Follow the latest coverage, related explainers and connected technology stories.
In its updated disclosures for its anticipated IPO, Anthropic hinted at approaching the stage of reverse AI self-improvement, raising questions about development speed and safety controls.
Dario Amodei, CEO of Anthropic, has proposed a three-stage plan to reduce the risks of AI development, including involving independent evaluators and broader coordination between companies and governments. The proposal is based on concerns about the accelerating ability of models to develop successor systems and the misuse of intelligent agents in cyberattacks or biological attacks.
OpenAI said it had achieved its stated goal of reaching the level of an “AI research intern” capable of carrying out research tasks that would take human researchers days, while human oversight of key decisions remains in place. Company data shows that the use of coding and research agents has risen sharply, but longer and more complex tasks still require human intervention.