@jxnlco it’s annoying if you want a quick answer between two longer tasks but you set reasoning to xhigh because it’s just too much friction to change it. something like “\low [prompt]” would be nice (ie only set reason for one prompt.
@OnodaCapital@zephyr_z9 demis seems quite good at reading people. this also reminds me of the story before pitching deepmind to peter thiel he talked about chess at a party with him.
@SemiAnalysis_ there definitely is “pretrainitis”. still pretraining make sense if you do something not language related. for example we at stripe have radar or pretraining for social media recommender engines, etc