> Its also fantastic for rubberducking
Given that the gold standard for rubberducking neither knows nor communicates anything, this is an easy win.
True, but in the past couple months, I've probably typed out 10+ truly complicated issues to ChatGPT and arrived at a solution before I ever pressed enter... just because I took the time to explain them in proper prose.
So far, the yi model is actually not bad, and by two weeks from now two new models will be out that are better. So it will continue until Altman figures out which politicians to bribe the hardest.