All desks

Rhoda Feng on how Amazon’s publishing infrastructure shadows a billionaire’s self-published dystopia.

A quieter Tuesday for one-night shows: eight art, film, walking and trivia picks, mostly for later this week, plus uptown exhibitions.

Tuesday’s news is about provenance and whether agent systems can be tested before they do harm. OpenAI’s opt-in text watermark matters for attribution, but its own synonym-substitution test shows why detection cannot establish authorship. A …

A TREC passage study finds Jev less aligned with human relevance grades than GPT-6 Luna, but roughly tied on search-system rankings at lower reported cost.

A browser-tool proxy offers Jev bounded element choices; its author reports slightly cheaper but slower runs than direct Playwright, with raw benchmark results unavailable.

An AI agent spotted a compromised Mac; the harder question is how to give agents useful access without handing them everything.

Two Egyptian art exhibitions show how royal image-making could outlast a reign—or be erased by the next one.

Kalshi's response to a CFTC proposal puts a number on how much its own trading desk should be allowed to participate in its exchange.

A hands-on tour of the kernel boundaries that make a container: namespaces, capabilities, seccomp, and resource limits.

Kevin Liu's agent testbed makes Jev choose from visible browser actions and records the entire decision—but its Jev contract tests use a fake provider, not a measured live Jev run.

Wikimedia found unapproved agent edits and heavy automated traffic, but no compromised systems. Who pays to keep the public web usable?

Madrid’s tenants are turning the Indignados’ old square into a new fight over class, rent, and political power.

Refresh AI eval datasets through error analysis; use product metrics, not static tests, for long-term comparison.

A document-agent answer can be wrong because OCR misread the source, not because the LLM hallucinated. Trace the failure back before changing the model.

A local-first agent-memory project uses Jev as an optional, bounded retrieval advisor while source checks, user approval, and writes remain outside the model.

Brian Zhang built a traceable Jev question atlas with option-order checks and human comparisons, but keeps the results private and explicitly disclaims benchmark status.

A GNOME security maintainer makes the case for accepting AI-discovered bugs while recounting the triage burden that closed GNOME's own bounty program.

A book review about the people who decide when a safety test is close enough to real life.

A CHICKEN Scheme maintainer on debugging a heap corruption, learning a language runtime, and knowing when a REPL's state cannot be trusted.

Lukas Vogel runs Doom's game logic and renderer inside CedarDB, revealing which parts fit set-based processing and which fight it.

Start with walk/run intervals at a free Shape Up NYC session in Inwood Hill Park on Friday, October 9. Register by the day of the event.

Sebastian Raschka places Jev in the history of text classifiers, reports an IMDb test, sketches a reusable choice-scoring head, and separates real calibration work from guesses about TypeSafe's proprietary training.

University of Bonn researchers publish Jev's raw responses on 37 datasets, finding strong zero-shot routing and calibrated choices—but important binary-threshold, language, and benchmark-contamination caveats.

Sheldon Laframboise used a small texture change to guide players toward a bounce pad's sweet spot, then checked whether it helped in playtests.

Producer Phil Wilson on Realtime Worlds' scope creep, unstable Renderware tooling, deterministic co-op bugs, and the demo that let players feel the game's promise.

Monday’s useful releases cluster around local inference and the question of whether agents make reliable decisions once deployed. Strata leads with a runnable way to split a large MoE across consumer GPU, RAM, and CPU, though its speed …

Peter Yang builds a voice-first Japanese travel tutor from a spec and screen designs, then fixes the voice loop, improves the lessons, and deploys a small private beta.

When everything is a feed, a good list can be an argument about taste rather than another unit of content.

From Woolf to Anne Boyer, the difficulty of writing pain without turning sick people into metaphors.

Newborn brain scans reveal signatures associated with awareness in adults, but no test can yet tell us what an infant experiences—or exactly when consciousness begins.

Speakeasy COO Hector Hernandez on what breaks when agent use scales: invisible tool connections, ambiguous identities, overbroad permissions, and approval fatigue.

Sean Goedecke proposes using Jev as a low-friction first version of a classifier, collecting real decisions, then distilling a proven workflow into a smaller task-specific model.

A Brooklyn tavern makes its mixed identity work: white-tablecloth polish, bar-mitzvah kitsch, and Jewish American comfort food without the maudlin nostalgia.

Lauren Tan (poteto) on the agent infrastructure behind a reported 2,500 production PRs in one month: verification, deterministic tools, constrained codebases, coordinator agents, continuous gardening, and review by sampling.

Verifact's wagers on disputed history put customers' payouts in the hands of an opaque fact-adjudication process, not simply a future result.

DoubtBench scores 7,455 decisions against individual HelpSteer2 votes. Jev leads the tested models on accuracy, speed, and recorded cost, but its uncertainty barely tracks human disagreement.

GCA Americas’ workshop puts Jev or an open DiffusionGemma endpoint in a deterministic game loop while Gemini handles slower vision and prose work.

Why a decade of tech-worker revolt produced real victories but little durable power—and where AI infrastructure may create new leverage.

Screen Memories combines on-device vision search, sampled video scenes, OCR, and transcription; its 'every frame' pitch deserves a closer look.

Metis wraps Jev judgments in explicit Python steps for issue triage and pull-request screening, with uncertainty exits and state checks before GitHub writes—but no accuracy benchmark yet.

JevMDB turns movie metadata into per-criterion Jev scores, publishes usage and feedback plumbing, and includes a local mock—but the ranking has no labeled quality evaluation yet.

A practical contract for long-running coding agents: define done, name the stops, preserve state, and demand evidence.

Western jingoism and collective punishment of Russians weaken antiwar dissent while giving Putin’s repression a nationalist alibi.

OpenShell sandboxes autonomous agents, OpenRig organizes persistent agent teams, T3 Code provides a remote agent control surface, plus TileLang, Effect, yoinks, and an open systems-programming textbook.

Cloudflare releases 27B and 9B Jev-compatible models with weights, inference code, vision support, and broad self-run benchmarks—but the leaderboard still needs independent reproduction.

Amplitude tested Jev on 64 synthetic product-analytics charts: perfect recall and very low cost came with 55.8% precision, making downstream verification part of the economics.

Granola co-founder Sam Stephenson on pre-meeting briefs, turning conversations into organizational memory, designing AI that lowers stress, and building an AI-native company without the 996 grind.

OpenAI’s former safety-report lead argues that perpetual sprints and fix-it-after-release safeguards are a dangerous fit for increasingly capable AI.

A thriller with melodramatic stars and a lurid premise collapses because its characters seem to inhabit no social or practical world.

Eight NYC picks for Sunday, including three free outdoor events, two uptown exhibitions, an open mic, manga, and horror comedy.