All desks

Johnnys.News explains an inspectable embedding-based decision demo, with cached options, browser inference and unusually explicit limits on its probabilities.

Bardia Pourvakil shares concrete failures in batching, citation support and test-quality judgments, plus a reusable design guide from GC AI’s hackathon.

Osita Nwanevu challenges literacy-collapse nostalgia: mass literacy is a democratic achievement, and serious reading needs a future beyond print.

AI can generate mathematical results. That does not settle whether students will have a place—or a livelihood—in understanding them.

Mundane programming work develops taste and judgment. Automating it away may remove the apprenticeship, not just the drudgery.

Ben Hylak explains change-focused agent testing: replay production traces in a reconstructed environment, inspect behavioral differences, and validate the simulation before trusting it.

Ten repositories spanning native game porting, coordinated coding agents, HTML-to-video, persistent memory, research tools, and self-hosted workout tracking.

AI can increase individual output while weakening the human exchanges that make intellectual work worth doing. An early HN discussion puts concrete relationships behind that concern—and challenges the inevitability of the conclusion.

Pusu’s browser-agent experiment exposes premature DONE, adds deterministic checks, and publishes a safety probe whose rules-only result we reproduced.

Private assessment providers can shorten individual waits while drawing staff away from NHS teams. New oversight addresses diagnostic standards, but workforce capacity remains the harder problem.

Competition creates wage inequality — but understanding its machinery can help workers change the terms of the fight.

AI-generated mathematics still owes readers an explanation: a set theorist argues that unreadable results offload the work onto everyone else.

Treat telemetry export as a product capability: OTLP destinations, preserved context, and an explicit trade-off between tenant isolation and operational overhead.

Forest Hills gets an Austin Street celebration on October 25, with free music, games and neighborhood events across all five boroughs.

A 15-minute visit to the Municipal Building's cupola costs nothing. Reserve ahead for a small-group view of Lower Manhattan.

Unsloth documents serving local Laya through /v1/systemone, with explicit warnings about confidence thresholds and trimmed choice descriptions.

Ves Stoyanov reports six classification experiments where editable rules compete with fine-tuning, and uncertainty sampling misses confidently wrong routing policy.

Working-class memory, the market for Southern nostalgia, and the reporting labor that a celebrated byline can conceal.

Medium Build at Webster Hall tonight, uptown exhibitions, autumn art walks, and upcoming neighborhood outings.

Friday’s useful AI news is about making agents safer and more predictable, rather than another frontier-model launch. Claude’s usage-policy update leads because it changes deployment requirements for high-risk recommendations and physical …

Sina Booeshaghi compares Jev and OpenAI on personal research relevance and RELISH, with a useful distinction between ranking scores and calibrated probabilities.

Shawn Xu's technical write-up shows how simulation and better decision objectives lifted an Ironclad agent's win rate—with a major hidden-information caveat.

Justin Chang on religious outsiders, institutional power, and solidarity without agreement in Cristian Mungiu's family drama.

Use AI to get unstuck, not to skip the practice that makes you capable of judging its code.

An AI-generated proof is not the same as a field learning something. Tao argues that the rewards need to change.

Proposed risk triggers and a ban on AI-customized bonuses would tighten sportsbook oversight, but can still miss compulsive betting at smaller amounts.

The Wall Street Journal reports an investigation into well-timed appointment wagers, highlighting the information advantage possible in political event markets.

The Fonagy review rejects the snowflake caricature and asks why mental-health support is locked behind a diagnosis that can take more than a year to obtain.

The Fonagy review proposes needs-led mental-health and neurodevelopmental care, protected access to diagnosis, and tighter oversight of private assessments.

Ofsted's own commissioned survey finds widespread reports of harm to staff wellbeing under its revised inspections, even as respondents praise inspectors' conduct.

A Telegram moderation implementation uses Jev to classify messages and link spam fragments, while keeping failures open and evaluating self-hosted alternatives in shadow mode.

A month of cheap coding-agent work suggests that 'good enough' changes what you delegate; practitioners disagree on where it stops being good enough.

When the mandate is 'add AI,' look for a workflow worth accelerating—not a chat box to bolt onto an existing screen.

A blind, small-sample writing comparison finds some improvements over Opus 5, but generic phrasing, misplaced detail, and bot-like replies survive.

Frost's late letters complicate the image of a serenely skeptical poet.

DSA can win elections; its harder task is building lasting working-class power beyond the ballot box.

A sponsored but useful head-to-head finds where Jev's cheap typed decisions hold up, and where an encoded shell command defeats a model-only safety check.

Stephen Solka tests Jev against political outreach mail, mines difficult counterexamples, then builds a tiny local classifier; his held-out results and threshold caveats are the story.

An email-security company's synthetic hard-case benchmark finds decision APIs promising, but deployment thresholds, class tradeoffs, and serving costs complicate the leaderboard.

A one-prompt design experiment produces a striking interactive cityscape, but the real question is what a designer still needs to do.

Jan Kandiyali argues that a socialist good life means useful work with others, not just a fairer division of its proceeds.

Zotero's origin story shows why understanding a workflow together takes longer than writing a prototype — and why that time can make software last.

An A/B test of 973 coding-agent threads found much lower median LLM cost, while its merged-PR metric could not establish either a quality gain or equivalence.

Four measured search experiments show better ranking, a relevance-versus-diversity tradeoff, a confident query-filter failure, and promising document splitting.

A newly compressed Darwin biography shows what rewriting can reveal—and what it can erase.

Tonight: free Ghibli trivia in LIC, a Halloween-TV mixer in Bushwick and Peter Wong’s Astoria stand-up; uptown exhibitions are still on, with an Upper West Side movie club ahead.

Thursday’s AI news is about the cost of narrow agent work and the tests that tell you when a cheaper route is safe. Claude Haiku 5.5 leads because its pricing could change how often hosted agents delegate small tasks, though its coding …

Long-running UK survey data link successive political and pandemic shocks to persistent psychological distress, with a warning against mistaking population estimates for individual diagnoses.

A large US student survey puts procrastination ahead of other reported obstacles to academic work, but its snapshot cannot explain what causes the delay.

A local Jev-compatible endpoint reads answer-token logits from Qwen rather than training a new classifier; its published results also show why confidence gates need testing.