The growing gap between what AI companies develop internally and what they release publicly is a significant safety and governance problem that requires mechanisms like maximum training compute ratios relative to released models.

8:03should expect chips to be priced going forward based on the value of running models on them and how the fact that the VA the models are getting so much better means that the value of running old chips is increasing. And thus we have, you know, rising prices for chips that when they were first bought and deployed, you know, were supposed to be basically end of life or, you know, kind

8:25of end of their depreciation at this point. So that was some pretty sharp analysis uh that has, you know, definitely served me well to keep in mind over the last handful of months. And yeah, I mean, he's pretty plugged in. So I think the um and we've obviously got some, you know, uh corroboration of additional models directly in the report. So I generally am kind of worried about this growing

8:54gap between what the companies have internally and what the rest of us get to see. Um you know this has been like so many things in AI we've gone through these phases in fast succession and there was a time around the kind of 01 to you know 03 to04 mini to um you know whatever uh I guess to GPT5 excuse me when there really wasn't it seemed much of a gap between the

9:30you know kind of finish of training and the launch of the models and I I remember a Rune tweet where he said, "You guys have no good no idea how good you have you know there there's like barely a gap at all between what we're using internally and what you guys are getting uh in the product surfaces and that now seems to have changed and and this is I think another kind of win

9:51for the you know cocatello AI safety school of uh forecasting because this is right chapter and verse pretty much for AI 2027 the gap is indeed growing and all the people that have said for a long time that private, you know, internal only deployments are going to be a major source of risk and uncertainty and, you know, who knows what. Um, you know, major base points for those folks, you know, based on what

10:28we've seen this summer, right? So um I am interested in some ways to to try to govern this. You know everybody pays lip service including anthropic. You know to their credit I think I think it's sincerely felt uh that they don't want to overly concentrate power you know and that they're quick to say oh the you know reports that we intend to be the only private company where it's just

10:54governments and anthropic left you know those are much exaggerated. Uh, and I'm sure that that is exaggerated relative to what they, you know, expect or or really aspire to. Um, and yet, you know, this gap is widening and we are indeed seeing kind of the most flagrant uh safety violations coming out of these previously undisclosed models, right? I mean nobody even knew that this next generation open AI model existed really

11:29I don't think at the time I mean obviously people you know assume that there's more models and training but I don't I don't think that they had even said anything really about it as of the time that Hugging Face went forward with their report about being hacked. So that is starting to be a a really weird world and I'm starting to think a lot about just what kind of governance