Hacker News. Daily summary. Top 20 stories.
China's open-weights AI strategy is winning, outpacing America's closed, proprietary approach.
Two camps: China's open strategy is market capture or smart competition versus enterprise adoption remains US-centric.
A minimalist browser-based ATC game where players draw flight paths for planes.
Fond nostalgia for Flight Control and comparisons to Mini Metro dominate the thread.
Chinese AI models' open weights are commoditizing intelligence, shifting advantage to lower serving costs.
Split on whether distillation is illegitimate or just fair use revenge for internet scraping.
Fable LLM discovered a counterexample to the 85-year-old Jacobian conjecture.
Confirmation of the counterexample, with surprise it was not found earlier.
A hacker breached Romania's land registry, wiping databases and halting real estate transactions.
The thread compares Romania's attack to similar breaches in Slovakia, Poland, and other nations.
Kimi Work is a local AI desktop agent with browser automation and 24/7 scheduled tasks.
UI copying accusations and data sovereignty concerns dominate the thread.
Jelly UI is a web component library that adds soft-body physics effects to native form controls.
The main sentiment is that Jelly UI's physics effects are invisible by default for users with reduced motion enabled.
EU negotiations with the US may force biometric data sharing for visa-free travel.
Split: data export to US as unacceptable risk vs. EU governments already abusive.
Xiaomi scales robot foundation models via 100K hours of embodiment-free pre-training.
Mixed views: impressed by scaling and uncut video, but skepticism about reliability and practicality of laundry tasks.
The ACLU documents Flock Safety repeatedly lying to city councils, police, and the public about its ALPR system capabilities.
Commenters highlight Flock's pattern of lies and note citizen action can still influence ALPR policy.
AI models are finding counterexamples to major open conjectures in mathematics, with Lean verification.
Debate: are AI counterexamples trivial finds or genuine breakthroughs in automated mathematics?
A researcher used an LLM prompt to find a WordPress pre-auth SQLi leading to RCE.
Skeptics doubt the $500k payout and note WordPress's old code has many bugs.
OpenCode has severe security and usability flaws due to its design.
Agreement with critique; alternatives like Pi recommended; project has 3,690 open issues.
New open models Kimi K3 and Qwen 3.8 challenge Anthropic's position as a top-tier AI vendor.
Two camps: those who think brand loyalty and workflow integration protect Anthropic, and those who foresee open models undercutting it.
Nativ is an open source Mac app to run AI models locally on Apple Silicon.
Mixed reception: welcome for an open source Ollama competitor, but skepticism over calling local models 'frontier.'
Jellyfin's founder Joshua Boniface and two other core team members have resigned from the project.
Gratitude for founders' work, with camps comparing Jellyfin's stability issues to Plex's increasing hostility toward self-hosted users.
Five US tech giants amassed $1.65 trillion in hidden debts through opaque AI funding schemes.
Thread splits on whether SPV debt represents real risk or a technical accounting distinction.
Qwen-Image-3.0 generates complex layouts, authentic details, and multi-language text.
Closed weights and NSFW SEO keywords dominate the thread.
Article body wasn't reachable. HN discussion still summarized.
Hardware video decoding on Nvidia GPUs may draw more power than software decoding without a specific environment variable.
Article body wasn't reachable. HN discussion still summarized.
Comparisons draw parallels to observables, signals, Excel, and build systems; Jane Street's packaging praised.