Your briefing

Naval (Naval Podcast) lays out a model of persuasion centered on credibility, honesty, and intrinsic motivation rather than conventional salescraft. He references Robert Cialdini’s CLASSR framework and anchoring as a useful checklist but insists that effective “selling” is really about being authentic, explaining simply, and building long‑term trust so people don’t feel they’re being sold to. He practices “rational empathy” and often employs a “yes, and” stance to understand others’ positions before agreeing or contradicting them.

Naval frames charisma as the simultaneous projection of confidence and love, and privileges honesty over politeness while still arguing kindness is necessary for effectiveness. He admits to being wrong about 80% of the time, to avoiding brute-force sales tactics, and to struggling with firing people—preferring to reassign or help them find better roles. He emphasizes leadership as inspiring people to want the work, preferring small, high‑trust teams (10–20 up to a few hundred for major projects) and uses the stag‑hunt metaphor to explain why cooperation scales value. Practically, he advises only selling things you’re obsessed about, timing fundraising to internal excitement (citing his startup Impossible), and feeding good intellectual obsessions rather than chasing formulas or motivational noise.

Jeff Bezos generated so many ideas that operations leader Jeff Wilke warned they would ‘destroy the company,’ advising Bezos to limit releases to what the organization can absorb. The post ties this to Little’s Law — excess inputs raise WIP, cycle time, and rework — and notes Bezos changed by prioritizing and keeping lists, while many startups still flood teams with new initiatives.

The post argues that autonomous coding agents create a 'ratchet' effect: repeated agent loops ship tests, docs, and evals with code, producing compounding, reliable progress. Examples: Garry Tan merged 14 PRs in 72 hours adding 29,000 lines; steipete ran 50 Codex instances to close 4,000 issues and released 'clawsweeper.' The author advises adding 'and write tests too' to Codex prompts.

Siri was unveiled October 4, 2011 as a flagship iPhone 4S feature—two days before Steve Jobs’s death—and, according to the author, has become a 2026 punchline many users disable. The new video argues Apple deliberately underinvested in AI while rivals (Microsoft, Google, Meta, OpenAI) pour $30–$50 billion annually into AI, explaining Siri’s decline.

WorldLens introduces a comprehensive benchmark for driving world models, spanning five aspects and 24 dimensions to evaluate pixels, 4D geometry, closed-loop driving, and human perceptual realism. The authors benchmark six models, show no method excels across axes (best attain 2–3/10 human realism), and provide WorldLens-26K plus WorldLens-Agent for scalable, explainable evaluation. (Summary based on abstract; full text not reviewed.)

3DMaxConvNet extends the MaxConvNet magnetic potential-field planner into 3D by training a convolutional autoencoder to generate obstacle-aware potential fields from LiDAR-derived 101^3 voxel grids. In Cosys-AirSim experiments (100 randomized closed-loop trials on two urban maps) it reached 100% success without retraining, matched A* path quality with ~2× lower runtime, and ran ~200× faster than RRT*(3k).

Heather C. Miller amplifies Gary Marcus's claim that Claude Code, while "still not AGI," is a major neurosymbolic advance: it pairs a state‑of‑the‑art LLM with 53 symbolic tools and roughly 500,000 lines of symbolic code. She argues this vindicates integrating classical AI/CS into compound systems rather than relying on pure LLMs, and links to a ccunpacked.dev dissection.

An AI PM interview question about a hiring tool recommending 15% fewer candidates argues candidates should prioritize immediate harm mitigation over technical root-cause debates. The right product move is to pause auto-rejects, route affected recommendations to human review, and fix the model offline (as Prasad Reddy did at Deaher). Audits take ~10 days; legal exposure can cost years and hundreds of millions, so leaders must lead with response and timeline.

Gary Marcus (2026-05-12) insists his position is that LLMs sometimes “partially regurgitate” and that fluent repetition doesn’t equal understanding, endorsing SE Gyges’ wording. He claims Geoffrey Hinton has attributed a different, unfounded quote to him on Hinton’s website, calls that fabrication or out-of-context citation, and urges intellectual honesty from a Nobel laureate.

ByteDance enacted an April 2026 rollback that eliminated 30% of AI projects (e.g., 猫箱, 星绘, parts of Dreamina) after failing to produce any of three promised 10M+ DAU products in 2025. With AI inference costs >¥80B (2.3× incremental revenue), slowing overseas growth and regulatory headwinds, the company is refocusing on 豆包, hardware (PICO and AI glasses), and cutting broad app bets.

Ethan Mollick (2026-05-12) reports OpenAI told him Study Mode remains live and reachable via /study and /learn shortcuts, despite its absence from most ChatGPT menus. He argues the silent removal is “a big mistake,” citing evidence that assistant-mode AIs can hurt learning by giving answers, and says Study Mode offered a simple mitigant for parents and teachers.

Aakash Gupta (tweeted 2026-05-12) argues that bloated PM artifacts—exemplified by a 30-page PRD that never ships—result from missing upfront checks in generator skills. His solution: simple three-line gates at the top of Claude skills, paired with a table that anticipates rationalizations. He validated this in 75 tests over 25 skills and publishes 10 laws plus two actionable skills (/improve-skill, /create-skill).

Consumer investors are recalibrating: Eric Newcomer (Substack) reports on May 12, 2026 that a Menlo Ventures–hosted dinner with VCs like Faraz Fatemi, Kirsten Green, Saar Gur, Ivan Zhou, and Mike Duboe reflected a shift—the consumer category is evolving to encompass prosumer AI, including coding assistants and other productivity-focused AI tools—rather than being dead; full article is paywalled.

Om Malik reflects on Pete Larson’s decision to stop making videos for his Just a Few Acres Farm YouTube channel, a project Larson ran for roughly six years after leaving a 20‑year architecture career to become a small cattle farmer. Malik notes Pete and his wife — both in their 50s — intend to “find a new way of doing things” and accepted not knowing the next step. Despite building a channel with hundreds of thousands of followers and a revenue stream, Larson chose to walk away; in his final clip he said two words, “I’m sorry,” recognizing how much viewers relied on that steady rhythm. Malik frames the exit as a quiet, characterful counterpoint to the spectacle of technology (citing the OpenAI lawsuit) and expresses personal envy and a longing for a less rancorous life.

LLM-based digital personas were evaluated as substitutes for human survey respondents using the LISS panel: personas were built from background variables and pre-2023 survey histories and tested on held-out post-cutoff answers. Across four persona architectures, three LLMs, and two prediction tasks, personas improved distributional alignment (notably for stable attributes) but struggled with individual prediction and multivariate respondent structure; retrieval augmentation helped. Summary is based on the abstract (full paper not reviewed).

Hiten Shah (Product Habits) ran an A/B email test on 2026-05-12 — half of subscribers saw one version and half another, with some calling one variant AI-sounding. Facing a year-long drop in click rates, Shah hypothesizes readers detect “empty” AI-style prose and will run a live Reddit AMA to probe responses and defend human writing.

Pirate Wires (Ryan Hassan, May 12, 2026) examines how a March 31 Google paper re‑shapes the quantum threat to Bitcoin by presenting a theoretical quantum‑computer design that could extract a private key from an exposed public key in ~9 minutes—nearly matching Bitcoin's ~10‑minute average block interval. The paper's resource estimate (~500,000 qubits) is roughly 1/20th of earlier forecasts that assumed millions of qubits, making a Shor‑algorithm attack (Shor, 1994) significantly more feasible. The article explains the quantum advantage (superposition and parallelism), contrasts the >1 billion‑year effort required by classical means, and warns that wallets with exposed public keys—including early Satoshi‑era addresses—could be immediately vulnerable once such hardware exists. It compiles views from ten crypto/quantum experts on timelines, mitigation paths (post‑quantum upgrades, key hygiene), and the risk of market and consensus disruption around 'Q‑Day.'

OpenSquilla: @hasantoxr ran a 10+ turn research workflow (published 2026-05-12) and says smart routing routed simple prompts to cheaper models and complex ones to full power, cutting his token bill by ~67% versus his usual agent while preserving memory and output quality. OpenSquilla (open-source, Python, local‑first) claims 60–80% cost reduction and launched the #10MTokenChallenge (30 winners × 10M OpenRouter credits).

PriorVLA introduces a prior-preserving adaptation method for Vision-Language-Action models that freezes a Prior Expert and trains an Adaptation Expert, using Expert Queries to inject pretrained scene and motor priors. By updating only 25% of parameters, it outperforms full fine-tuning and SOTA baselines on RoboTwin 2.0, LIBERO, and eight real-world tasks, with strong OOD and few-shot gains.

The paper addresses anonymous MAPF by formulating it as a Markovian MMOT that exactly reduces an exponentially large problem to a polynomial-size LP, with proven feasibility and total unimodularity yielding collision-free integral {0,1} transports. For large-scale instances the authors use a Schrödinger-bridge entropic regularization solved via Sinkhorn iterations to produce fractional templates used in a reduced LP, trading negligible optimality loss for major computational savings; experimental results demonstrate strong optimality and scalability. Accepted to ICML 2026 as a spotlight paper.

RoboMemArena is a large-scale benchmark addressing robotic memory shortcomings by offering 26 long-horizon tasks (avg >1,000 steps) with multimodal memory annotations and paired real-world evaluations; a VLM-based pipeline composes subtasks and generates trajectories via atomic functions. The authors introduce PrediMem, a dual-system VLA with recent/keyframe memory buffers and a predictive coding head that outperforms baselines on this benchmark.

UniSteer tackles efficient real‑world adaptation of diffusion-based vision‑language‑action models by combining human corrective interventions with noise‑space RL. The method inverts a frozen flow‑matching decoder to convert human actions into supervision for a noise‑predicting actor, which is concurrently improved with RL. Real‑world tests on four manipulation tasks show a jump from 20% to 90% success in 66 minutes on average. Full paper was not available, summary based on the abstract.

Counterfactual stress testing for medical image classifiers uses causal generative models to create realistic, anatomically-preserving ‘‘what-if’’ images by intervening on attributes like scanner type and patient sex. Evaluated on chest X‑ray and mammography across three architectures and several shift scenarios (Stammel et al., 2026), the method more reliably predicts real OOD performance than simple brightness/contrast perturbations, improving robustness assessment prior to deployment.

Rosamund Pearce’s May 12, 2026 Off the Charts newsletter links two strands: hard news visualised and guidance on technical diagramming. On the news side, the piece notes the Strait of Hormuz has been effectively closed for over two months, producing a major energy‑supply shock that has raised US petrol prices and disrupted trade in items from plastics to pistachios; it also flags SIPRI data (reported by Sondre Solstad) that, on a purchasing‑power basis, America’s allies now outspend the US on defence for the first time since 2001, and reminds readers Britain’s Rent Act began on May 1, 2026 with possible unintended effects. On the craft side, Pearce explains projection choices — oblique, isometric (equal axes), dimetric/trimetric and perspective — gives concrete examples (Chernobyl reactor, Hong Kong flats, geothermal schematic), and advises when to use 2D versus 3D tools (Blender, three.js) and why projecting text often reduces readability.

Artificial Analysis' τ‑Voice benchmark extends τ²‑bench into speech-to-speech agent evaluation using an LLM-driven simulated customer and realistic audio (accents, noise, packet loss) across Airline (50), Retail (114) and Telecom (114) tasks. xAI's Grok Voice Think Fast 1.0 tops the ranking at 52.1% (5.6 min); OpenAI and Gemini models follow in the high‑30% range, and overall S2S still resolves roughly half of realistic customer-service scenarios.

Megan Gafford’s May 12, 2026 essay 'Towards a New Aesthetic' announces that she won a New Aesthetics grant—prompted by Patrick Collison and Tyler Cowen’s end‑of‑2025 Call—to pursue a project to invent a new ornamental architectural style through drawing. Her approach is practical and methodological: she will travel to cities notable for Art Deco and Brutalist buildings, make life sketches (not photo studies), and pair drawing studies with archival and observational research published on Fashionably Late Takes. Gafford frames drawing as a historically decisive tool—arguing Le Corbusier’s and Paul Rudolph’s drawings shaped modernism despite uneven built results—and intends to use prolonged, on‑site observation (she invokes a six‑month moon‑drawing project as precedent) to synthesize past motifs into novel ornament. She also critiques public discourse (Noah Smith, Matthew Yglesias) for limiting the design conversation to modernism versus European pastiche, urging exploration of American aesthetic resources instead.

Sam Altman's cross‑examination transcript excerpt (posted 2026-05-12) shows Musk's lawyer repeatedly pressing him about honesty and trustworthiness—Altman hedged several times, eventually amending an answer to 'yes.' Gary Marcus reshared the exchange with the one‑line caption 'speaks for itself,' framing the testimony as damning or self‑evident.

@emollick reports that ChatGPT’s Study Mode was silently removed for most accounts (reported 2026-05-12), while Claude and Gemini still offer similar options. He argues Study Mode provided a simple, recommendable setting for parents and teachers to reduce learning harms from assistant-style answers and notes OpenAI’s Study Mode page/link remains online.

Aakash Gupta argues $32,604 (the 2026 equivalent of Sears' 1910 $938 kit) no longer buys a Sears kit home: California's average impact fee is $37,471 and Fremont's hit $157,000, while Los Angeles and San Francisco charge $18k–$40k in permit fees. Modern lot-size rules, fire and contractor codes, grading standards and 12–24 month permitting make owner-assembled kit homes effectively infeasible today.

Fraser McIlwraith reports that Armenia is visibly pivoting toward the West under Prime Minister Nikol Pashinyan but remains politically and socially divided. High-profile events on May 4–5, 2026 — an EPC meeting, Armenia’s first EU bilateral summit and the Yerevan Dialogue where Emmanuel Macron praised Pashinyan — acted as an unofficial pre-election boost ahead of the June 7 vote, with Civil Contract leading Strong Armenia by over 15 percentage points. Yet pro-Russian currents persist: oligarch Gagik Tsarukyan is erecting a giant Jesus statue in Zovuni and leads a minor, Russia-friendly party (3% in a February poll), while oligarch Samvel Karapetyan, under house arrest, urges closer ties with Moscow. After crushing defeats since 2020 Armenia is negotiating peace with Azerbaijan, but deep economic dependence on Russia and decayed institutions mean Pashinyan must pursue gradual trade diversification and reforms to consolidate a pro-European trajectory.

Hiten Shah's newsletter "AI broke the defaults" (May 11, 2026) synthesizes recent research and practitioner notes showing how cheap AI creation is changing defaults across software, content, and organizations. Key findings include an Internet Archive sample estimating ~35% of new sites by mid‑2025 were AI‑generated/assisted (with semantic contraction and a positivity bias), Chicago Booth tests of four detectors on ~2,000 passages that fail for snippets under ~50 words, and a study finding ~380,000 vibe‑coded apps publicly reachable with ~5,000 leaking sensitive data via simple URLs. Shah frames agents as tool‑loop executors, warns Copilot‑style usage creates private learning trapped in loops, and cites Microsoft and survey data (13% rewarded for reinvention) plus Clay’s role‑plasticity as structural advantages in the AI era.

Zvi Mowshowitz argues that contemporary math education is failing because weak research, grade inflation, and policy choices have removed objective reality checks. He amplifies Kelsey Piper’s exposé of Jo Boaler’s influential work, listing concrete methodological failures: undisclosed sites, inappropriate comparison groups (top quartiles vs. middle quartiles), assessments two to three years below grade level, incorrect grading, lack of predictive validity with standardized tests, and likely pre/post practice effects on short interventions. Those research flaws, he contends, fed policy choices that removed rigorous admissions signals (e.g., UC removal of SAT/ACT in 2020) and allowed students to accumulate transcripts that don’t reflect actual skill.

Mowshowitz cites UC San Diego data showing remedial Math 2 enrollments climbed from 32 (fall 2020) to ~1,000 (fall 2025), about 12% of students; many remedial students reported completing precalc/calculus in high school despite scoring below middle‑school levels (42%), and over a quarter had a 4.0 math GPA (avg. 3.7). Between 2017–2023, such students had 24% D/F/W in Calculus 10A and 30% D/F/W in 10B. He links these patterns to wider phenomena—grade‑scale inflation, dubious classroom grading practices, and experimental curricula (e.g., novel long‑division methods)—and urges restoring objective testing, stronger accountability for grade/reporting accuracy, and an end to “cargo‑cult equity” that masks skill deficits.

Author @0xd1namit adopts cryptovcdegen's six-step Polymarket weather-bot plan, emphasizing staged validation over jumping to live trading. The workflow mandates reading the Rules on 10 markets, building a data layer (ensemble forecasts + CLOB prices), 30 days of historical Brier-score checks, a bias-corrected signal layer, ≥2 weeks of paper trading, then live risk on 2–3 validated cities.

Nick Huber argues that the core failure of many small-business owners is time allocation: they constantly solve urgent problems and never do the uncomfortable, important work that scales a company. He illustrates the point with a concrete anecdote about a friend who runs a 1,500 sq ft pizza restaurant making roughly $250,000 annually yet is still required on-site every Thursday–Saturday to handle staffing, equipment, and customer issues. Huber contends this reflects ownership of a job rather than an organization because the owner spends most of his time in urgent quadrants instead of the important-but-not-urgent quadrant that produces long-term growth.

Huber lays out a four-quadrant time-management matrix (urgent/important; urgent/not-important; not-urgent/not-important; important/not-urgent) and prescribes delegation, hard conversations on hiring and sales, and system-building as remedies. He also shares operational claims: he’s hired 400+ offshore staff across his portfolio at roughly 80% lower cost than U.S. equivalents and advertises a $500 hiring discount. Practically, he promotes a May 19, 2026 workshop on profitability and cites business metrics (386 self-storage rentals in the first eight days of May 2026 vs. 206 last year) to underscore applying these principles.

Gary Marcus (posting 2026-05-13) pushes back on absolutist takes about AI and jobs: he prefers saying there won’t be an immediate AI-driven jobpocalypse, but rejects the certainty that one will never occur and calls the converse claim — that AI will trigger a broad “jobapalooza” — even less believable. Marcus reproduces Andrew Ng’s longer thread, in which Ng argues “There will be no AI jobpocalypse,” noting a 4.3% U.S. unemployment rate and continued strong software-engineer hiring as signs that AI so far tends to create more jobs than it destroys. Ng also lays out economic incentives that drive hype: frontier labs benefit from dramatic narratives, AI pricing can be anchored to $100K salaries (making $10K/user pricing plausible), and firms may cite AI as a convenient explanation for layoffs. Ng concludes optimistically that AI will produce many new AI-engineering jobs and changing skill demands, while Marcus urges caution about overconfident predictions on either extreme.

AGL — following a May 11, 2026 post‑earnings investor relations call — clarified its 2026 cash guidance is “at least $125 million” (versus $376M cash at end‑2025), which, together with the Q1 print and further analysis, left the author incrementally bullish about upside to EBITDA and the company’s balance‑sheet conservatism. The note presents simple math: ~18 million diluted shares, 2026E net cash ~$105M and a TEV of $885M, while the stock trades at under 0.2x TEV/Revenue and 2.4x Medical Margin — levels the author says reflect an outsized market fear of insolvency. The memo also highlights a presumed $250M prior‑year claims outflow that may be overstated, and corporate incentives (CEO grant vesting at $50/100/150 after grants made when shares were < $30). The author contends shares must rally to at least $150+ (on a longer path to $250) before being fairly priced as a going concern.

Handwritten Bangla compound character recognition is improved via a confidence-guided diffusion augmentation approach that synthesizes class-conditional samples, enhances U-Net blocks with Squeeze-and-Excitation residuals, and filters generated images using pre-trained classifiers. Fusing filtered synthetic images with real data yields consistent gains across ResNet50, DenseNet121, VGG16 and ViT, with a top accuracy of 89.2% on AIBangla. Only the abstract was available for this summary.

John F., born without arms, is a fully licensed driver who steers with his left foot and controls gas/brake with his right, restricted to automatic transmission and power steering. After seven years in a Model 3 and worsening hip arthritis, he upgraded to a Model Y with Full Self-Driving (FSD) Supervised; he says it dramatically reduces fatigue and preserves his independence.

Lake Cornelia Commentary posted a 32-minute recording of a live video on May 11, 2026 (© Judd Arnold) covering macro and oil commentary and discussing specific tickers/topics listed in the title: BW, AGL, TOI, EVH, ASTH, COMP, CAR, and LPRO. The piece is paywalled on Substack; the author thanks named viewers for tuning in.

HarmoWAM presents an end-to-end World Action Model that unifies predictive and reactive control by conditioning a predictive expert and a reactive expert on spatio-temporal priors from a world model, with a Process-Adaptive Gating Mechanism to coordinate switching. Reported zero-shot generalization covers three unseen real-world environments and six manipulation tasks, improving over VLA and WAM baselines by 33% and 29%, respectively. Full paper text not available (abstract only).

@KayvonJafar cites Andrej Karpathy’s line “I don’t think I’ve typed a line of code since December” and Garry Tan’s design prompt to have one person run a whole software team. He points to the open-source gstack as that answer, with Garry claiming ~810x higher pace; gstack wires AI roles (CEO, eng, QA, browser operator) into workflows like /office-hours, /autoplan, /qa, /review, pair-agent and parallel sprints, reframing development as operating systems of AI workers.

Marc Andreessen presents AI as an ongoing economic experiment where, contrary to zero‑sum Luddite expectations, programmers are working longer hours and feeling exhausted yet euphoric; he asserts higher marginal productivity expands work. He discussed this on MTS with Erik Torenberg, in a 00:00–52:25 episode covering Anthropic, AI doomers, jobs (coder→builder), AI psychosis, polls, UFOs, and career advice.

Aakash Gupta argues the biggest disruption of 2026 is mid-market web builders — the people who charged, for example, $150/hour to build dentist websites — not writers or designers. Tools like Bolt and Lovable let non-developers ship production apps in an afternoon (landing page, backend, auth, payments, admin) replacing what used to require “four years of CS plus three years of frameworks.” Market evidence: Upwork basic web rates dropped 60% in 18 months as buyers learned to self-serve. Gupta frames this as a recurring pattern when tools collapse the gap between intent and execution (Photoshop, Squarespace precedents) and says winners become 10x users of the tools. He plugs an AI Skills conf on May 14 (conf.cosprints.ai/?32) — free, 5+ hours, 8 AM PT/11 AM ET/4 PM London — with 20+ speakers and 5,000+ registered.

Sparse recovery from mixed-quality measurements (n1 high-quality, n2 low-quality) is studied via information-theoretic and algorithmic sample-size conditions. The paper proves a linear 'Price of Quality' trade-off: agnostic decoders cap one high-quality sample at no more than two low-quality ones, whereas informed decoders can exploit variance knowledge to make the price unbounded. LASSO matches homogeneous-noise thresholds, depending only on average noise. Published ICLR 2026 (arXiv 2026-05-11).

@freeaiguides' seven-section prompt structure (Task; Context Files; Reference; Success Brief; Rules; Conversation; Plan) is presented as a repeatable template for every Claude job. The thread assigns explicit roles to each section, urges clarifying questions before execution, and argues that prompt design — not the model — is almost always the deciding factor in useful outputs.

Ibrahim Khan condemns Emmanuel Macron’s Nairobi “Pan-Africanist” posture, citing a 2017 humiliation of President Kaboré, the 2022–2024 expulsions of French troops (Operation Barkhane), and the CFA franc’s 50% reserve rule as evidence of French domination. He labels Macron’s Gaza stance hypocritical and describes eleven Nairobi deals (railway to fiber optics) as PR-driven outreach after France’s Sahel defeats.

Pixal3D (Li et al., SIGGRAPH 2026) presents a pixel-aligned 3D generation paradigm that back-projects multi-scale image features into a 3D feature volume to create explicit pixel-to-3D correspondences, generating assets aligned with the input view. The method reportedly substantially raises fidelity—'approaching the fidelity level of reconstruction'—and extends to multi-view and scene synthesis.

IShowSpeed’s viewbotted stream in the Dominican Republic sparked platform and industry action after the creator revealed a reported 1.9M peak concurrent viewers was actually about 300K, with ~1.6M automated viewbots inflating the count. Viewbot services sell falsified views for roughly $0.28–$10 per 10 views; YouTube said Speed didn’t buy bots and warned channels face termination for repeat violations while Twitch CEO Dan Clancy announced temporary caps on concurrent viewers for streams using bots. Research from StreamCharts flagged ~10% of Twitch channels averaging 50+ viewers (16% on Kick) for suspicious activity, a problem that undermines brand measurement. The newsletter also profiles Dhar Mann Studios’ CEO Sean Atkins—who used direct access, community-led proposals, and collaborative 'and not or' deals to land partnerships with NFL, Fox and Samsung—and notes creator-economy headlines: Druski reinvested $100K into a sketch with 150M+ views, OnlyFans reached a $3.15B valuation after a $535M partial sale, and Hunter Peterson amassed ~371K non-binding pledges (~$337M) for a jokey Spirit Airlines buyout.

Profound's Marketing Engineering Hackathon (June 6, 2026, Union Square, NYC) will host 50 builders for a one-day challenge to automate inhuman-scale marketing processes, offering $40,000 total in prizes and judges from Ramp, Stripe, and MongoDB. The platform-agnostic event allows Claude Code, Cursor, n8n, Python, LangChain, or raw APIs; winners receive cash, Profound credits, and interviews. Application required.

Variational Inference for Lévy Process-Driven SDEs via Neural Tilting introduces a neural exponential tilting framework that reweights the Lévy measure with neural nets to build a flexible variational family preserving jumps. A quadratic parametrization gives closed-form normalization; a conditional Gaussian representation for stable processes enables simulation and scalable, symmetry-aware Monte Carlo optimization. Summary based on the abstract (full text not reviewed).

GBrain underwent a rapid three-day sprint: 14 PRs merged and nearly 29K net added lines, pushing the project from v0.31.2 to v0.32.4. Key work focused on hot-memory and real-time fact extraction (#885, #796), a major rewrite of take handling (#795), routing/resolver improvements, embedding recipes, auto-upgrade and community fixes.

GeoProto addresses cross-domain few-shot medical image segmentation by enriching prototypical matching with explicit geometric priors. The method's core, GAPE, augments appearance prototypes with learned ordinal offsets from an Ordinal Shape Branch trained under an ordinal-consistency loss (no extra labels beyond segmentation masks). According to the abstract, GeoProto attains state-of-the-art results across seven datasets and three settings (cross-modality, cross-sequence, cross-context). Only the paper abstract was available for this briefing.

Author @emollick praises frontier model writing for its sense of style, tone, varied sentence structure and memorable phrasing, but criticizes its fiction weaknesses and repetitive tics, saying ubiquity has made it clichéd. Roon (@tszzl) concurs on clarity and tics but rejects the claim that model prose lacks analytical or informational value.

Redistricting in 2026 has tilted modestly toward Republicans after a combination of red‑state map redraws (notably Florida) and two legal blows to Democrats: the Supreme Court’s Callais decision that weakened the Voting Rights Act and the Virginia Supreme Court’s procedural nullification of a Virginia referendum that had appeared to favor Democrats. Nate Cohn’s district estimates move the national map from near‑parity (median district ≈ 0.1 point R of the country) to a Republican advantage of roughly R+2.5 to R+3.9; Silver argues the Virginia “+4” story was probably a 2.5–3 seat probabilistic gain. Democrats still lead the generic congressional ballot by 6.1 points in Silver’s tracker, but Polymarket cut Dems’ House‑majority probability from ~87% to ~78% after the rulings.

Silver emphasizes that courts — especially a 6–3 conservative U.S. Supreme Court and a modest Republican edge among state high courts (his AI‑assisted tally: R 264 EVs, D 252, tossup 22) — will shape how aggressive partisan maps fare. He recommends Democrats avoid both despondency and complacency: pursue state legislative and judicial gains, use probabilistic modeling (Silver Bulletin’s midterms model due in ~6 weeks) to assess outcomes, and balance aggressive redistricting in blue states with sensitivity to public opinion and potential legal scrutiny to reduce long‑term vulnerability heading into 2028.

Howard Schultz argues that Seattle and Washington have become hostile to the companies that built the region’s prosperity, saying Starbucks will move hundreds of corporate roles to Tennessee and that he no longer resides in the state. He attributes the shift to political rhetoric from Seattle Mayor Katie Wilson and to state fiscal choices—highlighting a 10.55% sales tax in Seattle and a tax system he calls regressive—that he says discourage business growth. Schultz points to slowed hiring at Microsoft and Amazon, downtown vacancies, chronic homelessness, budget deficits and falling school outcomes as signs the ecosystem is fracturing. He calls for tax‑code reform, accountable public spending and pro‑entrepreneurship policies (referencing a bipartisan NGA initiative) while noting his family foundation remains invested in Washington’s future.

France–Kenya agreements: @FrenchResponse says the ten agreements signed between France and Kenya on May 12, 2026 contain 'nothing secret' and offers to publish all details. New Direction AFRICA alleges Macron and President William Ruto secretly signed 11 instruments at Nairobi's State House with no parliamentary debate, calls it 'Françafrique 2.0' and warns of lost sovereignty and resources.

When you trade on Polymarket with PolyHelper, you automatically farm two airdrops — Polymarket’s and a PolyHelper airdrop — while PolyHelper reimburses trading fees (they refunded all fees paid to Polymarket on a randomly chosen day, May 8). PolyHelper also adds extra UX info/features and is free; the author (@0xd1namit) urges there’s no reason not to use it.

The paper studies finite-sample hypothesis testing for the type I/type II error trade-off function between two unknown distributions. Without structure testing is impossible; they identify exact attainability of Neyman–Pearson rejection regions by a class S and show finite VC-dimension of S is necessary and sufficient. They give a test with nonasymptotic type I control and uniform power under explicit separations, produce simultaneous confidence bands, analyze local rates (monotone likelihood-ratio) with near-matching lower bounds, and extend to approximate attainability (e.g., univariate log-concave).

Pirate Wires Daily's 'Three Morning Takes' (published May 12, 2026) delivers three punchy cultural notes: a weekend campaign gaffe by Steve Hilton — a British‑American former Fox host and one of 62 gubernatorial hopefuls — who sparked online ridicule for calling a hard‑shell taco a 'street taco'; a housing snapshot linking last October's OpenAI secondary sale (more than 600 employees sold roughly $6.6 billion in shares, with about 75 reportedly netting ~ $30 million each, per the WSJ) to escalating San Francisco rental extremes (the newsletter cites a 1BR in Alamo Square at $8,000/month and reports of bids requiring six months' rent upfront from The Standard); and a profile of Rachel Dolezal/Nkechi Diallo pivoting to sex coaching amid OnlyFans activity and a Daily Mail‑reported $8,847 welfare‑fraud charge. The piece mixes sourced facts with sarcastic commentary on politics, housing, and culture.

Author @aakashgupta (2026-05-13) shows that Claude's routing relies on a single-line skill description, causing sub-100-character descriptions to be invisible. He audited 25 skills (75 runs), fixed a failing recipe skill by rewriting its description with explicit triggers and boundaries, and provides a six-rule post-load checklist plus an eval prompt and audit kit for reliable skill design.

Goma has been living under M23 control for more than a year, and the city’s inhabitants—what the author calls the forgotten people of eastern Congo—face a mix of order and acute suffering. M23 has introduced measures modelled on Rwanda (a new police force with Rwanda‑style uniforms and compulsory Saturday cleanings) that have reduced petty crime, even as everyday chaos persists. The author visited in early May 2026, was given access to M23’s leadership and shown a slide deck marketing the group’s mineral resources to the Trump administration. Médecins Sans Frontières has resumed a clinic on Nyiragongo’s slopes treating pneumonia in infants and feeding queues of malnourished children, but many aid groups have withdrawn and the UN’s World Food Programme deems much of the region too dangerous for full‑scale operations. Diplomacy (US Washington Accords; Qatar mediation) and March 2026 US sanctions on Rwanda have not yet produced a lasting ceasefire.

The Polymarket market “Putin out as President of Russia by December 31, 2026?” was used by @0xd1namit to farm LP rewards (reward pool > $100, ~4% yields) and capture small spreads. He bought, relisted 1¢ higher (+1.14%) and sold quickly; he considers the outcome low-risk and thinks the fair price exceeds $0.89.

CloakBrowser is a MIT‑licensed stealth Chromium (pip install, ~200MB) that the author says achieves human-like anti-bot fidelity — 0.9 on reCAPTCHA v3 versus stock Playwright's 0.1 — by patching Chromium's C++ in 16 spots (canvas, WebGL, audio, fonts, hardwareConcurrency, GPU strings, WebDriver flag, TLS fingerprint). It claims to pass 14/14 detection tests, auto-resolve Turnstile, and replace commercial $200–$500+/month stealth browsers or costly custom builds.

CapVector introduces a method to decouple auxiliary-objective finetuning into two parameter-space components: general capability enhancement and task-specific action fitting. By finetuning two small-scale models with different strategies and taking their parameter difference as a capability vector, then merging it into pretrained VLA models (plus an orthogonal regularizer), standard SFT attains auxiliary-like gains with lower compute. Full text not provided here.