|
📖 Read In Depth
|
DeepSeek V4–almost on the frontier, a fraction of the price
DeepSeek V4 is near-frontier performance at a fraction of the cost — Simon Willison's breakdown covers the model's capabilities and what the pricing gap means practically. Relevant both for understanding where the open/low-cost model frontier is and for thinking about when frontier pricing is actually justified.
hn/Best Stories
|
UPDATE: The method from the proof generated by GPT-5.4 Pro for Erdos Problem #1196 was successfully applied to other problems including another 60 year old Erdos conjecture.
GPT-5.4 Pro's proof method for a 60-year-old Erdős problem has now been successfully applied to additional unsolved problems, including another Erdős conjecture. If confirmed, this is a meaningful data point on frontier models doing genuine mathematical research — not just pattern-matching — and has implications for AI-assisted discovery.
reddit/r/singularity
|
Uber torches 2026 AI budget on Claude Code in four months
Uber burned through its entire 2026 AI tooling budget on Claude Code in four months. The story raises concrete questions about whether AI coding tools are delivering ROI at scale or just generating unchecked spend — and what it means when adoption outruns budgeting processes. Useful signal on how AI infrastructure costs are actually landing inside large engineering orgs.
hn/Best Stories
|
LinkedIn is scanning browser extensions
LinkedIn is enumerating installed browser extensions during site visits and building profiles from that data. The post explains the technical mechanism and how the data gets used. Worth knowing if you care about what data surfaces through your professional presence.
hn/Best Stories
|
City Learns Flock Accessed Cameras in Children's Gymnastics Room as a Sales Demo
Flock Safety accessed cameras in a children's gymnastics space as an unauthorized sales demo — and the city renewed the contract anyway. Connects directly to the broader Flock/license-plate-reader surveillance story and illustrates how surveillance infrastructure gets normalized despite documented misuse. The HN comment thread on LPR stalking by police officers adds more context.
hn/Best Stories
|
New research suggests people can communicate and practice skills while dreaming
New research suggests targeted memory reactivation during sleep can reinforce learned skills, and some subjects have demonstrated two-way communication while dreaming. The New Yorker piece digs into the science and the ethical/philosophical questions around deliberate sleep manipulation — genuinely interesting territory at the edge of neuroscience.
hn/Best Stories
|
AI uses less water than the public thinks
Pushback on the popular narrative that AI data centers are a major California water crisis driver — the actual numbers put AI water use well below agriculture or residential use. Relevant both as a corrective to hype and as a framework for thinking about how to evaluate infrastructure externality claims.
hn/Best Stories
|
How an oil refinery works
Construction Physics does its usual deep-dive treatment on how oil refineries actually work — fractional distillation, cracking, the economics of different crude grades. With oil prices surging from the Iran war context and energy infrastructure suddenly relevant again, this is good foundational knowledge delivered in the style Xinyu would appreciate.
hn/Best Stories
|
|
🎬 Check It Out
|
‘Moby Dick’ Review: Robert Wilson’s Last Masterpiece
Robert Wilson's adaptation of Moby Dick, with music by Anna Calvi, is playing a short run at the Brooklyn Academy of Music — reviewed as potentially his last masterpiece. Wilson's visual theater is genuinely unlike anything else; if you're in the Bay Area it's a drive, but worth knowing about if you travel to NY. Check BAM for dates.
nyt/Arts
|
|
âš¡ FYI
|
ARC-AGI-3 Update (GPT-5.5 High and Opus4.7)
ARC-AGI-3 benchmarks are in: GPT-5.5 High scores 0.43%, Opus 4.7 scores 0.18%. The new benchmark is significantly harder than ARC-AGI-2, and current frontier models are basically at floor. Useful calibration for where genuine abstract reasoning actually stands.
reddit/r/singularity
|
Maryland Is First to Ban A.I.-Driven Price Increases in Grocery Stores
Maryland became the first state to ban AI-driven personalized pricing in grocery stores — using consumer data to charge individuals more. First real regulatory precedent for algorithmic pricing discrimination; worth watching for how other states respond and what it means for ML-driven pricing systems.
nyt/Business
|
Sam Altman No Longer Believes In Universal Basic Income
Sam Altman has publicly walked back his earlier strong support for UBI, saying he no longer believes in it 'as much as he once did.' Notable shift given that OpenAI's original nonprofit mission was partly premised on distributing AI benefits broadly — and Altman was one of UBI's most prominent tech-world advocates.
reddit/r/singularity
|