funny that altman couldn't really the answer question about how a lab ought to prepare for incoming GPT6 seems pretty clear to me that the only way to really structurally benefit from model advances is to systematically make every step in your production process machine-legible

personally i'm not convinced that model progress is going to be this fast (or that available compute will keep up), but if you are really bullish you should probably start recording and storing video footage or screen recordings of everything too

definitely you should be laying the groundwork with your staff and coworkers, and thinking about how to introduce a lot of surveillance for process knowledge without it feeling like actual surveillance.