Hacker News. Daily summary. Top 20 stories.
Mathematician Terence Tao uses ChatGPT to rederive a known Jacobian Conjecture counterexample.
Impressed by AI as a mathematical collaborator, but see limited practical impact from this specific result.
Bento packs an entire presentation editor with collaboration into a single HTML file.
Single-file approach praised; technical details on encryption, CRDT, and auto-update explained.
John C. Dvorak, columnist and No Agenda Show host, has died at 74.
Mourning a PC journalism legend, with a side debate over his correct age (74 vs. 80).
A satirical site pitches an AI CEO replacement for $4,699, mocking executive pay and layoffs.
Debate: Can AI replace a CEO's networking and PR role, or is it just a tool for meritocracy?
Analysis shows AI labs likely not gaming the "pelican on a bicycle" SVG benchmark.
Consensus: labs aren't gaming the benchmark, though some question the LLM judge methodology's reliability.
Article body wasn't reachable. HN discussion still summarized.
Tokenization speed matters for pretraining, TTFT, and latency-critical decisions.
Reddit requires login for old.reddit.com, citing safety through reduced scraping of plain HTML.
Blame falls on Reddit's AI licensing deals, not safety, as the real motive for the login wall.
Passkeys confuse consumers because engineers prioritized security over user understanding.
Two camps: passkeys as phishing-resistant improvement vs. opaque, big-tech lock-in with poor portability.
SIMD follows a simple five-step pattern accessible to any programmer.
Two camps: trust the compiler's auto-vectorizer or write SIMD by hand for predictable speedups.
A startup engineer's practical guide to moving beyond basic Postgres knowledge.
Thread urges adding backups and monitoring to the guide; pgBackRest recommended by many.
A researcher built a searchable index of 6,500 prize-winning non-fiction books as an antidote to AI-generated content.
Commenters debated whether book prizes are meaningful signals and whether using AI to build a quality-focused tool undermines the project.
A take-home interview assignment contained Git hooks that deployed malware to steal cryptocurrency.
Two camps: those urging phishing-style validation of recruiters, and those noting official channels are often unreachable.
The author argues that AI-generated output lacks the personal fulfillment of making something oneself.
Two camps: LLMs destroy the joy of making vs. enable projects that would otherwise languish.
A Filipino-American diner is disappointed by AI-generated menu images at a local restaurant.
AI menus seen as the new signifier of low effort, like chalk signs or MS Word menus.
Codeberg votes to ban cryptocurrency-related projects from its platform.
Split between critics decrying the vague policy and supporters of anti-crypto stance.
Late.sh is an SSH-based multiplayer terminal with chat, games, and an artboard.
Dominant sentiment is enthusiastic praise for a nostalgic, feature-rich SSH-based BBS.
A evidence review asks whether creatine supplementation improves cognitive function, concluding maybe a little.
Split between those skeptical of nootropic claims and those accepting the cautious conclusion
A user returns to Kagi after trying alternatives, citing superior search quality.
Strong user loyalty to Kagi, with some reporting rising AI usage costs and debating LLM vs. search.
US official alleges Moonshot AI distilled Anthropic's Fable model for its K3.
Commenters broadly see hypocrisy in US AI labs complaining about distillation after training on the open internet.
Silicon Valley's AI obsession is driven by fear of falling behind, not rebellion.
Split between those agreeing with the critique and those defending AI as a tool.