“What they did in this report is that they had a lot of access to Anthropic data. They were provided access by the company to a bunch of internal data, researchers, et cetera. And they found that overall AI progress at Anthropic has accelerated to roughly twice the rate that we saw historically, specifically last year. And they estimate that about 80% of that speedup is directly attributable to AI. So essentially, AI being used to accelerate progress in AI.”

Yeah, so there have been big developments in the last week, week and a half in the AI safety space as well. And in particular, this report that was published by METER, which is an AI evaluation and research nonprofit organization on November 5th, related to AI RD. So, what they did in this report is that they had sort of a lot of access to Anthropic data. They were provided access by the company to a bunch of internal data researchers, et cetera. And they found that overall AI progress at Anthropic has accelerated to roughly twice the rate that we saw historically or specifically last year. And that they estimate that about 80% of that speedup is directly attributable to AI. So, essentially, AI being used to accelerate progress in AI. So, this is raising major concerns amongst a number of people who follow AI issues closely. And the reason for this is because there's long been this idea that there's this thing called recursive self-intelligence, recursive self-improvement. And what that means is that once AI learns how to substantially improve itself, you have this sort of runaway effect where the progress of AI then proceeds incredibly quickly, rapidly sort of leaves the ability for humans to oversee it. And then we enter situations where…