Also, the METR report that was one big AI analysis itself - quote from the research:
>Our subjective impressions are likely colored by analysis agents’ biases. Throughout this report, we describe a number of anecdotes of agent behavior that were compiled and summarized by analysis agents, where we were not able to read the transcript deeply enough to manually verify what occurred. We found that GPT-5.6 Sol would often uncritically adopt the perspective of the agent in the transcript it was reviewing
It’s the units that make it seem unintuitive. 12 days is roughly 0.032 years, and 17 minutes is likewise 0.000032 years. It’s relative to how we were taught to view smaller lengths of time. Of course you could argue measuring time by the rotation of heavenly bodies is more intuitive than counting on your fingers for the hours in a day and days in a lunar cycle.
Going from a million to a billion isn't some special relationship. It's just * 1000. It's the same as going 11000 or 101000.
I find it frustrating that people seem to think that someone going from 1M to 1B is somehow different to other numerical operations. It's not. It's 1000 times bigger, it's not a huge deal.
If you really wanna worry about all of these things you better not use any open source from now on. Or even private software as well, because you can't even audit and know what "vibe code" touched.
Being generous in my assumptions, you may have missed that my intent by "recent sprawl" indicates, "things are getting worse, at a rapid rate". At no point did I mean malware didn't exist 15 years ago.
It is getting worse, faster. That's objectively true.
The "independent" METR that is composed by... Checks notes... Previously employees from the top labs.
reply