OpenAI's math model releases 722 papers, straining expert review
Each result takes hours of compute but days of human verification, turning the mass release into a debate over reviewer capacity in AI research.
Tech stories and the sources behind them.
268 stories
Newest first
Friday
Each result takes hours of compute but days of human verification, turning the mass release into a debate over reviewer capacity in AI research.
The publishers are seeking over $250 million, and the case could set a defining precedent for how news archives are reused to train AI.
Carriers saw their stocks slide on the pending deal, which would extend satellite-to-phone coverage to places cell towers reach poorly.
Only one model finished the course, offering a rare look at how AI handles physical-world delays, safety refusals, and unexpected capabilities.
Thursday
Its models surfaced far more open-source vulnerabilities than humans can review, highlighting both the speed and the limits of automated security reports.
The dispute over whether the researchers were punished for raising concerns about unmonitorable model reasoning tests how the lab handles internal dissent.
The top score on real physics experiments included a robot arm recovering from a missed grasp, a behavior that never appeared in its training demos
Attribution of last week's breaches to a single operator using open-source hacking tools and several frontier models resets assumptions about how much damage one person with AI can
The relaunch drew six-figure attention within hours, a sign of how closely builders are watching AI coding tools.
Demis Hassabis's AI drug-discovery spinout from Google DeepMind could command as much as $50 billion, per Bloomberg.
The partnership puts a major stablecoin inside a mainstream consumer phone wallet.
The day's most-read take argues shoppers will want agents that buy across the whole web, not one Amazon supplies, handing startups a rare opening.
Showing 85 to 96 of 268 stories