Technology news and discussion on TalkFuse.
METR finds Anthropic’s AI is accelerating the company’s research progress
An investigation by AI safety research group METR found that Anthropic’s AI models are helping the company develop more capable AI at a sharply faster pace. The findings suggest AI is already accelerating its own development, raising concerns about how quickly progress could continue and how much of it may eventually happen without human direction.
METR estimates Anthropic achieved about two years of historical progress in one year, with most of the acceleration attributed to AI rather than increased computing power. Employees reported that the company’s most capable internal model made them about 12 times as productive, roughly double their reported gain a few months earlier. The report says much of the work still involves AI assisting researchers, but notes signs of more autonomous contributions ahead; it also says Anthropic can redact parts of the report, limiting public insight.
In short
- AI is helping Anthropic advance its models about twice as quickly.
- Staff reported a twelvefold productivity gain with an internal model.
- METR warns autonomous AI contributions could grow.
Steelman: both sides, in good faith
The strongest fair case each way — we don't pick a winner.
The case for
AI assistance could help researchers identify promising ideas, analyse results and improve systems faster, potentially accelerating useful applications and safety research alike. If people remain responsible for setting goals and reviewing work, greater productivity may be manageable, while transparent measurement can help organisations prepare for further changes.
The case against
When AI helps build more capable AI, progress may compound faster than people and oversight systems can adapt, increasing the chance that important decisions or risks are overlooked. The reported productivity gains and signs of more autonomous contributions make it reasonable to question whether human direction will remain effective, especially when redactions limit independent scrutiny.
Explained simply
Anthropic is a company that develops artificial intelligence (AI) systems, including models designed to help with research and coding. METR is an independent research group that studies how capable AI systems are and how they may affect the pace of technological progress.
AI can contribute to its own development when researchers use it to write code, analyse results or carry out experiments that help build more capable models. This is different from AI developing itself entirely without people: researchers still direct much of the work, though the balance could change as systems become more capable.
The pace matters because faster development could bring benefits as well as challenges, while making it harder to assess and manage changes in AI capabilities. Public understanding also depends on what companies disclose about their internal research and tools.
AI Business Companies Research Science Technology
Read the full article at the source →
Originally published by The Verge as “Why a new report on Anthropic automating AI development is alarming researchers”.