@jxnlco You already did this in chatgpt. Auto switch there was pretty good back in the days when chatgpt was good, and you could turn it off if you wanted, but mostly in codex you'd want to choose the model.
@thsottiaux You should loop in another LLM for a second opinion sometimes, an awful lot of bugs get out into production. Or maybe humans, I hear humans can be useful too. My cat told me to say this
Further evidence of my cabin fever:
Discussing the ins and outs of moose/abuse/loose vs lose/abuse with codex
Discussing bouncy balls with gemini
Noticing dinosaurs in my search misses so off to find dinosaur photos
I've had Covid for 64 days. I know this because I have a 64 day 1.2B Codex streak. 200ish commits, 3 new projects (one is an accidental SaaS), I've discovered the LLMs can make me animations, Claude appears to suck, Gemini is awesome, and I desperately need to touch grass
Codex agrees that I need dinosaurs so I'm going to get dinosaurs.
Its actually my user search data, but I just had to get my dose of sycophancy in and Codex was right there
@parth_develops@thsottiaux I've been using gemini to do the full stub of a new application, which it tends to do with working "fake" components or some working-ish real ones. Then get codex to make the fake stuff real without it undoing all the visuals
@thsottiaux Its a hideously slow thing and eats all my cpu. It also behaves quite different to CLI. I use CLI most and would be very upset if you mess up CLI. The app seems far worse at following instructions than CLI