Andrej Karpathy, OpenAI co-founder now at Anthropic, describes a widening gap: about 75% of people have barely used LLMs, while roughly 5,000 have access to internal frontier systems whose agent swarms do years of expert work. The public, he says, sees at best an odd chatbot.
The latest Import AI newsletter covers creativity and recursive self-improvement, open-weight models said to trail Chinese ones, and AI in space.
A study in Science reports that DeepMind's AlphaProof Nexus solved dozens of open problems, including nine Erdős problems, two of them open for 56 years. DeepMind had the work reviewed before publication, in contrast with OpenAI, which faces a boycott for posting unreviewed AI proofs on GitHub.
Meta has released two new models, Muse Wonder and Muse Wonder Full, which it says can compete with the industry's best. They follow Muse Spark, the first model from its revamped Superintelligence Labs, and the Muse AI agent.
TikTok chief executive Shou Chew expects AI to deliver major scientific breakthroughs within three years. He cautioned that growing capabilities will require stronger safeguards, especially as autonomous systems spread.
In an interview, Geoffrey Hinton said Donald Trump's fixation on the word superintelligence shows he knows nothing about AI. The pioneer also discussed whether AI could wipe out humanity and argued that a major challenge will be accepting that humans are not the only intelligent beings.