across materials and logistics. In late June 2026 Jiang Hai, China’s leading aluminum-capacitor supplier, issued its own notice raising prices for aluminum capacitors, film capacitors and supercapacitors because aluminum foil, chemical inputs, carbon powder and electricity costs have exceeded prior pricing models. The post highlights that Nippon Chemicon, Nichicon and Rubycon now all are raising prices, that these makers ship substantial volumes into AI server supply chains, and that the combined upstream cost pressure makes it easier for Taiwanese makers to pass costs on to customers.

By @jukan05
8 Twitter/X 2026-06-28 1 min read
Open

Gary Marcus (2026-06-28) wrote “hard to see how anyone makes much money from all…

Why it matters

Gary Marcus (2026-06-28) wrote “hard to see how anyone makes much money from all this in the long run,” likening the AI sector to the airline industry with “very small margins and big expenses” and calling the “pouring trillions in” probably unwise.

  • Xiaoyin Qu claims China’s playbook is to “kill OpenAI and anthropic” by providing free, high-quality models and exporting cheap compute powered by low-cost electricity; she says chips are the current blocker but “Huawei would catch up soon,” making inference cost “almost zero” instead of “hundreds of billions.”

Gary Marcus (2026-06-28) warns AI will be low-margin and costly—“more like the airline industry”—and calls pouring “trillions” into it likely unwise. He amplifies Xiaoyin Qu’s scenario that China could undercut OpenAI and Anthropic by giving away top-tier models and exporting cheap compute (cheap electricity plus Huawei closing the chip gap), driving inference costs from “hundreds of billions” toward “almost zero.”

By @GaryMarcus
9 Twitter/X 2026-06-28 1 min read
Open

Friend runs an AI SaaS at $70k MRR with ~30% profit margin (~$21k/month) after…

Why it matters

Friend runs an AI SaaS at $70k MRR with ~30% profit margin (~$21k/month) after paying for infrastructure, AI agents, and Google/Meta ads.

  • Author's B2C mobile apps deliver higher profit margins despite lower MRR compared with the AI SaaS.
  • Key trade-off: the AI SaaS may be sold for millions (more sellable asset), while B2C mobile apps are harder to sell at strong valuations — choose cashflow now or a more sellable asset later.

The author recounts meeting a friend who runs an AI SaaS generating $70k MRR but only ~30% profit (~$21k/month) after infrastructure, AI-agent, and Google/Meta ad costs. He contrasts that with his higher-margin B2C apps at lower MRR and frames the strategic choice: prioritize immediate high-margin cashflow or build a more sellable SaaS asset.

By @alexcooldev
10 Twitter/X 2026-06-28 1 min read
Open

John Loeber (2026-06-28) asserts all future programs will be formally verifiable…

Why it matters

John Loeber (2026-06-28) asserts all future programs will be formally verifiable, claiming formal verification is necessary for cybersecurity and to enable codebases 100x+ larger as AI drives software scale.

  • Jesse Michael Han cites Gwern's view that 2026-era LLMs may yield Lean implementations with worse baseline loss but better scaling exponents, implying Lean rewrites could eventually 'win' and justify large investments to improve program correctness and global security.

John Loeber (2026-06-28) argues that all future software must be formally verifiable to make cybersecurity possible and to support codebases 100x+ larger as AI increases scale. Jesse Michael Han quotes Gwern saying 2026-era LLMs give Lean worse baseline loss but superior scaling exponents, so Lean rewrites could eventually deliver major correctness and security gains.

By @johnloeber
11 Twitter/X 2026-06-28 1 min read
Open

Defaulting engineers to open-weight models (GLM 5.2 and Kimi 2.7) via an LLM…

Why it matters

Defaulting engineers to open-weight models (GLM 5.2 and Kimi 2.7) via an LLM gateway lowered cost pressure—91% of employees never hit usage caps, so the team moved to cheaper defaults instead of lowering caps or adding friction.

  • Better routing and caching cut costs: LibreChat's cache hit rate rose from 5% to 60% after changes, and preprocessing prompts to route planning to frontier models and execution to cheaper models helped cut AI spend nearly in half while token usage keeps growing.
  • Operational rules to reduce wasted tokens include starting fresh sessions when switching tasks, scoping file context narrowly, disconnecting unused tools, making requests cache-aware, and automating model selection rather than having humans pick models.

Brian Armstrong outlines operational tactics to keep AI spend flat while token usage grows: default engineers to open-weight models (GLM 5.2, Kimi 2.7), preprocess and route prompts to the right model, and implement cache-aware requests (LibreChat hit rate 5%→60%). These changes reportedly cut AI spend nearly in half without suppressing usage.

By @ccatalini
Quick skim

Scan these for facts, links, or weak signals worth tracking.

64 items · open
1 Twitter/X 2026-06-28 1 min read
Open

El 28 de junio de 2026 @martinvars afirma que España volverá a registrar más…

Why it matters

El 28 de junio de 2026 @martinvars afirma que España volverá a registrar más muertes que nacimientos en 2026 y que la población activa que sostiene a los jubilados se está achicando.

  • Afirma que las respuestas habituales —importar gente o prometer cheques por hijo— no tocan la raíz: la vivienda es carísima, el primer empleo llega tarde y mal pagado y la vida estable es un lujo; propone construir mucho más, regular mucho menos y dejar que la gente se quede con lo que gana para que formar familia antes de los 40 vuelva a ser normal.

El tuit de @martinvars (28-06-2026) avisa que España tendrá otra vez más muertes que nacimientos y que la fuerza laboral que sostiene a los jubilados se reduce. Rechaza soluciones como inmigración o cheques por hijo y exige atacar raíces: vivienda asequible, empleos tempranos y mejor pagados; su receta política es construir mucho más, desregular y reducir cargas para que los jóvenes puedan formar una familia.

By @martinvars
2 Twitter/X 2026-06-28 1 min read
Open

@bcherny (2026-06-28) identifies five archetypes on the Claude Code team…

Why it matters

@bcherny (2026-06-28) identifies five archetypes on the Claude Code team: Prototyper (many new ideas, most don't ship), Builder (turns prototypes into production), Sweeper (cleans UI/code, unships, optimizes), Grower (iterates to improve PMF), and Maintainer (owns mature system reliability/efficiency).

  • He observes individuals commonly span 2–3 of these archetypes and that the archetypes cut across job functions — designers, engineers, PMs, and data scientists at Anthropic map to different archetypes rather than fixed roles.
  • He prescribes role mixes by product stage: pre-PMF products need archetypes 1+2+3; growing/early-PMF products need 2+3+4 with some 5; mature, PMF-strong products need 3+4+5 with some 2.

@bcherny maps five product-role archetypes he sees on the Claude Code team—Prototyper, Builder, Sweeper, Grower, Maintainer—and argues people often span 2–3 archetypes. He notes these archetypes cross engineering, design, PM and DS functions and recommends specific mixes (1+2+3, 2+3+4+some5, 3+4+5+some2) depending on product stage.

By @bcherny
3 Twitter/X 2026-06-28 1 min read
Open

Anton Osika (@antonosika) says his friend Kristian Rønn (@Kristian_Ronn) is…

Why it matters

Anton Osika (@antonosika) says his friend Kristian Rønn (@Kristian_Ronn) is building cryptographic verification to confirm superintelligence is not misused, enabling both open-source projects and commercial labs to prove compliance without exposing model secrets or private data.

  • Osika frames AI as a trust/verification problem—'no one can verify what anyone else is doing'—and links to Billy Perrigo’s TIME article 'Can the Cold War Teach Us How to Slow Down AI?' published June 23, 2026, which covers Kristian's work.

Anton Osika highlights that his friend Kristian Rønn (@Kristian_Ronn) is developing cryptographic verification to ensure superintelligence isn't misused, letting both open-source projects and commercial labs prove compliance without revealing secrets or private data. Osika casts the issue as a verification shortfall—'no one can verify what anyone else is doing'—and links to Billy Perrigo’s TIME piece (June 23, 2026).

By @antonosika
4 Twitter/X 2026-06-28 1 min read
Open

Black Spark claims it infiltrated the Alabuga special economic zone in Tatarstan…

Why it matters

Black Spark claims it infiltrated the Alabuga special economic zone in Tatarstan for several months, exfiltrated databases on personnel involved in assembling Geran-2 and Geran-3 drones, and hacked Alabuga’s website (archived copy dated 27 June 2026).

  • Black Spark says it embedded "surprises" into recent UAV batches that will cause "unpleasant consequences" when operators attempt launches; Alabuga reportedly shut down servers and FSB agents are checking staff. Michael D. Weiss links theins.ru for context.

Black Spark claims it infiltrated the Alabuga SEZ in Tatarstan, stole personnel databases tied to Geran-2 and Geran-3 assembly, hacked the site (archived 27 June 2026) and secretly embedded "surprises" in recent UAV batches that will produce "unpleasant consequences" on launch; Alabuga allegedly shut servers and FSB agents are now checking staff, per Michael D. Weiss.

By @michaeldweiss
5 Twitter/X 2026-06-28 1 min read
Open

Levie (2026-06-28) claims 'mythos level' cybersecurity models will soon be open…

Why it matters

Levie (2026-06-28) claims 'mythos level' cybersecurity models will soon be open and available to anyone, which will enable alternative tech stacks that shift economic value and control away from the US tech stack.

  • He argues that gatekeeping advanced models (not releasing them) won’t improve security or strategic position if such models become public anyway, and that creating gates asymmetrically disadvantages the gatekeepers versus staying at the frontier and driving future AI architectures.
  • Polymarket reports a new Chinese model from Zhipu AI reportedly matches Claude Mythos’ performance at finding security bugs, which Levie cites as evidence China can catch up and that regulators should not assume China will fall behind.

Levie (June 28, 2026) warns that imminently open 'mythos level' cyber-security models will spawn alternative tech stacks and shift economic/control leverage away from the US; he says withholding models won’t make the US safer and instead risks asymmetric disadvantage, and points to a Polymarket claim that Zhipu AI matches Claude Mythos at bug-finding as proof China can catch up.

By @levie
6 Twitter/X 2026-06-28 1 min read
Open

Ford rehired, newly hired, or promoted 350 experienced engineers after AI systems…

Why it matters

Ford rehired, newly hired, or promoted 350 experienced engineers after AI systems and adjusted design requirements failed to maintain product quality, according to Ford's VP of hardware engineering.

  • Ford has cut over 5,000 workers since 2020 and recalled more cars than any other U.S. automaker this year; after bringing engineers back it achieved JD Power's top quality ranking for the first time in nearly 20 years.
  • CEO Jim Farley continues to say AI will replace half of U.S. white-collar workers while Ford plans 100,000 additional AI-powered tests; Gary Marcus labels the episode 'The Klarna Effect,' arguing Ford proved AI couldn't replace those 350 engineers.

Gary Marcus highlights Ford's failed experiment replacing engineers with AI: after cutting more than 5,000 jobs since 2020 and relying on AI-adjusted designs, Ford rehired or promoted 350 experienced engineers when quality collapsed. The company now tops JD Power for the first time in nearly 20 years, plans 100,000 more AI tests, yet CEO Jim Farley still predicts mass white-collar displacement.

By @GaryMarcus
7 Twitter/X 2026-06-28 2 min read
Open

On June 23, 2026 the KOSPI plunged 9.99% (910 points) to 8,203.84 after Samsung…

Why it matters

On June 23, 2026 the KOSPI plunged 9.99% (910 points) to 8,203.84 after Samsung (-12.3%) and SK Hynix (-12.5%) triggered a 20‑minute circuit breaker; the index had closed at an all‑time high of 9,114.55 the previous day after tripling from ~3,000 a year earlier on AI‑chip strength.

  • Foreign and Korean institutional investors net sold roughly $8.5 billion on June 23 while Korean retail net bought a record 11.55 trillion won (~$7.16 billion); retail margin debt reached a record 38.48 trillion won four days earlier and regulators had approved 2x single‑stock ETFs on Samsung and SK Hynix one month prior.
  • Author's claim: Korean retail now holds the most leveraged position in market history at the highest valuation with "smart money" gone, meaning leverage (margin loans, pledged assets, 2x ETFs) — not the crash itself — is the primary risk; Surmount promotes rules‑based strategies to avoid forced, panic selling.

SurmountInvest recounts the June 23, 2026 KOSPI rout — a 9.99% drop to 8,203.84 after Samsung (-12.3%) and SK Hynix (-12.5%) triggered a circuit breaker — and warns that record retail net buying (11.55 trillion won) funded by a record 38.48 trillion won in margin debt and new 2x ETFs leaves retail overleveraged and at systemic risk.

By @hamptonism
8 Twitter/X 2026-06-28 1 min read
Open

Brad began at O'Reilly in July 1996 as a 17-year-old parts specialist in Wagoner…

Why it matters

Brad began at O'Reilly in July 1996 as a 17-year-old parts specialist in Wagoner, Oklahoma, worked every store role before age 20 (counterperson, night manager, assistant store manager), and became the company's fourth CEO in 2024 after 28 years.

  • O'Reilly grew from under 200 stores in 1996 to 6,300+ locations, 93,000 employees and 31 distribution centers; the stock returned more than 4,000% over the past 20 years.
  • O'Reilly intentionally promotes from store operations: all 650 district managers and all 75 region managers rose from running a store, and both prior CEOs were promoted internally, creating a system that grooms store-floor employees into executives.

Brad, the CEO of O'Reilly Automotive, started as a 17-year-old parts specialist in Wagoner, OK in July 1996 and rose through every store role before age 20. Being passed over for district manager at 23 taught him to develop others; he became CEO in 2024 as O'Reilly scaled massively and institutionalized store-to-executive promotion.

By @aakashgupta
9 Twitter/X 2026-06-28 1 min read
Open

On 2026-06-28 @cramforce warned there is a short-time window where the defensive…

Why it matters

On 2026-06-28 @cramforce warned there is a short-time window where the defensive play (making source code available) is superior to the offensive play (black-box pen testing).

  • He recommends running harnesses like deepsec (GitHub: vercel-labs/deepsec), a coding-agent-powered security harness, with available frontier models to harden apps now and reduce findings for future pen-test models.
  • Guillermo Rauch (@rauchg) stated Mythos/Sol capabilities are useful both offensively and defensively and cautioned that if adversaries obtain equivalent offensive capability it poses a serious threat to US companies.

@cramforce (2026-06-28) says a brief window exists where defensive measures—sharing source code and using automated harnesses—outperform black-box pen testing. He urges teams to run deepsec (vercel-labs/deepsec) with frontier coding models immediately to harden applications. Guillermo Rauch warns Mythos/Sol-style tools can be weaponized by adversaries, threatening US firms.

By @cramforce
10 Twitter/X 2026-06-28 1 min read
Open

In June 2026 the 'Mag 7' hyperscalers faced 'brutal' pressure, but @jvisserlabs…

Why it matters

In June 2026 the 'Mag 7' hyperscalers faced 'brutal' pressure, but @jvisserlabs asserts this is rotation, not collapse: capital is rotating into healthcare, small caps, equal-weight, AI infrastructure receivers, memory, biotech, and agentic commerce.

  • He frames the market as in an 'AI midcycle slowdown' where 'easy money in the spenders is over' and the market is separating firms that spend on AI from firms that benefit from AI; he warns crash fears will continue.
  • He identifies Micron and memory shortages as the real AI bottleneck, predicts token usage may explode because of loops and tags, and forecasts a move from AI infrastructure to AI applications—citing Stripe Sessions, agentic commerce, and GLP-1–related healthcare/biotech as beneficiaries.

Author @jvisserlabs argues that June 2026 pain among the 'Mag 7' hyperscalers reflects rotation, not collapse: AI is consolidating as capital flows to healthcare, small caps, equal-weight, AI infrastructure receivers, and memory (Micron) amid shortages. He warns token usage could surge via loops/tags and predicts a shift from infrastructure to AI applications.

By @jvisserlabs
11 Twitter/X 2026-06-28 1 min read
Open

The author claims Anthropic's CEO called for regulation to shift legal and…

Why it matters

The author claims Anthropic's CEO called for regulation to shift legal and political responsibility away from the company if 'Mythos 5' later ends up in China's hands and causes damage.

  • The post quotes the alleged strategy: “See, US government, you regulated this poorly. Anthropic is not to blame.”
  • The author notes the CEO pushed for regulation 'few months ahead of an IPO', arguing the motive was strategic (avoid blame) rather than a genuine desire to be regulated.

Chubby♨️ argues Anthropic’s CEO lobbied for AI regulation as a strategic move to shift blame away from the company if 'Mythos 5' later causes damage in China. The post alleges the CEO sought regulations 'few months ahead of an IPO' to enable blaming the US government rather than Anthropic.

By @kimmonismus
12 Twitter/X 2026-06-28 1 min read
Open

Josh Rauh (quoting Chamath Palihapitiya) claims California’s Property Seizure Act…

Why it matters

Josh Rauh (quoting Chamath Palihapitiya) claims California’s Property Seizure Act — labeled the "Billionaire Tax" — would require handing over 5% in cash of all tangible and intangible property every year, resulting in confiscation of assets, destroyed 401(k)s, and lost pensions.

  • He predicts a decade of litigation likely to reach the Supreme Court; meanwhile business flight from California will shrink revenue, deepen the budget hole, force more borrowing and higher taxes on everyone (hitting the middle class hardest), and could lead to a hard landing, near-bankruptcy, and pension "reset/retrade."

Josh Rauh warns that California’s Property Seizure Act — labeled the 'Billionaire Tax' and highlighted by Chamath Palihapitiya and Hoover Institution media — would require a 5% annual cash levy on all tangible and intangible property, trigger decade-long litigation likely to reach the Supreme Court, spur business flight, deepen the budget hole and risk pension resets, higher taxes and near-bankruptcy.

By @joshrauh
13 Twitter/X 2026-06-28 1 min read
Open

Five-level AI maturity framework

Why it matters

Five-level AI maturity framework: Level 1 — AI as a thought partner (bouncing strategy off ChatGPT); Level 2 — AI as an assistant (drafting, summarizing); Level 3 — AI as a team (running creatives, handling first drafts, building reports); Level 4 — AI as the system (entire workflows rebuilt); Level 5 — agents managing agents with compounding feedback loops.

  • Author @aymanalabdul (published 2026-06-28) reports that most CEOs he talks to are stuck at Levels 1–2 while calling their companies 'AI native'.
  • He claims the companies 'pulling away' are those actively building toward Level 5 — autonomous agent hierarchies that get smarter each cycle without human intervention.

@aymanalabdul lays out a five-level AI maturity model: from AI as a thought partner (Level 1) to agents managing agents with compounding feedback loops (Level 5). He argues most CEOs remain at Levels 1–2 yet label themselves 'AI native', and that true competitive separation will come from firms advancing toward Level 5 autonomy.

By @aymanalabdul
14 Twitter/X 2026-06-28 1 min read
Open

On 2026-06-28 @kimmonismus flagged claims that “a new model from Zhipu AI” is at…

Why it matters

On 2026-06-28 @kimmonismus flagged claims that “a new model from Zhipu AI” is at least as strong as Fable5 for cybersecurity, but their search found only a Wall Street Journal article naming GLM-5.2 as a recently released model.

  • The WSJ reports GLM-5.2 can match top U.S. models in some bug-finding scenarios; separately, China’s 360 Security says its Tulongfeng tool is comparable to Anthropic’s Mythos (Claude Mythos).
  • The author concludes either GLM-5.2 is stronger than commonly believed or circulating reports are conflating/overstating different models (i.e., misleading coverage).

Kimmonismus (2026-06-28) challenges claims that “a new model from Zhipu AI” matches Fable5 on cybersecurity, finding only a Wall Street Journal piece about GLM-5.2. WSJ says GLM-5.2 can match top U.S. models in some bug-finding tasks, and 360 Security claims Tulongfeng rivals Anthropic’s Mythos, suggesting either underestimated GLM-5.2 or misleading reporting.

By @kimmonismus
15 Twitter/X 2026-06-28 1 min read
Open

Argentina closed its first sustained fiscal surplus in more than a decade…

Why it matters

Argentina closed its first sustained fiscal surplus in more than a decade, according to the post dated 2026-06-28, a result the author credits to Milei's government.

  • The post says the surplus was achieved by cutting public spending, eliminating entire ministries, and liberalizing prices; monthly inflation fell to single digits, and there were 'no rescues' or 'accounting tricks.'
  • The author argues a state that spends less than it collects practices 'arithmetical' discipline that produces prosperity, and that lesson applies to Spain and the United States.

Argentina closed its first sustained fiscal surplus in more than a decade, the post asserts, after measures including big spending cuts, elimination of entire ministries, and price liberalization. The author reports monthly inflation dropped to single digits without bailouts or accounting tricks and argues Javier Milei demonstrated that fiscal discipline—spending less than revenue—produces prosperity, a lesson for Spain and the U.S.

By @martinvars
16 Twitter/X 2026-06-28 1 min read
Open

At age 5 in Villa Gesell, Argentina, the author got lost on the beach; a very…

Why it matters

At age 5 in Villa Gesell, Argentina, the author got lost on the beach; a very tall man put him on his shoulders, a crowd formed and clapped, and his parents appeared instantly—described as a spontaneous local system to reunite lost children.

  • Now living in the USA with his own kids, the author claims Americans would not do this—saying people here are afraid to pick up someone else’s child and that US society has fewer social means to handle lost people.

Author @martinvars recalls getting lost at age 5 in Villa Gesell, Argentina, where a tall stranger hoisted him onto his shoulders and a clapping crowd quickly reunited him with his parents; he contrasts that with life in the USA (2026-06-28 post), asserting Americans wouldn’t intervene similarly and that US society lacks such informal safety mechanisms.

By @martinvars
17 Twitter/X 2026-06-28 1 min read
Open

@martinvars (publicado 2026-06-28) afirma que la principal aportación de la IA en…

Why it matters

@martinvars (publicado 2026-06-28) afirma que la principal aportación de la IA en medicina es resolver el problema de acceso: "millones de personas" viven a horas del especialista más cercano y esperan "meses" por consultas necesarias.

  • En Certuma observan que sistemas de IA que ayudan al médico a leer imágenes, priorizar casos y detectar riesgos tempranos multiplican su alcance y afinan su criterio, llevando atención a pueblos sin hospital y a pacientes en listas de espera.

Martín Vars sostiene que la IA médica debe orientarse a ampliar el acceso: al ayudar a leer imágenes, priorizar casos y detectar riesgos tempranos, estas herramientas multiplican el alcance del clínico y cubren "territorios donde hoy no hay ninguna" atención. En Certuma ven diariamente cómo esa tecnología llega a pacientes que viven horas de distancia y esperan meses por consulta.

By @martinvars
18 Twitter/X 2026-06-28 1 min read
Open

A clear glass disc spun at 3,000 RPM (50 rev/s) in front of a lens prevents…

Why it matters

A clear glass disc spun at 3,000 RPM (50 rev/s) in front of a lens prevents raindrops: drops near the edge experience roughly 500 g and are torn into mist by shear; only a tiny drop at the axis (zero tangential speed) can remain and is too small to register on camera.

  • Snow behaves differently: centrifugal throw can miss flakes, but the same rotation reportedly creates a thin vacuum that lifts flakes off the surface, so rain is flung away and snow is 'sucked' away.
  • The engineering challenge is precision balancing: any imbalance at 3,000 RPM causes vibration and blur and produces strong gyroscopic forces when panning; the idea was patented in 1917 by two British engineers for ships and is used today (e.g., race broadcasts) to maintain clear footage in heavy weather.

A spinning clear-glass disc mounted before a camera lens—rotated to about 3,000 RPM (50 rev/s)—uses centrifugal force (~500 g at the edge) and shear to atomize raindrops so images stay clear, while rotation can create a thin vacuum that lifts snow. The hard part is balancing the disc to near-zero wobble; the concept was patented in 1917 by two British engineers and is still used in modern broadcasts.

By @aakashgupta
19 Twitter/X 2026-06-28 1 min read
Open

@martinvars, fundador con 30 años de experiencia (publicación 2026-06-28), afirma…

Why it matters

@martinvars, fundador con 30 años de experiencia (publicación 2026-06-28), afirma que la IA da una ventaja enorme al fundador que entiende su negocio en detalle.

  • Ventajas prácticas: la IA permite rediseñar procesos completos — eliminar pasos innecesarios, fusionar traspasos y cazar errores antes de que cuesten dinero; quien solo le pone 'un chatbot encima' obtiene poca mejora.
  • Tesis central: la tecnología amplifica al que ya sabe cómo se crea valor; esta ola recompensará a operadores serios mucho más que a modas pasajeras.

El fundador y autor @martinvars (publicado 2026-06-28) sostiene que, tras 30 años creando empresas, nunca vio una herramienta que premie tanto al que ya sabe operar: la IA permite rediseñar procesos, eliminar pasos, fusionar traspasos y detectar errores antes de costo. Aplicaciones superficiales tipo chatbot cambian poco; la ola favorecerá a operadores serios sobre modas.

By @martinvars
20 Twitter/X 2026-06-28 1 min read
Open

On 2026-06-28 @deanwball argued the most important legal questions in AI now…

Why it matters

On 2026-06-28 @deanwball argued the most important legal questions in AI now center on the First Amendment, asserting creation, distribution, and use of 'frontier AI' (large language models) should be treated as protected expression.

  • He identified two litigation priorities: (1) develop fact patterns demonstrating standing for parties outside labs, and (2) move beyond 'code is speech' and posting-into-the-void tactics because courts will ultimately decide disputes arising over the prior two weeks amid national-security concerns.
  • He urged the 'best legal minds of our time' to focus on these constitutional and standing questions now.

@deanwball on 2026-06-28 argued that First Amendment litigation will decide whether creating, distributing, and using frontier AI (large language models) is protected speech; he called for concrete fact patterns to establish standing for non-lab parties, to abandon simplistic 'code is speech' approaches, and to marshal top legal talent despite national-security tensions.

By @deanwball
21 Twitter/X 2026-06-28 1 min read
Open

@deanwball (2026-06-28) says the First Amendment ("1A") will place real limits on…

Why it matters

@deanwball (2026-06-28) says the First Amendment ("1A") will place real limits on state intervention in AI, but argues those limits do not make AI regulation impossible; he warns 1A could "bite" if policymakers try to regulate alignment in high levels of detail.

  • He recommends focusing regulation on frontier labs as entities rather than on models, framing that choice as a hard but necessary tradeoff to preserve American popular sovereignty and the First Amendment’s protections.

@deanwball argues (2026-06-28) that while the First Amendment (1A) will constrain state intervention and could complicate finely detailed alignment rules, it does not make AI regulation impossible. He urges regulators to target frontier labs as entities rather than models, framing this limit as an acceptable tradeoff to preserve American popular sovereignty and 1A protections.

By @deanwball
22 Twitter/X 2026-06-28 1 min read
Open

Austria publicly invited Anthropic to 'strategically establish and participate'…

Why it matters

Austria publicly invited Anthropic to 'strategically establish and participate' in the EU, promising 'legal certainty, market access, capital and a set of values' (quote from a letter shared by Andrew Curran in the 2026-06-28 post).

  • Author calls the outreach 'an act of desperation,' arguing U.S. regulation has cut Europe off from the most powerful models and that if the U.S. treats Fable and Mythos as national-security assets, relocating to Europe won't prevent access being restricted.
  • Author asserts the invitation implicitly admits Europe cannot build comparable models and predicts Anthropic will remain where high-volume compute and supply are guaranteed — increasingly concentrated within U.S. borders; the UK made similar overtures months earlier.

Austria's outreach to Anthropic (reported 2026-06-28) offers legal certainty, market access and capital to host the company in the EU, but the author argues this is a desperate move: U.S. regulation and control of top compute (and possible designation of Fable and Mythos as national-security assets) would still keep Europe cut off, so Anthropic will stay near U.S. compute.

By @kimmonismus
23 Twitter/X 2026-06-28 1 min read
Open

Dean W.

Why it matters

Dean W. Ball (tweeted 2026-06-28) says the most important legal questions in AI are First Amendment issues: what fact patterns demonstrate that creation, distribution, and use of 'frontier AI' qualify as protected expression.

  • He asks who outside the labs has standing to bring such suits, urges moving beyond 'code is speech' copium and the impulse to 'post into the void,' and predicts courts will ultimately decide disputes that surfaced in the prior two weeks.
  • Acknowledging national‑security headwinds, he argues the underlying technology is a large language model and that should count heavily; he calls for the best legal minds to be working on these questions now.

Dean W. Ball argues (2026-06-28) that the central legal battles over AI will be First Amendment cases: courts must determine whether the creation, distribution, and use of frontier AI are protected expression and who outside labs has standing to sue. He warns against relying on 'code is speech' rhetoric, notes national-security complications, and urges top legal minds to address these questions given that the underlying tech is a large language model.

By @deanwball
24 Twitter/X 2026-06-28 1 min read
Open

African elephants carry about 2,000 olfactory receptor genes versus 396 in humans…

Why it matters

African elephants carry about 2,000 olfactory receptor genes versus 396 in humans and ~800 in bloodhounds, meaning an elephant's trunk has roughly five times the scent hardware of a human and twice that of a dog.

  • Elephants can scent water up to 12 miles away, lock onto a single ripe tree from kilometers off, and reportedly distinguish human groups by smell; in the shared clip the elephant continuously sweeps her trunk and had already identified each fruit on a table before turning around and entering backwards to let people be surprised.

The African elephant’s ~2,000 olfactory receptor genes give it far superior smell: it can detect water 12 miles away, locate a ripe tree from kilometers, and differentiate human groups by scent. In the posted video she constantly resamples the air with her trunk, had already mapped the fruit buffet, and entered backwards to let people think she was surprised.

By @aakashgupta
25 Twitter/X 2026-06-28 2 min read
Open

Laurel built a 200+ row spreadsheet that breaks every function into Categories…

Why it matters

Laurel built a 200+ row spreadsheet that breaks every function into Categories and Tasks (Product splits into five: Feature Work, Strategy & Roadmap, Customer Research, Product Analytics, Voice of Customer) and adds a one-line description for each task plus an explicit Negative Examples column that defines wrong versions of the work (e.g., Competitive & Market Analysis negative: ongoing tracking of competitor release notes).

  • Every task pairs a positive and negative example so the table becomes a labeled dataset a model can use to learn boundaries; agents can read a feature request in Slack and deterministically route it to the correct function, task, and PM, giving new hires the same lookup answer.
  • The spreadsheet formally treats doing work by hand as off‑spec in some rows (Agentic negative example: 'writing a spec manually without AI as the primary driver') and underpins a $100M AI startup's Company OS workflow described in Claude Code (video timestamps include 2:04 Company OS GitHub structure, 14:31 playbook→agent pipeline, 50:08 AI Ops and the Sasha model, 59:01 four levels of AI maturity).

Laurel's spreadsheet (described by Aakash Gupta on 2026-06-28) converts one person's tacit AI workflow into a 200+ row Company OS: each task has a one-line description plus an explicit Negative Example that draws boundaries (Product split into five subfunctions). That labeled structure lets agents and new hires route Slack feature requests and enforces AI-first norms (Agentic: manual specs are off‑spec).

By @aakashgupta
26 Twitter/X 2026-06-28 1 min read
Open

Used Qwen3.6-27B-MTP-GGUF on four 64×64 images

Why it matters

Used Qwen3.6-27B-MTP-GGUF on four 64×64 images: three processed plain and one with MTP, encapsulated by the claim “An image is worth 16x16 words.”

  • DGX Spark is described as the “prefill king” but “too slow” on decode for the author’s workflow; they warn prefill speed can mislead because real coding sessions send smaller chunks, not full 128K-token loads, relying on SSD cache.
  • Author @ivanfioravanti (published 2026-06-28 14:34:17+00:00) is previewing results while debugging 128K-context support on M5 Max and DGX Spark using llamacpp.

@ivanfioravanti (2026-06-28) previews experiments with Qwen3.6-27B-MTP-GGUF on four 64×64 images (three plain, one MTP), claims “An image is worth 16x16 words,” and says DGX Spark excels at prefill but is too slow at decode. They emphasize real-life use sends smaller chunks cached on SSD and are working to enable flawless 128K context on M5 Max and DGX Spark with llamacpp.

By @ivanfioravanti
27 Twitter/X 2026-06-28 1 min read
Open

Ramp's product-design workflow now begins in an LLM, includes user calls and…

Why it matters

Ramp's product-design workflow now begins in an LLM, includes user calls and AI-generated prototypes, then moves into Figma and finally iterates against live behavior—relegating Figma to 'step 4 of 5'.

  • By 2026, generative interfaces that can be produced in minutes have dissolved the decade-long craft moat that dated to Figma's rise in 2016; scarcity has shifted to taste, product judgment, and deciding which interfaces deserve to exist.
  • Frontier designers are shipping PRs straight through Codex, with Figma serving as the finisher; Aakash Gupta and Rohan Varma (first PM at Cursor, now PM on Codex) argue that the real decisions now happen upstream in the LLM, prototype, and customer call (published 2026-06-28).

Aakash Gupta argues that Figma — the defining product-design tool since 2016 — has been demoted to step 4 in a new five-step workflow (LLM → user call → AI prototype → Figma → live-iteration) popularized by Ramp. He claims generative tooling (2026) removes the craft bottleneck, shifting value to taste, systems thinking, and judgment; Codex-powered PRs exemplify this upstream shift.

By @aakashgupta
28 Twitter/X 2026-06-28 1 min read
Open

Scott Wu claims AI will so transform work that future generations won't even…

Why it matters

Scott Wu claims AI will so transform work that future generations won't even recognize today's jobs as 'work', contrasting modern office tasks — 'pushing buttons' and 'sitting in a room and talking' (meetings) — with ancestral manual labor like farming and making clothes by hand.

  • @lulumeservey posted a short video titled 'Me nodding to every point Scott makes' on 2026-06-28 endorsing Scott's argument; the post cites a David Senra tweet quoting @ScottWu46.

Lulumeservey posted a short reaction video on 2026-06-28 agreeing with Scott Wu's claim that AI will so change the nature of work that future generations won't recognize today's office jobs as 'work.' The post highlights Scott's contrast between 'pushing buttons' and meetings versus ancestral manual labor like farming and handmade clothing.

By @lulumeservey
29 Twitter/X 2026-06-28 1 min read
Open

@ivanfioravanti says they want to work at SpaceX specifically to try, play with…

Why it matters

@ivanfioravanti says they want to work at SpaceX specifically to try, play with, and interact with the latest models on SpaceX hardware.

  • Elon Musk (tweet dated 2026-06-28) announced Grok 4.5: built on a 1.5T V9 foundation model with Cursor data added in supplemental training, now in private beta at SpaceX & Tesla; early evals are "close to, perhaps exceeding Opus," RL and the Grok Build harness are improving, and SpaceX will release completely trained-from-scratch new models every month in 2026.

@ivanfioravanti expresses a desire to work at SpaceX to access and experiment with the latest models and hardware. They quote Elon Musk's June 28, 2026 tweet announcing Grok 4.5—based on a 1.5T V9 foundation model with Cursor data, in private beta at SpaceX and Tesla, with early evaluations near or exceeding Opus, ongoing RL gains, and monthly from-scratch releases in 2026.

By @ivanfioravanti
30 Twitter/X 2026-06-28 1 min read
Open

Dan McAteer predicts GPT-5.6 will ship and Fable will return the week of June 28…

Why it matters

Dan McAteer predicts GPT-5.6 will ship and Fable will return the week of June 28, 2026, saying Axios reports the government may lift export controls on Fable “as soon as this week” and is unlikely to favor Anthropic over OpenAI.

  • McAteer claims Fable is currently the most capable model but is slow and expensive; GPT-5.6 (Sol) is nearly as capable, far more efficient, and its output tokens cost 40% less than Fable's.
  • He notes community speculation that GPT-5.6 Sol helped crack the Erdős unit-distance problem and emphasizes the value of using Fable and GPT-5.6 Sol together, quipping “The Singularity is delayed by two weeks.”

Dan McAteer predicts GPT-5.6 will ship and Fable will return the week of June 28, 2026, citing Axios that export controls on Fable could be lifted immediately and won’t favor Anthropic over OpenAI. He argues Fable is most capable but slow and costly, while GPT-5.6 Sol is nearly as strong, 40% cheaper per output token, and may have aided solving the Erdős problem.

By @daniel_mac8
31 Twitter/X 2026-06-28 1 min read
Open

Austria (and the UK earlier) has formally invited Anthropic to establish a…

Why it matters

Austria (and the UK earlier) has formally invited Anthropic to establish a presence in the EU; a quoted letter (Andrew Curran) urges joint exploration to 'strategically establish and participate of Anthropic within the European Union' and praises Anthropic's safety-first values.

  • The author acknowledges the EU lags the US in AI compute infrastructure but argues Europe already has nonzero capacity and is committed to expand it; the post calls Americans who insist otherwise 'American idiot.'
  • The post claims the stronger barrier is political: the US government would likely block an outright Anthropic relocation, making an EU-controlled Anthropic subsidiary the more plausible compromise to apply EU law, limit Chinese AI influence, and preserve US market ties, though the author expects Anthropic to remain where compute and supply are concentrated.

Austria's bid to make Anthropic 'European' argues Europe has some existing AI compute and will expand, but the decisive barrier is likely US government consent. The author proposes an EU-controlled Anthropic subsidiary as a realistic compromise—granting EU jurisdiction and reducing Chinese influence—yet predicts Anthropic will probably stay in the US where compute and supply are concentrated.

By @daniel_mac8
32 Twitter/X 2026-06-28 2 min read
Open

On 2026-06-28, OpenAI previewed GPT-5.6 in three tiers—Sol (frontier), Terra, and…

Why it matters

On 2026-06-28, OpenAI previewed GPT-5.6 in three tiers—Sol (frontier), Terra, and Luna—and at the U.S. government's request all three will begin as limited previews to approved partners rather than immediate public releases; OpenAI positions Sol as its most capable model and strongest for cybersecurity.

  • The post ties this to a wider pattern: Anthropic's Fable 5 was pulled after three days and Mythos never went public, arguing that the most capable models are increasingly launching locked and 'government-gated,' turning access into an application process and prompting whether Sol will outperform Fable's brief public run.

OpenAI's GPT-5.6 preview (announced 2026-06-28) unveils three tiers—Sol (frontier), Terra, and Luna—but at the U.S. government's request none will go straight to the public; Sol will launch as a limited preview for approved partners and is billed as the company's most capable model for coding, science, and cybersecurity. The post links this to Anthropic's Fable 5 (pulled after three days) and Mythos (never public), arguing frontier models are increasingly invite-only and asking whether government-gated access is now the norm and if Sol will surpass Fable's brief appearance.

By @alex_prompter
33 Twitter/X 2026-06-28 2 min read
Open

Corey Ganim (post 2026-06-28) proposes five 'Second Brain as a Service'…

Why it matters

Corey Ganim (post 2026-06-28) proposes five 'Second Brain as a Service' verticals: Law firm case intake, Tax firm client-doc, B2B sales call, Restaurant franchise SOP, and Tutoring center curriculum — each stores specific artifacts (e.g., intake notes and jurisdiction rules; client checklists and deadline rules; call transcripts and objections) to automate prep and follow-ups.

  • He provides a 5-step, 45-minute build process: create folders raw/ wiki/ outputs/; dump sources into raw/ (notes, articles, call transcripts, screenshots, SOPs, Slack exports); add one schema file that tells the AI how to organize pages/tags/links/summaries; ask AI to build the reusable wiki; use outputs/ for deliverables like content briefs, proposals, SOPs. Obsidian or local folders are acceptable.
  • Go-to-market and pricing: five target customer types (founder/creator, agency, consultant, local business, ecommerce brand) with Starter builds $1,500–$3,000, Advanced builds $5,000+, and Monthly maintenance $300–$1,000/mo; @TheViableEdge has commercialized this approach and will appear on the Build With AI pod next week.

Corey Ganim outlines five Second Brain-as-a-Service products (law, tax, B2B sales, restaurant franchises, tutoring), a 45-minute build recipe (raw/wiki/outputs; schema file; AI-built wiki), and a GTM/pricing plan targeting five customer types with starter builds at $1.5k–$3k, advanced builds $5k+, and $300–$1k/mo maintenance; he cites @TheViableEdge commercializing the model.

By @coreyganim
34 Twitter/X 2026-06-28 1 min read
Open

Chamath (@chamath) posted on 2026-06-28 that California’s Property Seizure…

Why it matters

Chamath (@chamath) posted on 2026-06-28 that California’s Property Seizure Act—marketed as “The Billionaire Tax”—is now on the ballot and links a 4-minute Hoover Institution video (piped.video/x6k4W5Qzg8U) explaining it.

  • He claims the measure would force a 5% annual cash seizure of all tangible and intangible property, predicts decade-long litigation likely to reach the Supreme Court and fail, and warns this will drive away business builders, cause large revenue losses, deepen California’s budget hole, raise taxes on everyone (especially the middle class), and risk near-bankruptcy and pension retrades.

Chamath (@chamath) warns California’s Property Seizure Act—marketed as “The Billionaire Tax”—is on the ballot and would, he claims, require a 5% annual cash seizure of all tangible and intangible property. He predicts decade-long litigation likely to fail in the Supreme Court, an exodus of business builders, large revenue losses, deeper deficits, higher taxes on middle-class Californians, and possible near-bankruptcy and pension resets.

By @chamath
35 Twitter/X 2026-06-28 1 min read
Open

Shruti (@heyshrutimishra) says her keyboard usage is down to 20% and she now…

Why it matters

Shruti (@heyshrutimishra) says her keyboard usage is down to 20% and she now dictates almost everything, claiming she moves through work 5x faster.

  • She recommends FluidVoice: a free voice-to-text app for macOS that runs fully locally, requires no account or cloud, and is open source under GPLv3.
  • She argues open-source local models eliminate the need to trust cloud services with microphone data ("nothing to send anywhere") and declares the keyboard "officially dead" after writing full docs, emails, and messages without typing.

Shruti (@heyshrutimishra) reports her keyboard usage has fallen to 20% as she dictates almost everything and claims a 5× speedup. She highlights FluidVoice, a free, local macOS voice-to-text tool (GPLv3) with no cloud or accounts, arguing open source plus local models removes trust issues and makes the keyboard "officially dead."

By @heyshrutimishra
36 Twitter/X 2026-06-28 1 min read
Open

On 2026-06-28 @jasonc_nc argued a suggested plan would

Why it matters

On 2026-06-28 @jasonc_nc argued a suggested plan would: use state power to bankrupt current owners; deploy public funds to selected new owners; have a government agency provide below-cost management; ignore how to fund lost property-tax revenue; and thereby halt new construction while expanding the population needing subsidies.

  • He proposes allowing tenants associations to take possession of foreclosed or distressed buildings and convert them to limited-equity cooperatives (citing the Housing Development Fund Corporation model) and also calls to 'fully fund transit' (ref: @crosstown_line).

Jason C. (@jasonc_nc) warns that a proposed housing approach would weaponize state power to displace owners, subsidize chosen new owners and run buildings below cost, eroding property-tax revenue, stopping new construction and increasing subsidy demand. He recommends letting tenant associations form limited-equity co-ops from foreclosed buildings (Housing Development Fund Corporation model) and urges fully funding transit.

By @jasonc_nc
37 Twitter/X 2026-06-28 1 min read
Open

@bscholl (published 2026-06-28) asserts that a common business evil is using…

Why it matters

@bscholl (published 2026-06-28) asserts that a common business evil is using government power under false pretenses to hamstring competition.

  • He names Boeing and claims it was behind most of the 1970s opposition to supersonic travel to protect itself after its supersonic jet was cancelled.
  • @bscholl argues current AI “doomerism” is a repeat of that tactic—incumbents wrap would‑be competitors in red tape as innovation accelerates and "tomorrow’s leading AI company could be one that hasn’t even been founded yet"—and says this should be illegal and socially unacceptable.

@bscholl argues he is pro‑business but warns some firms commit real harm by weaponizing government to block rivals. He claims Boeing drove 1970s opposition to supersonic travel after its jet cancellation and contends today’s AI “doomerism” is the same tactic—incumbents using red tape to stop fast innovation—and must be called out and outlawed.

By @bscholl
38 Twitter/X 2026-06-28 1 min read
Open

Jason C (@jasonc_nc) posted on 2026-06-28 that "virtually every great…

Why it matters

Jason C (@jasonc_nc) posted on 2026-06-28 that "virtually every great neighborhood" was created before the American Planning Association’s predecessors existed and claims that as planning became more controlling the quality of cities "steadily got worse."

  • Carol Walsh (@CarolWalshReal1) highlights Esplanade Street: a main road paired with small one-way side streets that include a bike lane and landscaping, set-back homes and businesses for safer driveway access, a mix of residential, business and multifamily uses, abundant trees, and strong walkability.

Jason C (@jasonc_nc) argues that virtually every great neighborhood predates the American Planning Association’s predecessors and that increasing planning control has steadily worsened urban quality (post dated 2026-06-28). Carol Walsh (@CarolWalshReal1) offers Esplanade Street as a counterexample, detailing one-way side streets, a bike lane, landscaping, set-back mixed uses, trees, and high walkability.

By @jasonc_nc
39 Twitter/X 2026-06-28 1 min read
Open

GLM-5.2, a Chinese open-weight large language model released by 2026-06-28, is…

Why it matters

GLM-5.2, a Chinese open-weight large language model released by 2026-06-28, is claimed to match the performance of currently available models from OpenAI and Anthropic.

  • The post labels GLM-5.2 the “second DeepSeek moment” and cites Innovation Council’s statement that China now has an open-weight model on par with Western offerings.
  • David Sacks (quoted) warns the U.S. must not put its own models in “purgatory” while the world pivots to Chinese technology, calling the situation a global AI race that requires keeping companies moving and not slowing development.

GLM-5.2, released by Chinese developers and highlighted on 2026-06-28, is presented as matching OpenAI and Anthropic models and is called the “second DeepSeek moment.” Quoting Innovation Council and David Sacks, the post argues this parity creates an urgent, competitive AI race and warns against delaying U.S. model deployment or hampering companies’ progress.

By @kimmonismus
40 Twitter/X 2026-06-28 2 min read
Open

@levelsio has coded almost exclusively on his VPS with Claude Code for almost a…

Why it matters

@levelsio has coded almost exclusively on his VPS with Claude Code for almost a year, live-editing his production server; it only 'messed up' twice in 12 months, causing about 10 seconds of downtime each time.

  • He cut deployment time from ~1 minute (local test → push → auto-deploy) to ~3 seconds after abandoning a local Nginx workflow on a new MacBook Pro, and now can start projects and go live in seconds and run tasks overnight with /goal.
  • He endorses cloud/server-first AI coding agents (citing @theo’s ~6-month prediction and @karpathy’s move to cloud), recommends using a staging server for team/regulatory contexts, and maintains 3-2-1 on-/off-site backups.

@levelsio reports coding almost entirely on his VPS with Claude Code for nearly a year, praising instant live edits, device switching, overnight runs via /goal, and rapid go-live from scratch. He experienced only two brief outages (~10 seconds) in 12 months, previously reduced deploys from ~1 minute to ~3 seconds, and argues AI coding agents will run in the cloud, while advising backups and staging for teams.

By @levelsio
41 Twitter/X 2026-06-28 1 min read
Open

Jason (@jasonc_nc) (posted 2026-06-28) argues fire departments should rebrand as…

Why it matters

Jason (@jasonc_nc) (posted 2026-06-28) argues fire departments should rebrand as 'emergency services departments' with a tiny ("really tiny") slice dedicated to fire to adopt a broader operational perspective.

  • He asserts minor amendments to equipment choices would (1) directly improve public safety and (2) allow better use of urban streets to fulfill emergency missions.
  • He accuses fire departments, including the LAFD, of treating speed of response as the ~sole~ safety metric, claiming that focus drives departments and unions to gaslight the public and fabricate impacts.

Jason (@jasonc_nc) argues fire departments should embrace being "emergency services departments" with only a "tiny" portion devoted to fire; by making minor equipment changes they could directly improve public safety and use urban streets more effectively. He specifically calls out the LAFD and says an overemphasis on response speed, treated as the sole safety metric, leads departments and unions to gaslight the public and fabricate impacts.

By @jasonc_nc
42 Twitter/X 2026-06-28 1 min read
Open

Sarvesh Shrivastava (@bloggersarvesh) claims on 2026-06-28 that using Claude…

Why it matters

Sarvesh Shrivastava (@bloggersarvesh) claims on 2026-06-28 that using Claude (Claude Cowork) for $20/month can 'run like a $10k SEO team' if users stop treating it like a beginner.

  • Operational fixes to achieve that: preload full business context (services, locations, keywords, competitors, ideal customers); ask strategist prompts such as 'What's blocking our organic growth right now?'; and run competitor gap analysis: 'Compare my website with my top 3 competitors. Show me what they rank for that I don't.'
  • Tactical guidance includes clustering keywords by intent (informational, commercial, transactional) rather than volume, and optimizing for AI search with prompts like 'Why would AI recommend this business over competitors? What should I clarify or add?' — plus a free pack of 20 Claude SEO prompts and an infographic.

Sarvesh Shrivastava (@bloggersarvesh) argues that Claude Cowork can replace a $10k SEO team for $20/month by shifting from vague chatbot prompts to a structured SEO system: preload full business context, ask strategist-level questions, run competitor gap analysis, cluster keywords by intent, and optimize for AI search. He published 20 free prompts and an infographic on 2026-06-28.

By @bloggersarvesh
43 Twitter/X 2026-06-28 1 min read
Open

Created a dynamic, auto-updating chart (levels.io/bitcoin-4-year-cyc…) that…

Why it matters

Created a dynamic, auto-updating chart (levels.io/bitcoin-4-year-cyc…) that visualizes Bitcoin's repeating 4-year cycles and keeps adding new cycles over time; the chart marks October 31, 2026 as the start of the next cycle and theoretically the cycle bottom.

  • Asserts Bitcoin crashes at the end of each 4-year cycle (listed crash years: 2014, 2018, 2022, 2026) and that the next up-cycle runs 2027–2031.
  • Attributes the 4-year pattern to the halvening (miner reward cut about every four years) and to recurring market sentiment dynamics—every 4th year people 'get super sad' and selling/pumping pushes price down so others can buy cheap.

levelsio published a dynamic chart that constantly adds Bitcoin 4-year cycles, marking October 31, 2026 as the next cycle start and theoretical bottom. He lists crash years 2014, 2018, 2022 and 2026, projects an up-cycle from 2027–2031, and links the pattern to the quadrennial halvening and repeated sentiment-driven selloffs.

By @levelsio
44 Twitter/X 2026-06-28 1 min read
Open

Codex usage increased 6x since February (2026), now exceeding 5 million weekly…

Why it matters

Codex usage increased 6x since February (2026), now exceeding 5 million weekly active users; nearly 100% of OpenAI employees reportedly use the Codex desktop app regularly.

  • Andrew Ambrosino, lead of the Codex desktop app at OpenAI, says his mission is to build "the best desktop app that has ever existed" and claims the app would have flopped if shipped in November instead of February because the underlying model changed.
  • Key topics from the conversation: OpenAI PMs' "zone defense" operating model, why AI performs poorly at design, the professional skill of "taste," how Ambrosino uses Codex to run workflows, and his vision for integrating Codex with ChatGPT.

Andrew Ambrosino, who leads the Codex desktop app at OpenAI, reports the product grew 6x since February to over 5M weekly active users and is used by nearly all OpenAI employees. He frames his mission as building "the best desktop app" and argues timing and model upgrades (November vs. February) determined the app's success; the interview covers PM "zone defense," design limits of AI, "taste," workflows, and Codex+ChatGPT plans.

By @lennysan
45 Twitter/X 2026-06-28 3 min read
Open

Chubby♨️ (@kimmonismus) asserts that Dario Amodei’s alleged “fearmongering” was…

Why it matters

Chubby♨️ (@kimmonismus) asserts that Dario Amodei’s alleged “fearmongering” was not the reason the U.S. embargoed Fable 5 and GPT-5.6 and calls that attribution mistaken.

  • The author criticizes Anthropic’s handling of U.S. authorities—citing February when Anthropic refused to cooperate with the U.S. Department of Defense and poor phone availability—but insists those failures alone wouldn’t make the U.S. impose an embargo.
  • Chubby♨️ argues the embargo was driven by national-security fears that Fable 5/GPT-5.6 could be exfiltrated to China (for example via distillation) and used for cyberattacks, espionage, or other major damage; they say regulators weighed those risks against economic harm and imposed enormous investment/mitigation requirements, likening them to being even larger than the Manhattan Project.
  • The author warns public access may be blocked as a result and advocates open source as the correct solution to avoid concentrated control.

Chubby♨️ (@kimmonismus) argues that the U.S. embargo on Fable 5 and GPT-5.6 was not the result of Dario Amodei’s so‑called “fearmongering” but rather a calculated national‑security decision: while Anthropic’s February refusal to cooperate with the U.S. Department of Defense and other communication failures are fair criticisms, the U.S. government—backed by advisers, NSA and other experts—would not halt a technology and impose massive financial and investment requirements solely because of a CEO’s statements. The author believes regulators fear the models could be distilled or otherwise transferred to China and used for cyberattacks, espionage, or large‑scale harm; they say the resulting mitigation demands are enormous (compared rhetorically to the Manhattan Project). Chubby♨️ expresses concern that public access may be curtailed and endorses open source as the preferable response.

By @kimmonismus
46 Twitter/X 2026-06-28 1 min read
Open

FreeLLMAPI (GitHub repo by tashfeenahmed) aggregates 11 free LLM…

Why it matters

FreeLLMAPI (GitHub repo by tashfeenahmed) aggregates 11 free LLM providers—Google, Groq, Mistral, OpenRouter, GitHub Models, Cohere, Cloudflare, HuggingFace, Z AI, Ollama, and Kimi—and claims the combined free tiers provide about 1.7 billion tokens per month.

  • The project exposes a single OpenAI-compatible /v1 endpoint with one API key and an automatic fallover router that picks the best model, falls back on provider rate-limits, tracks usage per key, enforces free-tier caps, stores keys encrypted, and supports chat, embeddings, images, audio, tools, and streaming.
  • Posted by @hasantoxr on 2026-06-28, the author claims FreeLLMAPI can "save you a stupid amount of money" for AI side projects by consolidating free inference behind one endpoint (repo: github.com/tashfeenahmed/fre…).

FreeLLMAPI aggregates 11 free LLM providers into one OpenAI-compatible /v1 endpoint and claims roughly 1.7 billion free tokens per month. It provides an automatic fallover router, single API key, encrypted key storage, per-key usage tracking, and support for chat, embeddings, images, audio, tools, and streaming to reduce inference costs for AI side projects.

By @hasantoxr
47 Twitter/X 2026-06-28 1 min read
Open

In 1955 Sony (only months old) showed the TR-55 radio; Bulova offered to buy…

Why it matters

In 1955 Sony (only months old) showed the TR-55 radio; Bulova offered to buy 100,000 units on the condition Sony's logo be removed — cofounder Akio Morita refused, overruling partner Masaru Ibuka and the Tokyo board who wanted to accept the order.

  • Morita predicted Sony would be as known as Bulova within 50 years and later called the refusal his best business decision; Bulova was sold to Citizen in 2008 for $250 million.

Akio Morita refused a 1955 Bulova offer to buy 100,000 TR-55 radios if Sony's name was removed. With Sony only months old, Morita overruled partners who wanted the revenue, predicting Sony would be as famous as Bulova in fifty years; he later called it his best decision, while Bulova was sold to Citizen for $250 million in 2008.

By @aakashgupta
48 Twitter/X 2026-06-28 3 min read
Open

Elon Musk turns 55 on 2026-06-28; the author frames his lifetime of projects as…

Why it matters

Elon Musk turns 55 on 2026-06-28; the author frames his lifetime of projects as driven by humanity-first motives rather than money and says his plans ‘span decades’.

  • Founded SpaceX and demonstrated reusable rocket landings despite widespread skepticism, with the stated goal of making humanity multiplanetary to ‘backup the light of consciousness.’
  • Founded Tesla, pressured legacy automakers toward sustainable energy, and (per the author) open-sourced Tesla patents so the whole industry could move faster.
  • Built Starlink and—according to the post—immediately switched it on to keep Ukraine connected when it was cut off; developed Neuralink that ‘put a chip in a man’s brain’ enabling him to move a cursor by thought; and is building Optimus to replace backbreaking human labor.

Elon Musk is presented as a 55-year-old ‘guardian Angel’ whose ventures—SpaceX, Tesla, Starlink, Neuralink, Optimus, and AI/Grok—are cast as deliberate, often confrontational efforts to secure humanity’s future. The post credits SpaceX with reusable rocket landings and a multiplanetary mission to preserve consciousness; Tesla with forcing an industry-wide shift to sustainable energy via open-sourced patents; Starlink with restoring connectivity to Ukraine; Neuralink with enabling a paralyzed man to move a cursor by thought; and Optimus with targeting scarcity by automating backbreaking work. It asserts Musk paid $44 billion to buy Twitter to expose censorship and restore a public square, created Grok as a truth-seeking AI, backed Trump at personal cost, exposed institutional coverups of child abuse, called out unelected judicial power, warned about declining birth rates, and emphasizes that his motivation has always been humanity rather than money.

By @tetsuoai
49 Twitter/X 2026-06-28 1 min read
Open

swyx asked what the AIEWF "Stress Curve" looked like, defining it explicitly as…

Why it matters

swyx asked what the AIEWF "Stress Curve" looked like, defining it explicitly as the Gini coefficient over time from 0 (ticket launch) to sold out.

  • AI Engineer announced the 2026 World's Fair is completely sold out: leadership track for CTOs & VP AI's, tomorrow's workshops, and all late-bird tickets are sold.
  • Organizers opened limited overflow tickets for the expo and engineering passes (no seating guaranteed), advertised 65 free side events across San Francisco, and scheduled New Engineer Orientation at Moscone tonight from 5p–9p; they reiterated a commitment to publish content free on YouTube.

swyx framed the AIEWF "Stress Curve" as the Gini coefficient over time from ticket launch to sold out; AI Engineer replied that the 2026 World's Fair is completely sold out — leadership track, workshops, late-bird tickets — while offering limited overflow tickets, 65 free SF side events, and a Moscone orientation 5–9pm.

By @swyx
50 Twitter/X 2026-06-28 1 min read
Open

@0xd1namit (published 2026-06-28) claims users demand features Polymarket will…

Why it matters

@0xd1namit (published 2026-06-28) claims users demand features Polymarket will never add—examples include copytrading, trader tracking, and extensive analytics.

  • PolyHelper is presented as the 'pro version' that layers those advanced features onto the Polymarket interface so users can keep Polymarket's retail UI while gaining pro tools.

PolyHelper positions itself as the 'pro version' of Polymarket, adding features Polymarket won't implement — copytrading, trader tracking, and advanced analytics — while letting users keep the Polymarket interface. @0xd1namit (2026-06-28) argues this is necessary because Polymarket targets retail and cannot overload its native UI.

By @0xd1namit
51 Twitter/X 2026-06-28 1 min read
Open

Liebig's 1835 method uses ammoniated silver nitrate (Ag⁺ suspended in ammonia)…

Why it matters

Liebig's 1835 method uses ammoniated silver nitrate (Ag⁺ suspended in ammonia) plus a reducing sugar (e.g., glucose); the sugar's aldehyde donates electrons so each Ag⁺ gains one electron and precipitates as metallic silver while the sugar oxidizes to gluconic acid.

  • Because the reduction runs slowly across the whole surface, silver plates out atom-by-atom into a continuous film about 100 nanometers thick (thinner than a single wavelength of red light), producing a mirror without polishing.
  • Before 1835, mirrors were made by pressing tin foil onto glass and flooding with mercury to form an amalgam; that produced cloudy finishes and toxic mercury vapor exposure, which Liebig's pourable bath replaced.

@aakashgupta explains that Liebig's 1835 ammoniated silver‑nitrate + reducing‑sugar method plates silver atom-by-atom across glass, producing a continuous film about 100 nanometers thick—thinner than a wavelength of red light. The sugar's aldehyde reduces Ag⁺ to Ag⁰ while oxidizing to gluconic acid, replacing earlier mercury‑amalgam mirror techniques that caused toxic exposure.

By @aakashgupta
52 Twitter/X 2026-06-28 1 min read
Open

The 2026 World's Fair is "completely sold out" (post published 2026-06-28) and is…

Why it matters

The 2026 World's Fair is "completely sold out" (post published 2026-06-28) and is billed as "the largest AI industry expo on earth"; the Leadership track for CTOs & VP AI's, tomorrow's workshops, and ALL late-bird tickets are sold out.

  • There are 65 side events still free across San Francisco; organizers have opened limited overflow tickets for the expo and engineering passes — no seating guaranteed and sessions are first-come, first-served.
  • New Engineer Orientation is scheduled tonight (June 28) at Moscone from 5p–9p; organizers expect extremely heavy last-minute registration, ask attendees to help load-balance across days, and pledge to publish all AI engineering content for free on YouTube.

The 2026 World's Fair is completely sold out and promoted as "the largest AI industry expo on earth," with the Leadership track for CTOs & VP AI's, workshops, and all late-bird tickets gone. Organizers opened limited overflow expo/engineering tickets (no guaranteed seating), note 65 free side events around SF, and host New Engineer Orientation at Moscone tonight (June 28, 5–9pm).

By @swyx
53 Twitter/X 2026-06-28 2 min read
Open

Yishan (@yishan) re-ran the paper authors' open-sourced evaluation framework…

Why it matters

Yishan (@yishan) re-ran the paper authors' open-sourced evaluation framework (github.com/aiden-ygu/health-...) and published his reproduction and results at github.com/ywong137/health-a..., successfully reproducing tests on the public datasets the paper included.

  • Baseline re-runs on the paper's older models showed little performance drift; the most capable current model tested (GPT-5.5 Pro) scored 79/100 versus the prior best 69/100 — a clear improvement that nonetheless remains below the paper's standard for clinical readiness.
  • The paper's 'fit for reliable medical use' criterion demands robustness to perturbation/bad data, correct refusal/abstention when information is insufficient, and clinically valid reasoning without hallucinations; Yishan was unable to reproduce those qualitative robustness evaluations in his automated reruns.
  • Some test data could not be open-sourced because of copyrights held by JAMA/NEJM, but partial open-sourcing of the evaluation code enabled rapid external reproduction; Yishan urges all AI-paper authors to release evaluation frameworks so others can re-run tests on newer models.

Yishan (@yishan) used the authors' publicly released evaluation framework (github.com/aiden-ygu/health-...) and posted his reproduction at github.com/ywong137/health-a..., re-running the paper's radiology-interpretation tests on both the original models and current frontier models. He found little baseline drift and measured an improvement from 69/100 (prior best) to 79/100 for GPT-5.5 Pro, but stresses this numeric gain still falls short of the paper's stricter clinical-readiness requirements—robustness under perturbation, proper abstention when data are insufficient, and clinically valid reasoning without hallucinations—tests he could not replicate qualitatively. He notes some testing data remained closed due to JAMA/NEJM copyrights, praises the authors for releasing what they could, and argues that open-sourcing evaluation code is essential so others can reproducibly run benchmarks on newer models over time.

By @emollick
54 Twitter/X 2026-06-28 1 min read
Open

On 2026-06-28, Fox News Media issued an apology after Kevin O'Leary retracted…

Why it matters

On 2026-06-28, Fox News Media issued an apology after Kevin O'Leary retracted claims about opponents of his planned Utah data center and stated the network is aware of no evidence they were funded by or acting in coordination with the Chinese Communist Party.

  • Gary Marcus reacted to Acyn's tweet reporting the apology with a single clown emoji ("🤡"); Acyn's post includes a video and links to the original tweet (nitter.net/Acyn/status/2071024356948590922).

Gary Marcus posted a single clown emoji ("🤡") in response to Acyn's June 28, 2026 tweet reporting that Fox News Media apologized after Kevin O'Leary retracted claims about opponents of his planned Utah data center. O'Leary said he has "no evidence" the opponents were funded by or coordinated with the Chinese Communist Party; Fox acknowledged the error.

By @GaryMarcus
55 Twitter/X 2026-06-28 2 min read
Open

cryptopunk7213: Over the next 6 months, the most valuable capabilities will be…

Why it matters

cryptopunk7213: Over the next 6 months, the most valuable capabilities will be per-task model choice, multi-model routing/harnesses, caching to reduce token spend, and AI-cloud management of memory/databases; the bottleneck is shifting from LLM capacity to aggregation.

  • Brian Armstrong: Coinbase is defaulting to open-weight models (GLM 5.2 and Kimi 2.7) via an LLM gateway to control cost rather than hard caps; 91% of employees never hit usage caps so cheaper defaults steer behavior without friction.
  • Armstrong operational wins: preprocess routing and cache-awareness raised LibreChat cache hit rate from 5% to 60%; combined better defaults, routing, caching, lean context, and visibility cut AI spend nearly in half while token usage keeps growing.

cryptopunk7213 and Brian Armstrong argue the near-term value lies in model selection per task, multi-model routing/harnesses, caching, and AI-cloud memory/DB management as aggregation becomes the bottleneck. Armstrong reports defaulting to GLM 5.2 and Kimi 2.7, 91% of employees never hit caps, a LibreChat cache hit improvement from 5%→60%, and nearly halved AI spend while tokens grow.

By @cryptopunk7213
56 Twitter/X 2026-06-28 1 min read
Open

Daniel Mac (@daniel_mac8) predicts GPT-5.6 will ship and Fable will return the…

Why it matters

Daniel Mac (@daniel_mac8) predicts GPT-5.6 will ship and Fable will return the week of 2026-06-28; Axios reportedly says the U.S. government may lift export controls on Fable as soon as that week and is unlikely to favor Anthropic over OpenAI.

  • Fable is claimed to be the most capable model but is slow and expensive; GPT-5.6 is described as nearly as capable and far more efficient, with output tokens costing 40% less than Fable's.
  • Independent tester Mikhail Parakhin reports GPT-5.6 (Max/Sol) outperforms Opus 4.8 across tasks (and is slightly faster depending on load), while Fable is stronger at coding and GPT-5.6 is stronger on agentic/experiment-running workloads—he used Fable to write code and 5.6 to run experiments.

Daniel Mac (@daniel_mac8) predicts GPT-5.6 will ship and Fable will return the week of 2026-06-28, citing Axios reporting that export controls on Fable may be lifted and the government won’t favor Anthropic over OpenAI. He frames Fable as the most capable but slow/expensive model, GPT-5.6 as nearly as capable and 40% cheaper per output token, and highlights combining Fable (coding) with GPT-5.6 (agentic experiments); tester Mikhail Parakhin corroborates 5.6 beating Opus 4.8 and being better at agentic tasks while Fable is stronger on coding.

By @daniel_mac8
57 Twitter/X 2026-06-28 1 min read
Open

ODNI is drafting an unclassified report, mandated by the FY2026 Intelligence…

Why it matters

ODNI is drafting an unclassified report, mandated by the FY2026 Intelligence Authorization Act and due before December 2026, intended to disclose personal wealth, offshore accounts, and global real estate of Xi Jinping and the 25-member Politburo.

  • The post and UnveiledChina cite Congressional Research Service data tracing over $700 million to Xi relatives and predict the ODNI report will reveal billions more hidden across Hong Kong, Macau, and Western banks.
  • @dennisw5 states Xi’s sister runs a vast business empire and argues that exposing these secret fortunes will shatter CCP anti-capitalist propaganda and undermine public trust in the Party.

UnveiledChina and @dennisw5 say the ODNI is drafting a public, unclassified report — mandated by the FY2026 Intelligence Authorization Act and due before December 2026 — that will expose hidden fortunes of Xi Jinping and the 25-member Politburo; the thread cites CRS data linking over $700 million to Xi relatives and argues revelations about Xi’s sister and offshore holdings will undercut CCP propaganda and public trust.

By @dennisw5
58 Twitter/X 2026-06-28 1 min read
Open

Paris Fire Brigade (serving a metro of 10M) posts average fire response of 12…

Why it matters

Paris Fire Brigade (serving a metro of 10M) posts average fire response of 12 minutes and EMS 7 minutes; Houston reports 90th‑percentile times of 13 minutes for fire and 15 minutes for EMS.

  • Smaller apparatus and alternative vehicles can improve medical response times without sacrificing firefighting capability; local US practice favors large apparatus and wide lanes that hinder such options.
  • Road safety measures matter more: a 10% reduction in road deaths in a typical US city would save more lives than all fire deaths; road diets/traffic calming cut crashes ~20–40% and fatalities/serious injuries ~30–40%, and the author claims deaths per fire have seen zero improvement for ~5 decades and safer streets could save 3–4× that number.

Jason (@jasonc_nc) contrasts emergency response metrics (Paris: 12 min fire, 7 min EMS; Houston 90th‑percentile: 13 min fire, 15 min EMS) with traffic‑safety opportunities: US lane standards (12–14 ft lanes, 26 ft two‑lane minimum) hinder smaller apparatus and calming measures, while narrower European lanes (Stockholm ~9.8 ft) and road diets cut crashes and could save far more lives than further shrinking fire response times.

By @jasonc_nc
59 Twitter/X 2026-06-28 1 min read
Open

Arvind Krishna (IBM CEO) said in a podcast that “Year one of enterprise AI is a…

Why it matters

Arvind Krishna (IBM CEO) said in a podcast that “Year one of enterprise AI is a net loss,” causing 80% of companies to currently “look like they have failed,” but “Year two is precisely when the 10x return kicks in.”

  • The author argues CSPs’ CAPEX can grow in absolute terms while CAPEX-to-revenue ratios decelerate once AI drives a 10x revenue inflection, creating a virtuous flywheel for operators.
  • The author rejects judging CAPEX solely by present cash flow and insists cash flow should be assessed by pulling forward expected future cash flow from AI-driven revenue gains.

Author @jukan05 cites IBM CEO Arvind Krishna’s podcast claim that enterprise AI incurs a net loss in year one (making 80% of firms look failed) but delivers a 10x return in year two. The post argues CSPs’ revenues will expand at that inflection, allowing absolute CAPEX growth with falling CAPEX/revenue ratios and a virtuous flywheel, so cash flow must account for future AI-driven gains.

By @jukan05
60 Twitter/X 2026-06-28 1 min read
Open

Andrew Ambrosino (@ajambrosino) leads the Codex desktop app team at OpenAI; Codex…

Why it matters

Andrew Ambrosino (@ajambrosino) leads the Codex desktop app team at OpenAI; Codex usage has grown 6x since February to over 5 million weekly active users, and nearly 100% of OpenAI employees use the Codex app regularly (not just engineers).

  • Andrew’s stated mission is to build "the best desktop app that has ever existed, full stop." He claims the Codex app would have flopped if shipped in November instead of February (same product — only the model changed).
  • The conversation outlines OpenAI’s "zone defense" PM model, explains why AI currently performs poorly on design, defines "taste" as a professional skill, describes how Andrew uses Codex in his workflows, and previews his vision for Codex + ChatGPT.

Andrew Ambrosino, who leads development of OpenAI’s Codex desktop app, reports a 6x increase in usage since February to over 5 million weekly active users and near-universal internal adoption at OpenAI. The interview covers product timing (November vs. February), OpenAI’s "zone defense" PM approach, design limitations of AI, professional "taste," and plans for tighter Codex–ChatGPT integration.

By @lennysan
61 Twitter/X 2026-06-28 1 min read
Open

Andrew Ambrosino leads the Codex desktop app team at OpenAI and states his…

Why it matters

Andrew Ambrosino leads the Codex desktop app team at OpenAI and states his mission is to build “the best desktop app that has ever existed, full stop.”

  • Codex usage has grown 6× since February 2026, reaching over 5 million weekly active users; nearly 100% of OpenAI employees use the Codex app regularly (not just engineers).
  • In a Lenny Rachitsky interview (published 2026-06-28), Ambrosino explains OpenAI PMs' “zone defense” model, argues why AI is bad at design, says the product would have flopped if shipped in November instead of February due to a model change, and outlines his view of “taste,” workflows, and Codex+ChatGPT integration.

Andrew Ambrosino leads OpenAI’s Codex desktop app and aims to build “the best desktop app that has ever existed.” Codex grew 6× since February 2026 to over 5 million weekly active users and is used by nearly all OpenAI employees. In a June 28 interview he covers PM “zone defense,” design limits of AI, timing effects from model updates, and Codex+ChatGPT plans.

By @jxnlco
62 Twitter/X 2026-06-28 1 min read
Open

On 2026-06-28 @kimmonismus asserted that “the open internet was the training…

Why it matters

On 2026-06-28 @kimmonismus asserted that “the open internet was the training ground for modern AI,” citing “billions of texts, forum posts, code repositories, Wikipedia articles, scientific papers, blogs, and discussions” as the collective data foundation for current models.

  • They argue open-source is essential to prevent AI from being locked behind proprietary walls and that if intelligence access is increasingly regulated or geopolitically controlled, open source becomes a question of power rather than merely a technical ideal.
  • Despite quoting a critique—“open source is kind of a distraction you still can’t really see what’s happening inside the model, so it’s not truly 'free'”—@kimmonismus says a referenced clip made them “even more convinced” of open source.

The post by @kimmonismus (2026-06-28) claims modern AI was built on the open internet’s “billions of texts, forum posts, code repositories, Wikipedia articles, scientific papers, blogs, and discussions,” and insists AI must not be sequestered behind proprietary walls. They frame open source as a geopolitical power issue and, despite transparency critiques, remain strongly pro–open source.

By @kimmonismus
63 Twitter/X 2026-06-28 1 min read
Open

Aakash Gupta: $60,000 for a private cross‑country flight buys 60 first‑class…

Why it matters

Aakash Gupta: $60,000 for a private cross‑country flight buys 60 first‑class coast‑to‑coast tickets; on a 5‑hour flight the total time difference is ~45 minutes, implying about $1,333 saved per minute by flying private.

  • Modern commercial premium cabins (examples: LAX–JFK Delta One) now offer lie‑flat beds, door‑closing suites, Michelin‑trained meals and extensive entertainment—Gupta says the private‑jet experience sold for $60k was commercial first‑class in 2019; private flying only mathematically favors legs under 2 hours, groups of 6+ or routes that require connections.
  • Rob Gronkowski (who earned roughly $100M in NFL contracts) says he flies commercial coast‑to‑coast because it costs ~$1,000 vs $60k–$70k for private; Gupta uses Gronk’s choice to illustrate that even high earners get better ROI flying premium commercial.

Aakash Gupta argues Rob Gronkowski is correct: a $60,000 private jet across the U.S. is poor ROI because $60k buys 60 first‑class coast‑to‑coast tickets. A 5‑hour flight saves ~45 minutes—about $1,333 per minute—while modern lie‑flat suites, Michelin meals and private‑like service make commercial travel sufficient; private pays only for <2‑hour hops, groups of 6+, or complex connections.

By @aakashgupta
64 Twitter/X 2026-06-28 2 min read
Open

Mixture of Agents (MoA) runs several LLMs in parallel and an aggregator…

Why it matters

Mixture of Agents (MoA) runs several LLMs in parallel and an aggregator synthesizes their answers; Together AI's MoA scored 65.1% on AlpacaEval 2.0 vs GPT-4o's 57.5% using only open models, and the exam used to claim superiority was not released and was authored by the lab.

  • Ensembles cannot create new capabilities: MoA only recovers more of what member models collectively know and reduces errors from the strongest model, so gains reflect coordination across the public frontier rather than new model knowledge.
  • MoA disproportionately helps benchmarks that use LLM judges and reward longer, fuller answers (AlpacaEval 2.0, MT-Bench); Nous/Hermes advertises '8% higher than Opus 4.8 and 11% higher than GPT 5.5' on an unshipped benchmark, while MoA increases latency (response time floors at the slowest model per sequential layer) and yields single-digit accuracy gains at much higher time cost.

Mixture of Agents (MoA) ensembles parallel LLM drafts and an aggregator synthesis; Together AI's MoA hit 65.1% on AlpacaEval 2.0 vs GPT-4o 57.5% using open models, but the exam wasn't released and was written by the lab. MoA recovers collective knowledge and favours judge-driven benchmarks that reward long answers, while adding sequential-layer latency for only single-digit accuracy gains.

By @aakashgupta
Less notable

Lower-priority items kept for completeness and search.

6 items · open
1 Twitter/X 2026-06-28 1 min read
Open

@ianmiles (2026-06-28) asserts the British public has voted AGAINST more…

Why it matters

@ianmiles (2026-06-28) asserts the British public has voted AGAINST more migration "election after election, for decades," claiming politicians repeatedly ignore this mandate.

  • @ianmiles accuses every governing party of "flooding the country with them in untold numbers" and demands immediate mass deportations and remigration, warning "Britain will cease to exist as it once was" if action doesn't start now.
  • The Home Office counterpoint notes new safe and legal refugee routes—community sponsorship, refugee study and work routes—with numbers starting small, strict screening, and approved sponsors to prevent abuse.

@ianmiles (28 June 2026) argues voters have opposed more migration for decades but successive governments continue to "flood" the UK with migrants. He calls for immediate mass deportations and remigration, claiming Britain’s identity is at stake. The post contrasts with a Home Office announcement of limited, screened safe/legal refugee and sponsorship routes.

By @ianmiles
2 Twitter/X 2026-06-28 1 min read
Open

The viral claim that “9 million women were burned as witches” is false; modern…

Why it matters

The viral claim that “9 million women were burned as witches” is false; modern historians estimate about 40,000–60,000 executions across three centuries in Europe.

  • Roughly 75–85% of those executed were women, and most victims were not burned alive—many were strangled first or hanged.
  • The 9-million figure is repeatedly circulated (especially every October) and went viral because it felt righteous; the author argues the inflated number does a disservice by making the true deaths easier to dismiss.

Author @meishato debunks the recurring claim that '9 million women were burned as witches,' noting it resurfaces every October. Citing modern historians, they put European witchcraft executions at roughly 40,000–60,000 over three centuries, 75–85% female; most victims were strangled or hanged. The author argues the inflated number undermines the real victims.

By @meishato
3 Twitter/X 2026-06-28 1 min read
Open

On 2026-06-28 @bswud (citing @vpostrel) argued modernism is more popular in…

Why it matters

On 2026-06-28 @bswud (citing @vpostrel) argued modernism is more popular in California and Chicago than in Europe or the US East Coast.

  • He attributes the transatlantic aesthetics divide to building type and quality: the US—especially California—has many premium private modernist houses, while UK postwar modernism was dominated by low-quality social housing and public buildings due to prolonged postwar austerity.
  • He cites The Priory in the Cator Estate, Blackheath as a rare British example of premium postwar modernism with high-quality materials and details; despite features like horizontal windows, uPVC, raw sills and lack of cornice, he expects a significant minority or maybe a majority of survey respondents would say they like it.

Popular modernism: @bswud (citing @vpostrel) argues modernist architecture is more popular in California and Chicago than in Europe or the US East Coast because American postwar modernism often appears in premium private houses—especially in California—while UK postwar modernism was dominated by low-quality social housing. He points to The Priory, Cator Estate, Blackheath as a rare premium British example.

By @bswud
4 Twitter/X 2026-06-28 1 min read
Open

At age 40, studios offered Meryl Streep three "witch" roles in a single year; by…

Why it matters

At age 40, studios offered Meryl Streep three "witch" roles in a single year; by then she already held two Oscars (Kramer vs. Kramer at 30, Sophie's Choice at 33), signaling industry typecasting of women over 40 as crones or villains by 1989.

  • In 1989 Streep rejected the title role in She-Devil and chose Mary Fisher, a glamorous romance novelist; that turn revealed her comic timing and earned her a Golden Globe nomination in the comedy category—her first after a decade of straight drama—and led to roles in Postcards from the Edge (1990) and Death Becomes Her (1992).
  • Streep holds a record 21 Oscar nominations (the next closest actor has 12); eight nominations came before she turned 40 and 13 after, demonstrating she spent roughly thirty years overturning Hollywood's verdict on actresses over 40.

Meryl Streep faced typecasting at 40 when studios offered her three "witch" roles despite already owning two Oscars. In 1989 she chose the glamorous Mary Fisher in She-Devil, exposing her comic range and earning a Golden Globe comedy nomination; subsequent choices (Postcards from the Edge, Death Becomes Her) and a record 21 Oscar nominations (8 before 40, 13 after) reversed Hollywood's verdict over three decades.

By @aakashgupta
5 Twitter/X 2026-06-28 1 min read
Open

Elon Musk (tweet, 2026-06-28) asserts critics who claim “millions” died cannot…

Why it matters

Elon Musk (tweet, 2026-06-28) asserts critics who claim “millions” died cannot cite a single named person who allegedly died.

  • Whole Mars Catalog (@wholemars) argues Musk was only an advisor to the President, not a ‘king,’ and claims that if alleged spending cuts were really killing millions of children the President, Congress, or courts would have intervened.
  • Whole Mars Catalog claims attacks were meant to neutralize Musk because he was cutting government “waste and fraud,” and that his low-cost space launches and profitable electric cars show he could have reformed the federal government if not targeted by a “fraudsters anti-viral response.”

Elon Musk and Whole Mars Catalog claim the accusation that he “killed millions” is baseless because no individual names have been produced. They argue Musk was only an advisor to the President, so institutional actors (President, Congress, courts) would have stopped any truly deadly spending cuts. They further assert critics targeted him to protect entrenched waste and fraud, despite his achievements in cheaper space launches and profitable electric cars.

By @elonmusk
6 Twitter/X 2026-06-28 2 min read
Open

In 1972 off California an orca bit a surfer, took one taste, released him, and…

Why it matters

In 1972 off California an orca bit a surfer, took one taste, released him, and the man survived after receiving over 100 stitches—used by the author as direct evidence that orcas reject unfamiliar prey.

  • Orcas hunt by culturally transmitted, population-specific 'inherited menus': Pacific Northwest resident pods eat salmon but ignore seals, while Bigg's (transient) orcas hunt seals but not salmon; the author asserts no pod has ever added humans to its prey list.
  • Orcas locate prey with echolocation from hundreds of meters, switch to eyesight at close range to match a learned 'search image,' and researchers have observed whales abort approaches when a target fails to fit that template—specialization reduces improvisation and likely explains rare predation on humans.

Orca hunting culture is presented as the reason humans are rarely prey: the author cites a 1972 California case where an orca bit a surfer, then released him after a single taste (the surfer needed over 100 stitches). Orcas, the post argues, inherit narrow prey preferences across generations, use long-range echolocation and close-range visual matching to a learned search image, and abort when targets don't match—so no pod has ever learned to hunt people.

By @aakashgupta