johnsmith1840

30 ms·

1. 331.▲
   by johnsmith1840 1y ago

   I'm positive they are pushing users to cheaper models due to cost. o1-pro is now in a sub menu for pro users and labled legacy. The big inference methods must be stupidly expensive. o1-preview was and possibly still is the most powerfu
2. 332.▲
   by johnsmith1840 1y ago

   Drop in replacement files per update should be done on the heavy test time compute methods. o1-pro, o1-preview can generate updated full file responses into the 1k LOC range. It's something about their internal verification methods tha
3. 333.▲
   by johnsmith1840 1y ago

   I seem to be alone in this but the only methods truly good at coding are slow heavy test time compute models. o1-pro and o1-preview are the only models I've ever used that can reliably update and work with 1000 LOC without error. I don
4. 334.▲
   by johnsmith1840 1y ago

   It's a new skill that takes time to learn. When I started on gpt3.5 it took me easily 6 months of daily use before I was making real progress with it. I regularly generate and run in the 600-1000LOC range. Not sure you would call it &q