2. We believe that a majority of this acceleration in capability improvement is likely now driven by AI being used in AI R&D. We aren’t sure of this, though, and did not find consensus about this among Anthropic’s staff. Non-AI factors likely played a large role in late 2025.
3. We believe there is a chance that capabilities progress could lead to even faster acceleration in the next 12 months, since we did not observe clear bottlenecks or strong diminishing returns. However, given our limited vantage point, we may have missed bottlenecking factors.
Since August, a team within METR has worked with internal Anthropic information to assess whether AI itself is accelerating AI development.
We are now publishing a report summarizing the results of this risk assessment.
If you ask a civil engineer why their bridge design is safe, they'd have a very clear and convincing answer, starting from saying that it meets the gov't design code, with a detour into why the design code is sufficient (or nearly sufficient). They can also show numerical analysis for safety, and that likely they or other people checked various different ways the bridge could fail, analysis with structural models, etc. An honest engineer (typical) would additionally have clear answers for the limitations for their assumptions and why their safety cases may fail.
Ditto for airplane engineers, who additionally have formal verification for increasingly large parts of the input space, demonstrations that catastrophic failure conditions are extraordinarily improbable, mandated incident reporting culture, etc. And these are ~typical engineering industries, almost all of whom actually take safety seriously: industries with critical infrastructure concerns and/or public fears, like nuclear power, typically take safety much more seriously, rather than less.
Software engineers (especially in consumer software companies) are the odd ones out here, mostly because historically software didn't affect the real world as much + failures are easier to roll back, and the laws reflect this prevailing (arguably outdated) notion. And AI "engineers" even more cowboyish than normal software engineers.
Next Saturday, Nov. 14, hundreds of us will be marching in Washington DC, San Francisco, New York City, Boston, Paris, and Montreal.
We're calling for an immediate international agreement to ban ASI.
Let's do this.
@RuhtraX99@wuyezi0 It is not your place to insist yue not care. Ed put a spotlight on the map, and whether by being "sarcastic" or not Ed invited people who love to hate into that convo. In replies to OG & quote tweets, people say rancid shit, not for no reason but bc Ed's attitude set the tone.
dressed up for the first time in years and sang a song from the Pirates of Penzance. it was a pretty good time! rare moment i feel like drinking to celebrate lol