All desks

Use reflective prompt optimization after the baseline, with a real budget and an untouched final test set.

HybridAgent mixes internal vector-based computation with text for tools and inspection. Matched-model tests show stronger search results, but the cost savings cover estimated generation compute.

A small code-search project makes hosted reranking optional and exposes a provenance bug: local fallback results still carry the Jev provider label.

An inspectable highlighting service separates whether a paragraph deserves attention from uncertainty about its colour and importance.

Alan Milburn and Peter Fonagy propose earlier, integrated help for young people. Financing it through welfare cuts raises a separate question about security while that help arrives.

Karri Saarinen on earned intuition, shared product judgment, and why individually sound features can make the whole product worse.

Gbenga Awodokun’s six-transcript comparison exposes different failure patterns, modest whole-workflow savings, and the limits of a clinical-safety headline.

Montreal’s game workers confront an industry built with public subsidies but governed by private owners, with passion used to excuse insecurity.

E. Ahmet Tonak remembers a teacher who made Marxist economics measurable and opened a path from political exile to a lifetime of collaboration.

Build a text parser from small functions, then read the HN discussion for the failure cases a friendly introduction leaves aside.

Two cerebral palsy advocates describe the loss of support at adulthood and call for lifelong care that includes specialist mental-health services.

A revival of Sarah Kane's 4.48 Psychosis gives depression six overlapping voices. Clive Paget asks what music reveals, and what it fixes too firmly in place.

GitHub Trending, week ending October 18, 2026: one new pick after excluding repositories already covered in recent roundups. storytold/artcraft — an AI image-and-video editor with 2D layers, 3D scene staging, and camera controls for artists …

Urban Park Rangers lead a free one-hour fall tree walk in the East Village. Meet at Avenue A and St. Marks Place; no registration required.

A free art installation at 279 Calyer Street gives every visitor one plush flower. Open 10 a.m.–6 p.m. through Monday, October 19.

Marcus Johansson compares photo-cropping methods against his own marked focal points, exposing the gap between finding an object and keeping the right subject visible.

A small hands-on review shows an injection rewrite bypassing Jev, an uncertainty fallback that never fires, and why hosted screening cannot make a prompt local-only.

Tim Brinkhof asks developers what long-form video essays offer: a language for player feelings, overlooked experiments, and ideas that mature between projects.

John Goodman on the physical craft behind his comic exuberance, Roseanne's working-class world, and learning to work without punishing himself.

Sunil Pai on recovering momentum through smaller work. Rhys Sullivan revisits YC and shows how clearer contracts made his coding agents more useful.

Cloudflare AI product lead Michelle Chen explains model launches, GPU economics, licensing, and why cheap inference subscriptions may tighten.

Dillon Mulroy, Rhys Sullivan, and Sunil Pai debate AI containment, then inspect a coding workflow built around scoped access and visible end-to-end tests.

Dillon Mulroy, Rhys Sullivan, and Sunil Pai argue for products that serve a customer's chosen agent. Their examples connect APIs, documentation, review loops, cloud environments, and burnout.

Dillon Mulroy, Rhys Sullivan, and Sunil Pai ask why more generated code has not produced better software. They compare architecture, token costs, model access, and autonomous loops.

Dillon Mulroy shows how Pi, Plannotator, and type-driven plans keep him involved in agent-written software.

Pi Durable, a live Lisp demo, and the cost of agents whose work is getting harder to inspect.

Wolfe’s weekly roundup contrasts Krea’s editing workflow, downloadable models, one-shot coding demos and the smaller updates behind product launch headlines.

Martin Keen maps four context failures to four fixes: validate mistaken notes, compact long histories, curate tools and remove superseded information.

A customer-support example walks through tool design, selective memory, testing, tracing and safeguards for agents that take real actions.

Free Harlem jazz, Open House New York’s final-day drop-ins and a Greenpoint block party lead Sunday’s outings, with museum exhibitions, self-guided walks and a Thursday folk gig to follow.

DeepSeek releases an updated Flash model DeepSeek updates Flash with downloadable weights and better coding results — Developers get a concrete alternative to test in coding assistants, with integration details to check before switching. An …

Victoria Guida's reported column connects housing frustration to betting's promise of a shortcut, while examining the savings displaced and the safeguards being proposed.

Michael Rubin's advertising forecast puts sportsbooks and prediction markets in the same contest for customers, backed by Fanatics' merchandise and collectibles profits.

A 1982 research agenda for better tools, trustworthy reusable components, and prototypes that expose misunderstood requirements before implementation.

An inspectable Gemma-based Jev alternative exposes its probability readout, public benchmark results, and a critical difference between its calibrated 12B and raw-probability 31B adapters.

Review 50–100 traces, find recurring failures, and show how fixes reduce errors. Build support for evaluation through visible product improvements.

A Windows UI Automation project makes Jev’s action menu inspectable. Offline tests pass, while mocked probes expose gaps between the checked decision and the executed command.

inference-benchmarks records complete model exchanges, separates valid output from correct decisions, and exposes the request-shape tradeoff in a Qwen comparison.

Workers rarely reach the ballot. Jared Abbott asks why a major new study discounts the unions that once helped them govern.

Caleb Brennan traces how Cold War nostalgia ties domestic renewal to military rivalry, and argues for a socialist foreign policy built around peace.

Brooklyn Dance Center brings seasonal performances to the Salt Marsh Nature Center on Sunday afternoon. Bring a blanket or chair; admission is free.

Marine Park's annual Festival of Lights gathering brings Pandit Ravi Doobay to the Carmine Carro Community Center. Free, with no registration required.

A local Jev-style router exposes a confidence-field mismatch, missing context, and the harder problem of proving that a bigger model helps.

A measured TweetTopic example combines typed labels, per-tag error counts, bounded concurrency, and explicit usage accounting. One broad tag stays unreliable even at a high threshold.

Tim Keenan let isolation shape Duskers' systems, then used player feedback to discover the reaction time its survival mechanics needed.

Andrew Lipstein describes how compulsive rereading consumed his writing life, and what it cost to leave a sentence unfinished.

Naomi Fry on a documentary whose invasive intimacy reveals how little access can tell us about another person.

Executable checks and a clear cost score let Claude redesign quantum circuits, with occasional human pushes toward different approaches.

A visual explanation of how language models remember, built around the interference that makes recall fail.

Free park music and Open House New York uptown drop-ins, a Culture Lab dance party tonight, and Harlem jazz tomorrow.