Claude helps solve a decades-old security mystery Anthropic uses Claude to trace how digital-signature settings were chosen — The result comes with reproduction code and an outside verification of a long-disputed piece of security history. …
Ben Carsley explains context-specific writing styles, the limits of prompt-only personalization, and shared model serving with a small adapter per person.
White paint carries histories of poison, design, and racial power. A skeptical review asks how those histories become an argument.
Anthropic measures agent-written code by human attention and reversibility. HN asks what happens to code understanding, security, and the token bill.
Nicole Noren's account of rejected investigations puts ESPN's sportsbook relationships alongside the reporting audiences lose.
The Next New Thing on REA, test-driven coding, HyperFrames, branded diagrams, context compression and text-to-CAD—with limits on the bigger claims.
A talking dog, tabloid Pop art, and the feminist ingenuity of a Bronx artist working across the boundaries imposed on her.
Mercor maps AI tests against American jobs and finds large blind spots. The coverage gap measures how little we know about models at work.
Keep evaluation inside the development loop, give a trusted domain expert the final call on quality, and build annotation tools around their work.
An inspectable testing framework pairs Jev with Playwright, measures complete workflows, and shows how missing field state can produce a false pass.
A Laya-based, Jev-style moderator exposes a useful failure: a local no-link shortcut can allow self-promotion before the model or confidence gate runs.
Anahid Nersessian argues that protecting people from sexual violence must also mean defending their sexual agency.
Rain presses for specific credit to Mononoke's design, then accepts Anthropic's added attribution. The discussion separates that correction from disagreement over the engineering achievement.
Cole Medin keeps agents building in parallel, with people defining each release and testing the result before production.
Sharp texels, palette-based lighting, and a useful reminder that a faster rendering backend can change the work it renders.
Sunday's Family Day pairs drop-in artmaking with Woody Guthrie's Songs to Grow On, an interactive performance designed for ages 2–5.
A measured prompt-injection test shows how untrusted policy-shaped text can suppress attack detection and lower an unauthorized tool call’s risk score.
A bilingual synthetic benchmark compares hosted and local decision models, with inspectable responses, confidence failures, and the operational cost of self-hosting.
Joshua Rothman questions the equation of concentration with bodily discipline: perhaps progress lets us sit still, rather than the reverse.
Max Norman argues that Hujar's intimate portraits challenge a photographic canon built around outsiders looking in.
A coding agent stalled on a Linux scanner driver until it could observe the working Windows version. The useful change was the feedback loop.
An illustrated monarch-migration exhibition opens uptown, with free rock, an acoustic art show and a free horror screening tonight, plus weekend and ongoing outings.
Muse Spark Open releases weights, with substantial deployment requirements Meta releases a model that needs at least 700GB of graphics memory — running it yourself calls for expensive hardware and a close look at the licence. The model card …
Richard Brody on secrecy as movie marketing—and why protecting a twist can stop critics from discussing what a film actually does.
Generating research is not the same as advancing understanding. A physicist asks what AI is accelerating—and who gets left checking the answers.
Discovered Labs' founders distinguish retrieval from citations and argue for clear positioning, genuine corroboration, and measurement that accounts for uncertainty.
Demetrios Brinkmann's everyday automation examples lead to harder questions about permissions, model-harness compatibility, and whose interests shopping agents represent.
Sportsbook-funded political groups are returning to state election advertising, with ads that leave their gambling interests unmentioned.
The NBA restricts bets tied to isolated player actions and adds harassment protections after gambling scandals exposed the risks to athletes and competition.
Jev is a strong default in Parallel’s human-labeled search tasks, but prompt tuning changes the comparison and a specialized reranker still wins.
Ottoman campaign memoirs and Ruhi Su's folk songs preserve different histories of Yemen: imperial strategy above, bereavement below.
Current Affairs draws a boundary around generated prose and illustration, while allowing tools that assist the people doing the work.
What if the apparent tangle inside language models comes partly from bad microscopes? A case for studying the original neurons, with important limits.
Hamel Husain argues for validating AI judges against human labels before adding another model provider: measure precision and recall, then optimize cost.
Nick Kuhn argues that agents still need the durable state, scoped identity, approved tools, isolation, and observability familiar from platform engineering.
A demand for public evidence meets a disagreement about whether better AI can quickly become cheaper goods and automated jobs.
Zach Landes tests decision models on trolley dilemmas, description order, and logically checkable controls. The useful finding is fragile judgment—not proof of a survival instinct.
Brussle’s evaluation separates yes/no, category, and ordered-scale decisions. Good average probabilities do not guarantee reliable thresholds on difficult cases.
AI-generated proofs raise a question beyond job replacement: what happens to the shared pursuit of understanding?
Football broadcasting offers a surprising counterexample to American anti-intellectualism: an audience that expects complexity rather than resenting it.
Guardian readers describe late autism and ADHD recognition, assessment queues and the gap between an explanation of their difficulties and practical support.
NVIDIA's Ming-Yu Liu explains sensor-specific adaptation, batch-size-one inference, and world models as tools for faster policy evaluation rather than automatic proof of safety.
Leadership coach Reed Frerichs offers a compact framework for changing roles without confusing confidence, identity, and evidence.
Daniel Whitenack and Chris Benson distinguish AI proficiency from independent understanding, and ask what schools should measure when polished output is easy to generate.
Hard Fork examines Anthropic's meetings with religious scholars, uncertain claims about Claude's consciousness, and the gap between moral consultation and demonstrable change.
The AI Automators tests screenshot-backed support tickets, separates confidence from calibration, and shows why a bounding box is not proof that the target exists.
For the Love of Fragile Things brings Caribbean horizons and colonial memory to a free Lower Manhattan gallery. Thursday evenings run until 7 p.m., with special closing-Sunday hours.
Aperture's new Upper West Side gallery opens with photographs by Diane Arbus, Dawoud Bey, Carrie Mae Weems and more. Free, with an amNY-listed run through November 1.
HiddenLayer’s checkout and toxicity-filter tests show how typed decisions still fail under manipulated input. The evidence establishes a failure mode, not a population attack rate.
A five-agent study separates cheap action selection from successful task delivery: recovery, auxiliary reasoning, and completion checks determine whether the savings survive.