Stanford's 2026 AI Index: frontier models fail one in three attempts, lab transparency is declining, and benchmarks are ...

A model can be 95% accurate and still be a disaster if it’s too slow or drifts. Don't just watch the model — watch the ...

Superconductivity—the ability of some materials to conduct electricity with no energy loss—holds immense promise for new technologies from lossless power grids to advanced quantum devices. A ...

At Sundance last week, Joseph Gordon-Levitt discussed one of the biggest problems facing indie filmmakers who want to use AI tools. The people whose work trains these models aren't getting paid. And ...

Clinical/professional psychologists are engaged in a debate about what constitutes the best identity and training model for the field. Below re-post a blog I did on the APA Division 24 website for ...

Relying on one giant AI model for everything is a trap; it’s too expensive and slow for simple tasks and too risky for the hard stuff when things go wrong. Whenever I see a new agent project kick off, ...

An ambitious effort to create a neurophysiological paradigm to explain near-death experiences has failed to capture many fascinating and often perplexing aspects of people’s brushes with death, top ...