| ▲ | nradov 13 hours ago |
| Lol that's what they told us last year. And they'll say the same thing next year. It's obviously useful for certain tasks but the number of errors and hallucinations are still quite bad for anything large or complex or really novel. That part is improving very slowly and some fundamental theoretical breakthroughs will be needed. |
|
| ▲ | bpodgursky 13 hours ago | parent [-] |
| I don't care what they told you last year, I didn't say jack shit. I'm saying it now. Cope or pretend it's not happening, I don't care, this is the year. |
| |
| ▲ | honeycrispy 13 hours ago | parent [-] | | Have you actually had it do anything substantial and then tried to work with the code it produces afterwards? It may "work" but it's a horrific mess. Good luck bug fixing that. | | |
| ▲ | wahnfrieden 13 hours ago | parent | next [-] | | Yes, with GPT 5.2 Codex and 5.2 Pro specifically. It’s not a mess because of the context I provide and the guidance and reattempts I apply. It’s working great, the resulting code is good when I accept it, and I’m getting much more done than in the before times. | |
| ▲ | bpodgursky 13 hours ago | parent | prev [-] | | Yes, Opus 4.5 is not business as normal. If you haven't tried it, you really need to, to update your opinion. | | |
| ▲ | honeycrispy 3 hours ago | parent [-] | | I am talking about opus 4.5. It made a complete mess out of my project. | | |
| ▲ | wahnfrieden 2 hours ago | parent [-] | | That's what it does. Switch to Codex (not someone else's harness) with 5.2 Codex XHigh | | |
|
|
|
|
