@six

There is an interesting phenomenon where you know the models are going to get better so the calculus on doing certain work earlier rather than later actually doesn't make sense. for example at a micro level, there is no work that rationally makes sense to do tonight when you can do it tomorrow when fable returns and finish it faster and at higher quality this days-length phenomenon may also exist at a years-length phenomenon e.g. some things do not make sense to work on today when you know that GPT-8 will be able to do them much better and faster in 2027

1 reply

0 recast

8 reactions