HN Flash

Hacker News. Daily summary. Top 20 stories.

2026-11-04

Postman silently auto-installs AI plugins and attempts to share all API data (GitHub)

Postman auto-installed AI plugins and modified developer tool configurations without user consent.

Migration to Bruno dominates as users flee Postman's enshittification after this severe trust violation.

1388 pts · 458 comments

The 'Astra can read and write weights to improve Qwen' post is misleading (honnibal.dev)

The original blog post misrepresents how Astra improved Qwen by conflating multiple techniques into one claim.

Two camps: those saying the original post was clear, those saying the abstract misleads about how Astra improved Qwen.

1000 pts · 204 comments

Stringman, an $800 open-source room scale gripper robot (neufangled.com)

Stringman is an open-source, four-motor cable robot that grips objects across an entire room for $800.

Commenters split between praise for its clever design and concerns about cable entanglement in homes.

674 pts · 212 comments

Qwen 3.8 27B on my laptop can interpret model weights, too (Simon Willison’s Weblog)

Qwen 3.8 27B running locally can identify games from raw neural network weights.

Commenters split on whether weight-space programming is genuine reasoning or a clever setup exploiting known games and specific model pathologies.

673 pts · 195 comments

Google will turn off search for schools that reject AI (nytimes.com)

Google will block Search for K-12 schools that disable its AI features.

Dominant view: Google is using its search monopoly to force AI on schools.

637 pts · 439 comments

OpenAI seems to be silently re-enabling model training on Codex chats

Users report OpenAI silently re-enabling data sharing for model training in Codex chats.

Many see it as proof checkboxes can't be trusted; others blame buggy software.

500 pts · 149 comments

Kalshi and its co-founders face felony charges for violating Utah gambling laws (Ars Technica)

Kalshi and its co-founders face 13 felony charges each in Utah for violating gambling laws.

Strong consensus that Kalshi and Polymarket are gambling sites, not legitimate prediction markets.

485 pts · 472 comments

Could a modern chess engine on a Pentium have defeated Kasparov in 1996? (nicholas.carlini.com)

Article body wasn't reachable. HN discussion still summarized.

Most commenters find the result unsurprising given mid-2000s hardware already surpassed Deep Blue.

476 pts · 234 comments

Safari introduces Adaptive Dark Mode in 27.3 (WebKit)

Safari 27.3 beta adapts light websites to dark mode, ignoring those with native dark mode support.

Commenters split: some praise the feature, others fear broken sites and note the privacy-driven lie to JavaScript.

465 pts · 242 comments

Mac at any scale (Namespace)

Namespace built custom Mac racks using MacBooks due to multi-month Mac Mini delivery delays.

Split on EULA compliance: Apple permits 2 VMs but restricts commercial resale of access.

397 pts · 216 comments

US tech giants' hidden AI debt swells 90% to $3.2T in 6 months (Nikkei Asia)

Hidden debt from AI investments by US tech giants surged 90% to $3.2 trillion in six months.

Two camps: those seeing rational FOMO-driven investment and those warning of a 2008-style bubble with bailout expectations.

381 pts · 156 comments

The Little Book of Generative AI Foundations: An Intuitive Mathematical Primer (arXiv.org)

Tianhua Chen's book offers a derivation-focused primer on generative AI mathematical foundations.

Comparisons to Fleuret's 'Little Book of Deep Learning' and practical reading tips dominate the thread.

362 pts · 56 comments

C++26 Disapproved (Andreas Fertig)

C++26 DIS ballot failed due to national bodies rejecting contracts feature.

Split between those who blame broken process on contracts proponents and those who see a wider trust breakdown.

358 pts · 255 comments

Jelly Sheet (mrdoob.github.io)

A 3D jelly-like cloth simulation demo by Three.js creator Mr.doob.

Two camps formed: those defending AI-assisted coding as a tool and those arguing it undermines creative challenges.

324 pts · 82 comments

The Internals of PostgreSQL (interdb.jp)

A detailed guide to PostgreSQL internals, updated through version 19-beta.

HOT updates praised as key section; debate over vacuum-based MVCC vs. alternatives.

316 pts · 23 comments

Nvidia – open-source auto-ontology (GitHub)

NVIDIA's open-source Auto-Ontology framework adds semantic context to help agents generate fewer incorrect SQL queries.

Skepticism about fixing unreliable agents by adding more unreliable agents, countered with examples of reliable systems built from unreliable components

282 pts · 138 comments

AI is forcing us to pay off 40 years of technical debt (IEEE Spectrum)

AI-aided researchers turned a PHP crash bug into a critical RCE vulnerability, signaling a wave of exploitation of legacy software.

Commenters split between welcoming the forced debt payoff and fearing the endless patching burden AI will create.

273 pts · 163 comments

Study estimates 75% of NFL players who died over 15-year span had CTE (NBC News)

A new study estimates 75% of deceased NFL players over 15 years had CTE.

Thread focuses on cumulative head trauma starting in youth sports, not just NFL play.

262 pts · 234 comments

Nvidia RTX Spark encrypted firmware slows inference speed to 35% on Linux (MindStudio)

Linux on RTX Spark laptops throttles GPU to ~780 MHz when plugged in due to encrypted firmware.

Commenters ask whether encrypted firmware will become a paywalled Nvidia feature.

244 pts · 137 comments

Kitty 0.50 (kitty)

Kitty 0.50 adds window docking, session restore on logout, and workspace-like tabs.

Users compare Kitty to Ghostty, praising Kitty's native session support and Ghostty's macOS polish.

233 pts · 144 comments