notdevin

GPT o1, so far, sucks for code as far as I can tell. It’s insanely verbose and goes off the rails just as fast, but now for longer, than previous models. It has been awesome for conceptual brainstorming and system architecture. I much prefer Claude 3.5 or gpt 4, I’ll take 4o for boiler plate specifically

Anyone else writing code having better experiences with o1 preview?

If I offended you, I was probably talking about you. If I’m wrong, update/inform/regale me with why, or don’t. 

https://backpack.network
https://anagency.xyz

giving more direct instructions, breaking down problems into smaller steps, or explicitly asking for minimal output can help reduce verbosity


Geophysicist || web3 || UI/UX || art || FCB ❤️ 

You're spot on. By trying to make these models more accessible to everyone, they risk watering them down to the point where they excel at nothing. It feels like we're losing the fine tuned precision that made earlier models powerful for specific tasks, like coding, in exchange for broader usability.

Sure, but now we’ve made it take more work to do the same thing. I think these guys are really missing the marks by focusing on making them usable by everyone, the natural outcome is that they’re going to drive the models to perform absolute mid at everything. Which apparently is the not so subtle goal of all things tech