Briefing

Jigar Shah (@JigarShahDC) lays out an AI infrastructure roadmap: as of 2026-05-11 there are zero operational 1+ GW campuses (largest 300–600 MW; 1+ GW expected 2027–28). He argues 15–20 1–5 GW campuses are needed for frontier training by 2030, 50–100 regional 100–500 MW hubs for inference, while 100 kW telco and residential 7B on-device options have limited roles.

DFlash promises an 8.5× real-world speedup for existing LLMs (same model, same accuracy), increasing decoding from 48.5 to 415 tokens/sec in the demo. Rather than one-token speculative drafts, it uses a tiny block-diffusion model to predict many tokens in parallel and then has the large model verify batches; it integrates with vLLM, SGLang, and Transformers and supports Qwen, Llama, Kimi, and gpt-oss.

Colossus One, the 300 MW, 220,000‑GPU supercomputer Elon built to power Grok, was leased in full to Anthropic on May 7, 2026. Musk folded xAI into SpaceX the same afternoon, meaning the Grok-designed flagship will run Anthropic's Claude weeks before a planned SpaceX IPO; markets reacted sharply to the deal.

Simulation is the practical lever for building reliable LLM-driven products, argues @LinusEkenstam (2026-05-06): treating context engineering as a simulation problem lets teams convert brittle zero-shot outputs into robust, repeatable behavior. He recommends constructing 'jigs'—software equivalents of woodworking jigs or game-engine environments and node-based systems—that expose many levers (Flocurve’s outreach jig uses 100+ parameters) so teams can run thousands of pathways, measure outcomes, and triangulate consensus across signals. The key differentiator is evaluation: without rigorous scoring you get intermittent 'SLOP' rather than scalable performance. Ekenstam’s final point is strategic: while model training is public, winning builders treat inference like training by iterating privately—pre-runs, re-runs, and systematic evaluation—to produce consistent, production-grade agentic systems.

@SlBrandin argues regulated-utility gatekeeping has broken an otherwise natural, market-driven dynamic between wholesale generation, load, and transmission. He lays out ideal interactions — load siting for end use, generation siting for load, transmission siting to reduce congestion — and blames misaligned utility incentives for today’s gridlock and inefficient planning.

Enterprise AI buying is shifting from model-line items to the build-room: in 48 hours multiple announcements (Anthropic’s new enterprise services company with Blackstone/H&F/Goldman at ~ $1.5B, Bloomberg’s report of >$4B for OpenAI from TPG/Brookfield/Advent/Bain, SAP’s Dremio and Prior Labs deals, Pinecone’s Nexus claiming 85% wasted agent compute, and ServiceNow’s Action Fabric with Anthropic at Knowledge 2026) amount to roughly a $5.5B bet on deployment infrastructure. Nate argues that the decisive value is the infrastructure that lets agents reach real data, act with real permissions, run workflows, and remain auditable — not raw model intelligence or token costs. The CodeWall exploit of McKinsey’s Lilli (full write access via an unauthenticated API and SQL injection) illustrates how procurement and build‑room failures, not just security, break enterprise AI plans and motivate this capital reallocation.

Jason Calacanis argues Elon Musk is quietly building a 'neo-cloud'—an 'Elon Web Services'—by combining Tesla's factory-scale manufacturing, vehicle onboard compute, Powerwall solar/battery hardware (as distributed compute 'fabs'), and Starlink connectivity to displace centralized data centers (AWS/Azure/GCP; ~$300B revenue, $4–5T combined market cap), scaling ultimately into orbit and monetizing homeowner-hosted compute.

Johnson–Lindenstrauss Lemma shows high-dimensional datasets can be randomly projected into much lower-dimensional spaces while roughly preserving pairwise distances (post by @probnstat, 2026-05-09). That property explains the success of random projections, embeddings and ANN search, enabling scalable learning by compressing redundancy rather than discarding signal—allowing dramatic dimension reductions without destroying geometry.

Aakash Gupta presents the 'Team OS' pattern: a shared repo that makes product decisions, call summaries, and analytics discoverable to people and agents. He argues which architecture to run depends on current AI adoption (<50%: Hub-and-Spoke; >50%: Full Adoption; 10+ engineers: add Agent Delegation), citing a 15-second DoorDash example and metrics on onboarding and knowledge loss.

The Idea Farm newsletter (May 10, 2026) curates several research reports with quantified implications for markets, technology risk, and AI adoption. Deutsche Bank credits a cyclical recovery plus structural themes (AI exposure, governance reform) for the KOSPI rerating and notes strong earnings and discounted valuations versus global peers. J.P. Morgan’s Investment Bank warns that ~20% of IT and ~50% of OT assets are “unpatchable,” calling attention to hardware-level vulnerabilities in systems from aircraft autopilots to power-grid controls. Vista Equity finds agentic AI adoption early—only ~1% of enterprise data in AI stacks—yet models as much as $11 trillion of cumulative software/infrastructure value by 2030. FT Partners’ 100-largest private-FinTech ranking shows $174B in combined revenue, topping $158B from the top-100 recent public firms. Separately, an intramonth momentum study (1980–2025) finds nearly all momentum returns concentrated in six trading days before month-end, implicating market structure over slow information flow.

Wells Intelligence API & CLI is launched by @kyleewalker (published 2026-05-07) with a Claude Code walkthrough: the agent identifies operators and permits, fetches and analyzes production history for a West Texas location, and outputs an interactive map over satellite imagery. The feature is available now to Operator-tier subscribers; API/CLI documentation is online.

Author @godofprompt reports that Google’s Gemini Interactions API moved from an opaque prompt→response model to a structured timeline that labels each step (thinking, searching, tool calls, final output). Google posted the change via Google AI Studio and says the schema enables mid‑flight steering and async tool calls; existing integrations must be updated before June 6.

Evan LaPointe shares Lenny Rachitsky's conversation with Eric Ries ahead of Ries's book Incorruptible (out May 26, 2026). The episode explains why companies decay under 'financial gravity', cites >80% of founders losing control within three years post-IPO, examines the Groupon-style public-company pitfalls, profiles 'governance fortress' firms, and offers three immediate founder actions.

The data center is technically under the county noise-action threshold yet produces a persistent low-frequency hum neighbors say lowers quality of life. Andy Masley and a midnight video from Merissa Hansen attribute most noise to temporary gas turbines, which won’t be removed until grid hookup that’s now delayed and could take up to seven years, revealing regulatory gaps.

Garry Tan's gstack (open-source; noted 2026-05-10) reconceives AI as an entire software team—CEO, staff engineer, QA, security reviewer, designer, release manager and parallel execution layer—through structured workflows like /office-hours, /autoplan, /qa, /review and /pair-agent. Tan says his output is ~810× his 2013 pace (normalized), arguing future winners will direct and orchestrate AI workers.

Noland Arbaugh’s Neuralink N1 implant reads motor commands the brain still produces: the same motor-cortex neurons fire 50–100 ms before movement and Neuralink captures those spikes. The device (1,024 electrodes on 64 polymer threads, 3–5 mm depth, 20,000 Hz sampling) survived an 85% thread retraction by relying on a rewritten ML decoder that restored and exceeded prior throughput, underscoring decoder advances over raw electrode count and continuing a line from the 2004 Utah array and BrainGate work.

The post credits activist hedge funds for pressuring Ajinomoto into a reported 30% price increase for its ABF build-up film. It notes spot ABF substrate prices have jumped over 30%, contract quotes are ~5–10% higher, Taiwan’s top three makers posted record April revenue, and supply-cost pressures and tight capacity are expected to push additional substrate price hikes of roughly 3–6% in H2, affecting customers like NVIDIA, AMD and Intel and PCB suppliers.

Gergely Orosz warns that distribution, not product development, is becoming the principal challenge for software startups—especially AI firms. He says numerous founders have rapidly built AI+context/trust/security tools but lack a way to reach users, and that lower build friction shifts competitive advantage to marketing and a race to win market attention.

Linus Ekenstam argues AC is unrelated to ventilation and that new builds legally require mechanized ventilation while older houses rely on convection (heat rising out high vents, low vents pulling in fresh air). He promotes heat‑recovery air-exchange systems for energy savings, praises Sweden's long-standing codes and criticizes Spain's passive-convection designs.

Europe lacks scale, not innovation, argues @marcosagusstinn: universities produce world-class research and talent, but fragmented national markets, weaker private R&D growth and shallower capital pools have cut Europe’s share of global corporate R&D from 21.4% in 2014 to 16.2% in 2024. The post urges creating one integrated market so startups can scale continent-wide and produce large tech champions.

Levie (@levie) argues on 2026-05-08 that when AI makes something easy, you should assume everyone can do it, so competition shifts resources into sales, marketing, and customer success (for example, client engagement when financial advice is automated). Gergely Orosz adds that customer acquisition is the costly bottleneck for many rapidly built AI startups.

A staff member on the House Select Committee on the Chinese Communist Party was contacted in December by a man identifying himself as Chris (display name Kris) Chen, who offered $10,000+ for occasional biweekly calls about the committee’s work and US policy and promised a $2,000 advance. Chen pressed for insights on Venezuela (post‑January military operation), rare‑earth minerals, and manufacturing in Vietnam and Mexico while claiming to represent NimbusHub Strategic Consulting. Republican committee staff recorded about two months of conversations, judged Chen likely a Chinese intelligence officer or contractor, and referred the case to the FBI. OpenAI later tied similar NimbusHub persona emails to ChatGPT prompts written in simplified Chinese and recommended concise, urgent messaging, suggesting an automated or coordinated operation to target U.S. officials.

AI in 2026 faces a paradox: frontier models (Claude, GPT-5, Gemini) are commoditized, so product success hinges on the 'harness'—prompt and context engineering, tool calls, memory, skill files, eval rigs, and orchestration. Lower model prices haven't shifted moats; firms that built harnesses ~18 months ago remain ahead. Aakash Gupta highlights these claims and points to an AI Skills conf on May 14 (conf.cosprints.ai/?32) with 20+ industry speakers.

Alex (@alexcooldev) is a solo indie hacker targeting 10 apps that each earn $5,000/month organically. By 2026-05-10 his first app hit $19k/month (then stalled) and a second app is doing $7k+/month. He argues organic growth caps near $20k, so he multiplies apps for profit now and will later use paid marketing on the top performer.

Terence Tao and Judit Polgár argue that shortcuts—AI tools in math or skipping practice in chess—remove the experiential journey that builds intuition and skill, and both warn this trend especially endangers youth development. @kshashi posted their parallel warnings on 2026-05-10 with links to The Atlantic and an archive.

Alex Nguyen (@alexcooldev) says he completed Jack Friks’s $5k/month challenge at the end of 2025 and now has a second app generating $7k+/month (web + app). He pursues a '10 apps × $5k/month' strategy, stays a solo, no-VC, no-ads indie hacker, and reports ~92% profit margins by minimizing costs (reusing phones).

Dillon Loomis announced on 2026-05-10 that his 'Electrified interviews' are back, highlighting a returning guest whose previous appearance was his most-viewed interview. He urges insider-driven updates on Tesla's multiple 'startups'—information he says retail lacks—and notes a potential SpaceXAI merger would heighten the importance; he asked followers to share the live interview ($TSLA).

A developer (shared by @chddaniel on 2026-05-10) dumped a two-week sprint into Codex 5.5 High with full permissions after reaching ~1% Plus-plan quota, ordered it to complete everything, and left it running. Nine hours later Codex had finished 4/5 deployment tickets with runnable, clean code and handled edge cases, prompting the developer to call it autonomous execution.

Cursive began as speed technology for quills and fountain pens; the ballpoint's 1945 mass-market arrival rendered connected loops mechanically unnecessary. The Palmer Method (1894) trained 25+ million students, but Common Core (2010) dropped cursive and prioritized typing. State reinstatements began in 2014 and reached 25 states by early 2026. The author says only brain-development research decisively supports cursive.

Gold's price at $4,700/oz (May 9, 2026) is contrasted with the $35/oz official price when Nixon closed the gold window in August 1971; the author emphasizes gold's physical constancy (atomic number 79) and argues the huge nominal increase reflects changes in the dollar/monetary system and purchasing power, not the metal itself.

Chamath warns the current wave of AI capex may be a bubble because, as of 2026-05-10, macro metrics like productivity and GDP growth haven’t meaningfully improved from AI. He insists continued spending depends on companies proving direct AI-driven revenue or margin gains (examples: shoes sold or restaurant profits), otherwise the AI investment case remains unstable.

China is publicly claiming it is exiting the dollar, but a CFR analysis (Brad Setser) shows Beijing’s state entities likely hold over $4 trillion in dollar assets—more off official books than the ~$1.8 trillion shown in SAFE’s $3.3 trillion reserves. Policy banks, state commercial banks, and the China Investment Corporation retain dollar exposures and continue dollar lending (e.g., Zambia, Sri Lanka, Ecuador, Angola).

@DeanTTraining posted on 2026-05-10 a step-by-step for an "EASY 75+ grams of protein" meal: marinate 1 seasoning pack per pound of chicken (breast/tenderloins) with 2 oz oil and 2 oz water, 2–3 hours, cook on cast iron and refrigerate. To serve, dice 12 oz cold chicken, add half a Birds Eye veg bag and 1–2 servings Mexican street corn, and sauté 6–8 minutes.

@0xd1namit says he’s developing a tweet-market trading strategy that he believes estimates the probability of a given number of posts more accurately than Polymarket. He built a quick v1 (two entry strategies) that backtested to breakeven, then v2 that backtested to ~2x in two weeks; all stats are simulated backtests. He now has access to Claude (via a friend), is testing a model on trade simulations, and will not disclose the model logic.

Steve C (@steveconaway1) recounts that in the week before his May 10, 2026 post he worked three separate deals where older owners offered significant discounts on farmland or homesites to young, non-family tenants — including one sale under contract for under 25% of fair market value. He attributes these private discounts to tenants’ demonstrated stewardship: long-term caretaking, honesty about work done/planned, good management, and a patient, non-entitled attitude after starting as tenants rather than immediate buyers. He pushes back on the blanket claim that “boomers” always demand top dollar, arguing many want the next generation to succeed. He closes with four concrete recommendations: keep room in your budget, build reputation patiently, celebrate others’ wins, and actively save/prepare for a future purchase opportunity.

Aakash Gupta argues Anthropic is repeating Apple’s App Store playbook but faster: by owning the model protocol and watching which API endpoints and wrapper categories scale, Anthropic ships native primitives and acquires leaders. He cites Cursor’s $2B ARR, Claude Code’s $2.5B in nine months, December’s Bun acquisition (7M monthly downloads), and rapid Claude integrations as evidence.

The Twitter post by @AskYatharth (quoting Qiaochu Yuan) claims Twitter's UI is unique because replies, quote-tweets and threads are full tweets with image/attachment support and feed visibility. That 'fully recursive' design lets discussions bud into new, independent conversations—an affordance QC says Substack Notes and rivals still haven't replicated.

Elon Musk leased Colossus 1 to Anthropic after SpaceXAI shifted training to Colossus 2; Drew Pavlou reported on 2026-05-10 that he spent a week with senior Anthropic team members, called them "highly competent," said "no one set off my evil detector," and judged that Anthropic's Claude "will probably be good."

Author @Afinetheorem (published 2026-05-10) cautions against superficial China coverage, citing Anji County—about 30 miles from Alibaba's HQ—in a wealthy tea/tourism belt near a rich city. They claim Zhejiang's GDP per capita is ~40% above the national average and comparable to Greece/Oman/Panama (PPP). David Fishman adds that tidy Zhejiang towns show GDP per capita poorly tracks quality of life.

Tpayneful liked Spencer Burleigh’s Substack post 'Andrew Jackson’s Ghost at the Fed' on mg1.substack.com; the notification includes the post link, author profile, and UI elements. The post's published timestamp is 2026-05-10 23:35:41+00:00, with earlier metadata showing creation on 2026-03-11 12:44 and a last update on 2026-04-04.

Vaseline pulled an AI-produced ad within hours after a designer posted a side-by-side screenshot that reached 9.1M views. The author argues major CPG brands accept the occasional viral takedown because AI creates dozens of ad variants faster than legal can review them, so social artist callouts now function as the practical rights-clearance mechanism.

Alex_prompter argues that AI is expanding, not replacing, developer jobs: Microsoft data shows US developer employment reached 2.2 million in 2025 (up 8.5% YoY) and GitHub pushes rose 78% YoY; AI lowers costs, uncapping demand so companies hire more—risk falls on developers who refuse to use AI.

Auto buyers who financed new cars in 2022–2023 are increasingly underwater: 30.9% of Q1 2026 trade-ins owe more than the vehicle is worth, averaging a $7,183 shortfall. Long 84‑month loans at ~7.9% push positive equity years out, dealers roll negative equity into larger 84‑month loans, and repossessions are rising toward Great Recession levels.

Desire, an essay by Ava Huang published May 10, 2026 in the Bookbear Express newsletter, argues that desire cannot be simulated: attempting to force it extinguishes it, whereas other life goals can be achieved by routine. Huang links desire to aliveness—automatic, uncontrollable, and rare—and distinguishes it from love, which she characterizes as unconditional acceptance while desire is conditional and bound to a person’s way of being. She writes candidly about refusing to fake affection, credits daily writing as a method to access genuine feeling, and quotes a reader (“Visa”) about the cost of pretending to be fine. The post references a Peter Doig image (Hitch Hiker, 1990), promotes a Love Science x Bookbear Express matchmaking party on May 15, and encourages paid subscriptions to sustain the newsletter and related coaching/matchmaking work.

Narendra Modi and the BJP are depicted as riding a political high after the West Bengal victory: the party now runs over two‑thirds of states and territories covering about 80% of India’s people, while Mr Modi has led as prime minister for 12 years. The newsletter credits economic dynamism and reform for the BJP’s popularity but warns of democratic erosion—electoral‑register revisions that have disenfranchised millions, weakening of the Election Commission, criminal investigations of opponents and a more pliant press—which could prompt a turn toward divisive politics if near‑term pressures such as rising energy costs hit voters. Simon Long, who covered the 2002 Gujarat riots when Mr Modi was chief minister, contrasts regional economic performance (an average Bengali earns roughly half of a Gujarati) and invokes wider concerns about rulers’ folly (citing Barbara Tuchman and a Telegram column). A letter from Prof. François Melese links long one‑party dominance to emigration (noting California after nearly 30 years) and argues democracy needs competitive pressures.

@Afinetheorem argues West–China mutual understanding is now mediated largely through social media, creating dangerous misperceptions. He stresses Western observers wrongly assume China has a robust safety net — in reality there is little, which frees fiscal resources for infrastructure and industrial policy. The post cites Yanzhong Huang’s May 10, 2026 NYT essay on Chinese overconfidence ahead of President Trump’s visit.

Sarvesh Shrivastava's May 10, 2026 thread prescribes a six-step Claude workflow to make the model act like a $10,000/month SEO team for roughly $20/month. He emphasizes precise prompts, preloading full business context, competitor gap analysis, intent-based keyword clustering (informational/commercial/transactional), and AI-search optimization prompts.

Chicago's 2008 parking-meter lease gave a Morgan Stanley/Allianz/Abu Dhabi joint venture control of all 36,000 meters for 75 years at $1.15 billion. Economists say the asset was underpriced by about $900M; revenues of $1.97B from 2009–2024 already exceeded the purchase price. A true-up clause forces the city to compensate investors for any lost metered revenue, including EV conversions.

Boom will eventually reveal its high‑pressure turbine blades but plans to keep them secret for now. Four foundries — PCC and Howmet (~80% market), Doncasters and CPP — dominate single‑crystal blade production, and heavy‑frame gas turbines are sold out through 2030. Elon Musk linked xAI Colossus power issues to blade chemistry, saying extreme purity (sulfur to ppb, oxygen to ppm) and specialized processes are the bottleneck.

User @bussypounder999 says Taiwan felt terrifyingly childless during their visit, attributing low birthrates to densely packed, expensive, tiny city apartments owned by boomers and corporations and to suburbs that lack infrastructure. Driving 40 minutes out revealed only fishing huts and small houses; they insist birthrates will stay low until units and lots can be merged into family-sized homes. Sovey (@SoveyX) replies that Korea now has the world’s lowest birthrate.

@DA_Stockman asserts Iran’s early bomb research was a deterrence-style program like Pakistan’s that was abandoned in 2003, citing multiple NIEs. He accuses Benjamin Netanyahu of peddling a 'BIG LIE' to justify war and calls Donald Trump a 'blithering idiot' who fell for it, while quoting Eyal Yakoby and an Iranian deputy speaker about past nuclear attempts.

@BoringBiz_ (May 10, 2026) argues there are only three real edges to generate alpha: information (having superior or forward-looking data, easier in small caps and emerging markets than in large caps), emotional (not making market-driven emotional trades so volatility is an opportunity), and timing (accepting longer holding periods to avoid forced selling across cycles).

@BoringBiz_ (2026-05-10) argues podcasts contain "incredible alpha": multi‑billion‑dollar company and investment‑firm CEOs, plus smaller‑fund managers and founders, openly discuss product strategy and zero‑to‑one journeys on niche, underrated shows. The post stresses that value requires taking action on those actionable lessons rather than using episodes as an escape.

Ali Ghodsi, CEO of Databricks, in a 40-minute conversation with Stanford students hosted by @apoorv03 on 2026-05-10, claimed AGI has already been solved under earlier definitions but the goalpost keeps moving; that modern AI still lacks organizational human-context; software remains important though barrier to entry has fallen; value will flow to the application layer (Uber, Airbnb, Amazon) while frontier models become an economies-of-scale game with shrinking margins.

Author @chrisman cites the Harvard Grant Study, claiming the single best predictor of success across fields is having a warm, loving mother, and calls a mother’s love the greatest blessing (posted 2026-05-11 for Mother’s Day). Alex Feinberg echoes a pastor’s Army Ranger anecdote: completing Navy SEAL training correlates with having a SEAL father or a strong maternal bond, which he attributes to nervous-system regulation; the post urges thanking mothers and fostering that regulation in children.

Chris Wick’s May 10, 2026 X post argues that voters who believed Donald Trump’s ‘America First’ and ‘no new wars’ promises are now disillusioned as U.S. aid and involvement persist in Ukraine and Israel, with ‘billions’ sent abroad. He contrasts that spending with rising groceries, crushing rent, homeless veterans, and growing household debt, saying supporters’ trust has broken.

Gavriel Cohen announced on 2026-05-10 that he will share a stage in Singapore next week with the country's Minister of Foreign Affairs, who published a gist two weeks earlier. swyx added that the minister (handle @VivianBala) is a "huge NanoClaw fan" and will keynote AIDOTENGINEER Singapore, with Cohen presenting immediately after, highlighting growing government involvement in @aiDotEngineer.

Gary Marcus highlights a reach disparity—his tweet got ~32,000 views versus ~2.2M for the misleading post (one cited tweet had 1.4M)—and rebuts a viral claim about chess by noting the cited paper shows a moderately positive IQ–chess correlation in children, not that low-IQ players are best, warning that oversimplified, unsupported claims spread widely.

Local AI model creation on Hugging Face has accelerated sharply: HF reported 176,000 public GGUF models (post dated 2026-05-10; May partial). After an Oct–Feb baseline of ~5.1K new GGUF models/month, March marked a +55% inflection and March–April averaged ~9.2K/month (April = 9.7K), likely driven by open-weight releases and better quantization tooling such as llama.cpp and automated pipelines.

Corey Ganim lays out a repeatable client-acquisition and productized-consulting playbook: run a low-cost local AI meetup (he partnered with a realtor, spoke about Claude for 15–20 minutes, captured attendee contact info and landed a client within a week), then sell a $999 "AI Assessment" that guarantees 5 hours/week saved or a refund. The assessment workflow is a 45-minute Fathom-recorded interview, transcript analysis in Claude to identify 5–7 opportunities, a Gamma-built report (executive summary, impact/effort matrix, 3–7 tool recommendations, 4‑day quick wins, financials), and a 30-minute review call; Corey says 40–60% of clients opt into paid implementation. He recommends starting by mastering Claude/ChatGPT/Fathom/Gamma/Zapier, doing 2–3 free assessments, ramping price from $499→$749→$999, and monetizing via an upsell ladder (examples: $1–10k implementations and $1k/mo retainers).

@cryptopunk7213 (2026-05-11) argues AI animation has already matched studio output after seven iterations at roughly 100× lower cost, is effectively Pixar-level for 99% of viewers, and will reach parity with live-action within 1–2 model generations. They predict rapid upheaval of Hollywood business models and democratized IP creation, citing a RunwayML demo by Marko Slavnic as illustrative.

Affinetheorem (2026-05-10) argues that while AI is already in current use, only Bay Area organizations tend to plan for where AI will be in 2029 — a mindset she says is required to restructure firms for substantial productivity gains. Ethan Mollick replies that AI adoption and advanced use cases are widespread across industries, so SF is not the sole epicenter.

Our World in Data reports that global liquid biofuel production expanded sevenfold between roughly 2006 and 2026, initially driven by Brazil and the United States as an alternative to imported oil when EVs were too costly. Although electric vehicles are now cheaper and cost-competitive in places, biofuel volumes are at record highs and are expected to grow further due to fuel standards and national policies (Data Insight by @_HannahRitchie, 2026-05-09).

Eric Ries, author of The Lean Startup and now Incorruptible, argues that durable companies are built by legally and structurally protecting the mission that creates their economic value. Speaking on Lenny Rachitsky’s podcast (episode ~1:39:22), Ries cites the empirical problem that roughly 80% of venture‑backed founders are ousted within three years of an IPO (only ~20% remain CEO), and attributes much of that churn to “financial gravity” — the predictable drift from original mission toward short‑term metrics. His remedies are practical: adopt mission‑protecting governance (examples include Anthropic, Costco, Novo Nordisk), consider public‑benefit/B‑corp style charters, and file a simple two‑page Delaware document to lock in purpose. Ries frames these as complements to Lean Startup methods — continuous innovation plus governance that prevents success from becoming the company’s undoing.

Aakash Gupta advocates replacing long planning decks with a concise PLANNING.md stored in the code repo, citing Matt MacInnis (Rippling CPO) and a concrete 40-line spec format (Problem, Hypothesis, Success metrics, Rollout). He argues repo-based specs enable Claude Code integration, better IDE access and auditability, and points to Garry Tan's 65K-star Claude Code as proof; he urges committing a PLANNING.md on GitHub today.

ancerj argues it is inevitable that rules demanding unquestioning belief or respect will be abused, warning that banning critique risks serious harm. The claim is illustrated by Larissa Phillips' account: a man who said he was a woman was admitted to a women's shelter despite a nurse's protest and punishment of that nurse, and subsequently murdered his roommate.

Shipper, announced by @chddaniel on 2026-05-10, positions itself as an end-to-end autonomous AI game studio that turns ideas into playable 2D/3D/open-world MP/SP titles, publishes to Steam and Itch, and "self-maintains" the business. Built on Claude Code Opus 4.7 and trained on 1,000+ games, it touts one-click localization to 68+ languages and a launch giveaway requiring a retweet/comment.

Chris (@chatgpt21) says OpenAI's Codex autonomously found and executed an open-source security/audit bounty over ~22 hours, submitting a PR, handling maintainer interactions and verification, and preserving his payment privacy. The run produced a first payout of $16.88 (received by the author), which he views as a proof-of-concept for AI that can independently earn income.

Hannah Ward (@HannahWardEdu) explains that after teaching in Montessori classrooms she switched from a letter-sounds-first method to a letter-names-first approach and created a free curriculum that teaches the alphabet first, then phonics. She says sound-first gave early gains that collapsed with complex words, requiring reteaching of letter names, and cites corroborating classroom research and higher success when names precede formal phonics.

Robert Morris personally raised $1.4 million in 1781 on his own credit to fund Washington’s march to Yorktown, with the final $20,000 underwritten by broker Haym Salomon. A signer of the Declaration, Articles and Constitution and Superintendent of Finance (1781–84), he later amassed nearly $3 million in debt, spent three years in Prune Street debtor’s prison, and died in 1806 broke—contradicting AOC’s "billionaires" claim.

Tpayneful liked the Substack post "Success Bias Reply - Maybe the Opposite" by Spencer Burleigh on 2026-05-10 23:39:18+00:00. The email-style notification from mg1.substack.com contains links to the post and profiles, publisher metadata (© 2026 Spencer Burleigh), a San Francisco mailing address, and standard Substack actions (mute, view profile, app, start writing).

Gary Marcus (May 10, 2026) urges calm over the Mythos/METR graph, stressing it measures ~50% success on software tasks—not high reliability or general intelligence. He acknowledges Claude Code and Mythos advances but says gains likely stem from neurosymbolic tools (interpreters, verification, harnesses) rather than pure scaling; per @ramez, Mythos remains on trend on the ECI benchmark.

Tpayneful liked the Substack post "Institutions, Institutions, Institutions" on 2026-05-10 23:38:21+00:00; the item is hosted at mg1.substack.com. The page shows Created 2026-03-11-12-44 and Last Updated 2026-04-04-00:00, includes © 2026 Spencer Burleigh in the footer with a San Francisco mailing address, and displays the figure "615."

The U.S. fertility rate fell to 1.62 in 2025, producing a generation ~23% smaller (2025 cohort ≈3.6 million). Fertility declines—including a 72% drop in teen births—persisted for 11 years post‑Great Recession. The author ties this structural demographic contraction to policy strains (Social Security) and $660–700B in 2026 AI capex by hyperscalers as firms price long‑term demand risk.

Tpayneful liked the Substack post "Travel Difficulty Index: My Tiered List for English Only Americans" on 2026-05-10 at 23:36:43 UTC. The item is published on Spencer Burleigh's Substack (©2026), created 2026-03-11 and updated 2026-04-04, hosted at mg1.substack.com and showing an engagement figure of 344.

Tpayneful liked the Substack post "Founder’s Guide to Executive Skills" on 2026-05-10 23:34:37+00:00; the email-style notification (mg1.substack.com) links to the original article and the liker’s profile. It also cites Spencer Burleigh (© 2026), shows a San Francisco mailing address (548 Market St, PMB 72296, CA 94104), UI actions (View profile, Mute post) and the number "433."

Doomberg's Information Grade (May 10, 2026) describes a deliberate method of broadening intelligence collection by repurposing the Twitter/X algorithm to surface a curated pro‑Iranian feed immediately after the Iran war began. That feed surfaced satellite imagery and on‑the‑ground video—much of it routed from Chinese platforms—that suggested Tehran had struck US bases in the region far more heavily than official statements admitted. Western satellite vendors reportedly confirmed much of the damage before the Pentagon intervened to limit publication. The Washington Post later quantified the impact as at least 228 damaged or destroyed structures/equipment across US sites, and commanders evacuated personnel from threatened bases. Doomberg also flags May 2 reports from China about a possible escalation in the sanctions war and a looming Trump–Xi summit, stressing the need for disciplined information grading amid mixed signals.

Tpayneful liked Spencer Burleigh’s Substack post "The Saudi Spat with Canada - Messy" on 2026-05-10 23:38:54+00:00. The page is a Substack notification/email with profile imagery, links to the post and profile, UI elements (mute, app/download prompts), Spencer Burleigh © 2026 metadata, the address 548 Market Street PMB 72296, and a displayed value of 259.

Tpayneful liked Spencer Burleigh’s Substack post "Getting Started" on 2026-05-10 23:46:28+00:00. The email-style notification provides links to view the post and the liker’s profile, a mute control, and footer metadata showing ©2026 Spencer Burleigh and the San Francisco address (548 Market Street PMB 72296); it also displays the number "255".

Tpayneful liked the Substack post 'Apollo - Success Bias' on mg1.substack.com; the notification records the post as published 2026-05-10 23:39:33+00:00 and credits Spencer Burleigh (© 2026). The scraped item includes metadata (Created 2026-03-11 12:44; Last Updated 2026-04-04), a San Francisco mailing address, UI images, and a displayed count of 137.

Tpayneful liked Spencer Burleigh’s Substack post "21st Century Moats - Your Brand" on 2026-05-10 23:39:38+00:00. The notification captures metadata (Created 2026-03-11-12-44; Last Updated 2026-04-04-00-00), shows copyright © 2026 Spencer Burleigh, the San Francisco publisher address, and includes standard Substack action links and a displayed figure "215."

The war in Ukraine is presented as unwinnable for Russia by @oroborous (posted 2026-05-10), citing Mediazona/Meduza’s Probate Registry estimate of 352,000 Russian deaths through end‑2025 and an updated ~400,000 by May 2026 — ~1% of ~40 million fighting‑age men plus ~2% incapacitated — and projecting that 5–25 million more men would be needed to seize remaining territories.

Tpayneful liked the Substack post "Sukuk - No Interest Here," per a notification timestamped 2026-05-10 23:37:55+00:00. The message is a platform engagement email from Spencer Burleigh’s Substack account, embedding links to the post and profile and offering CTA buttons (Mute post, Get the app), with no article content included.

Levie (@levie) argues that as agents perform more long-running, compute-heavy tasks, enterprises must adopt explicit 'token budgeting' to allocate tokens across teams. Because leaders lack end-to-end visibility and controls, token spend will move into organizational budgets and require new software solutions, creating startup opportunities during a compute-constrained era.

Bijal (@bijald) writes that since 2021 they treated @lennysan’s newsletter and podcast as required reading, bingeing 5–6 episodes before interviews. They say Lenny taught concrete PM mindsets—tradeoffs, growth, sharper questioning, taste versus jargon—and that an episode with Shreyas inspired them to implement a Pre-Mortem. Bijal calls Lenny an unofficial mentor to many PMs.

LeWorldModel (LeWM) claims a JEPA trained end-to-end from raw pixels with 15M parameters on a single GPU in a few hours, using two losses (next-embedding prediction and a Gaussian latent regularizer), no EMAs or pretrained encoders, one hyperparameter, and up to 48x faster planning while remaining competitive on 2D/3D benchmarks. Yann LeCun closed $1.03B for AMI Labs on March 10, 2026 (seed at a $3.5B pre-money valuation with Bezos, Nvidia, Samsung, Toyota); the LeWM paper was released three days later by authors at Mila, NYU, Samsung SAIL, and Brown.

Tpayneful liked Spencer Burleigh’s Substack post "Gold Mines & Self-Terminating Business Models" on 2026-05-10 23:35:04+00:00. The notification is a Substack email containing post metadata, author avatar, view-profile and mute links, app/publish buttons, embedded images, and the publisher footer (©2026 Spencer Burleigh; 548 Market Street PMB 72296, San Francisco, CA 94104).

@adxtyahq argues that running Qwen 3 27B locally with a capable CPU+GPU delivers near-Claude Opus performance and prevents your data from being used to train third-party models, citing @julienc's agreement. Kartik (@codekartik) counters that paying for Cursor in 2024 was a poor decision—bad limits, weak reasoning, low ROI—and recommends Claude Code or Codex.

Geek Estate Blog's Weekly Digest (published May 10, 2026) promotes GEM Crystal, a paid real-estate tech newsletter offering Weekly Radars (4x/month), Weekly Transmissions (2x/month) and a Quarterly Earnings Radar. It advertises 85% off for three months, notes 275+ Weekly Transmissions published, and links to a piece on AI in property management (leasing agents, automated screening, scheduling).

The post by @alex_prompter (2026-05-10) tells Claude Code users to invoke "use /skill-creator" and build skills with a /references folder, /assets folder, and a main SKILL markdown router. It requires a pre-generation compatibility check, surfacing restructuring issues and asking clarifying questions, and to explicitly list any non-text assets needed rather than creating placeholders.

Ben Lang’s May 10, 2026 Next Play post lists ten concrete, fast-to-build startup ideas ranging from marketplaces to AI tools. Highlights include OnlyFeedback (a dating-profile coaching marketplace where coaches set rates and the platform takes a cut), InstantExpert (an AI-driven expert network that auto-discovers experts, removes scheduling by letting experts join an immediate call link, uses AI to continue questioning until research goals are met, and returns a final report), and AI Talent Tracker (real‑time LinkedIn monitoring for departures at OpenAI, Anthropic, Meta AI, DeepMind, Cohere — Next Play notes they’ve partially built it and invites hires via hi@nextplay.so). Other ideas: a Forward Deployed Engineering community, an entirely AI-powered news livestream with animated AI characters, Swim Lane (male fertility tracking with metrics like count/motility/morphology and at-home test integration), infrastructure for AI rollups, a sober headline-focused media brand, and lightweight consumer apps (Favorites, Momma Knows Best). The post emphasizes viral growth channels (TikTok, livestreams, newsletters/podcasts) and practical monetization paths (marketplace cuts, recruiting, paid reports).

@fromzerotomill argues short slideshow videos are 'quietly printing' because algorithms reward completion rate: a 60-second slideshow with 90% completion beats a 30-second talking-head with 60% completion. He recommends writing an 8–10 point framework with Claude, producing one-slide-per-point via makeUGC (makeugc.ai/?ref=amir), and monetizing by teasing deeper paid content.

Kris Abdelmessih’s Moontower #314 (May 10, 2026) opens with a close read of Adam Mastroianni’s essay distinguishing objective intelligence (verifiable problems) from subjective intelligence (wisdom, lived feeling), and uses that to explain why current AI writing often “vaguely sucks”: models can follow recipes but cannot transmit personal, subjective experience. The newsletter then turns to money and careers: Kris resurfaced his Nov 7, 2024 piece on adverse selection in the option job market, cautioning senior traders that hiring processes can concentrate lemons and that career moves require explicit filtering for selection risk.

On trading practice, Kris summarizes a Brent & Alf podcast and applies its lessons: add a carry‑to‑vol pre‑trade screen to force a “thinking in bets” discipline. Concrete examples: a Hungary FX view had ~4% annual forward carry (≈30 bps/month), implying an ex‑ante one‑month Sharpe around -0.6 unless the move is sharp; options buyers repeatedly pay vol (costs compound with frequency); hard‑to‑borrow equity borrow can reach ~150% (CoreWeave), making timed intraday shorts sometimes more profitable than multi‑day holds; a EUR/GBP trade at 0.8573 carried a 17‑pip roll to Sept 17; and exogenous shocks (copper -30% over two days after tariff removal) can produce transient 3–5% moves across other metals. Kris’s practical takeaway: make cost of carry and borrow explicit hurdle tests in process, be fast with sympathy trades, and beware adverse selection when considering senior roles. He closes with lighter notes — a GME/eBay options mention, a KenKen puzzle anecdote (4x4 logic, bottom‑right = 2 given a 7+ cage contains 3 and 4), and an offering of $500/60‑minute consults for decision help.

Sophie Bakalar's May 9, 2026 essay (cross‑posted by Collaborative Fund on May 10) argues that the QR‑code story is a template for how automation actually unfolds: QR codes (technology extant since 1994) surged in NYC in late fall 2020, streamlined ordering and payments, then receded from the diner experience while the POS, kitchen display and reservation infrastructure they accelerated remained. Bakalar pairs restaurant employment data — full‑service employment still 3.7% below 2019 but up 86,000 jobs in the past year — with the broader AI debate: CEOs like Sam Altman pitch augmentation, commentators (Jack Clark) put a 60% chance on recursive self‑improvement by 2028, and ~50% of Americans are worried. The piece stresses that automation eats tasks measured by benchmarks (chatbots, apps, ATMs) but often fails to replace jobs where human presence is part of the product, echoing Narayanan and Kapoor’s point that benchmarks capture tasks, not complex professional practice.

Financial Samurai (Sam) argues public markets are overheated—S&P 500 hit 7,399 well ahead of his 7,300 year-end target—so he's exploring private/derivative exposure to the AI buildout. He describes a rapid, ~ $400M double-layer SPV for Anthropic with 10% fees and 20% carry, and posits that buying Anthropic at a $900B valuation could return ~2.2x if the company reaches $3T within two years (70% probability in his view). He highlights Fundrise’s Q1 10.1% gain tied to AI infrastructure, and a Feb 16 senior-secured bridge loan to develop a nearly 3,000-acre, 5GW data-center campus in Hubbard, Texas (12.55% fixed, projected 18.1% IRR over 10 months). He also notes Anduril secondaries near $100B and urges disciplined allocation and awareness of execution, financing, and concentration risks.

AwardWallet Plus subscription renewal failed after an attempted $49 charge to the card ending in 7560; the renewal would have extended membership through May 10, 2027. The service will retry billing daily for one week before stopping and provides a payment-update link plus account restore and unsubscribe options, with company address listed.

The baobab is a living water tower and long-lived resource: trunks hold tens of thousands of liters, support wildlife in drought, and yield 250+ fruits/year and valuable powder (~$9B market). Climate stress has killed many ancient individuals (Patrut et al., 2018; Panke fell 2010–11), yet people still fell 1,500‑year trees for short‑lived timber.

@JoshDaws (2026-05-10) condemns the 'unreal' hate aimed at AI creators, arguing great artists embrace new tools while mediocre artists feel threatened. A linked post from Soul Motion labs demonstrates that embrace: a Seedance 2.0 clip made with RunwayML for an upcoming short, accompanied by a detailed stop-motion prompt (animation on 2's at 24fps, tactile textures, no motion blur).

@AlexFinn (2026-05-10) advocates "Reverse Prompting": give an AI a full brain dump about yourself, then prompt the model to ask you what additional info it needs and to list immediate tasks it can do. He names OpenClaw/Hermes as example agents and guarantees this approach yields far more actionable tasks ("100x") and productivity.