AI models, services and robotics
OpenAI announced that 10,000 agents solved the Navier-Stokes problem. It and a mathematician who released a proof in the same direction a day earlier are telling different stories about authorship and data access.
AlphaGenome Atlas precomputes every single-base change in the genome as a 1-petabyte dataset, now open to researchers worldwide
An NYU mathematician says he used AI help to prove blow-up in the Euler equations, and has published a statement detailing a call with OpenAI just before he announced it. He wrote that OpenAI proposed dropping a co-author who works at Anthropic.
Here is what it takes to qualify for the Google AI Plus student offer, and what to watch for with automatic billing
Insilico Medicine gave patients its AI-designed pulmonary fibrosis drug rentosertib for 12 weeks, and their biological age dropped on all six aging clocks tested — by more than three years in some cases. The drug was made by AI, and its effects were measured by AI. So how much should we trust these numbers?
Four days after receiving a "critical" cyber risk rating, OpenAI showed the model solving hacker puzzles using parallel agents
A Hugging Face-linked security incident halted training, but reliance on agents kept climbing
Anthropic's Claude wrote 13 million lines of Lean code over 11 days to formalize one of mathematics' toughest problems
Two or three units connect directly with QSFP cables, but four require a switch. NVIDIA's official documentation now includes an approved cable list and an automatic setup tool
A gold-colored taxi with no steering wheel and no pedals has arrived, but 70% of Americans still find the idea uncomfortable.
In a new test that hands AI a year of running an online store, GPT-5.6 Sol took first place while an older Alibaba model went bankrupt four times
OpenAI's ARC-AGI-3 results sheet lists 62.7% and 99.9% side by side for the same model. Tracing why one model ends up with two scores reveals who actually deserves credit for this result.
A model trained on a European cohort actually lost accuracy as the target population's sample size grew
OpenAI begins rolling out Astra, a model that handles computer tasks on your behalf, starting today
One grey-box level produced three finished prototypes with different themes, and first-try success rates jumped sharply
A managed Code Review service and a GitHub Action — either way, approval stays with humans
Anthropic released a tutorial showing how to bundle skills, hooks, and MCP servers into a single-line install
Anthropic's fifth Claude Academy lesson shows how to set up hook events that enforce CLAUDE.md rules
Ten days after teaming up with France's Mistral, Saudi Arabia's state AI company has now built on a Chinese open model
Anthropic's Claude Academy lecture covers how to manage long coding sessions, from plan mode to worktrees.
Across 7.1 million verification images, a training-free discriminator matched or beat existing detectors
Built with HHMI Janelia, the map covers 166,000 neurons and 125 million synapses, making it the largest connectome ever produced
The UK's first autonomous ride-hailing service launched in London on September 3, local time. A British startup's AI reads one of the world's most complex road networks without lidar or HD maps.
Built by three founders from Instagram, Uber, and Google, Fambot reads Gmail and Google Calendar to organize your family's schedule for the next day
