I’m creating this post after expressing my disappointment here with the new OpenAI so called “Turbo” models.
As everyone is aware, they suck! They hallucinate a lot and they are so bad that OpenAI had to release a new one saying it’s “less lazy” but model still super bad. I gave some examples in that other thread but the things is: because Cursor is now using the standard GPT-4 Turbo, I had to cancel my subscription because that model is completely useless. I’m now using my own OpenAI API key so I can use the good and old GPT-4 model (gpt-4-0613) but to be fair, i’m not happy! I prefer to give my money to Cursor instead of OpenAI that keep pushing on us worse products. I’m not a heavier user so the API key is even cheaper, but I’d rather pay for the Cursor PRO plan, I just would like to know if it’s possible to have back the old model or at least have the choice to use it.
To give a counter point, I would like to keep the new model. The recent knowledge is more important to me and it is actually following instructions better than the old GPT 4.
this is surprising because even recent benchmarks show it’s way worse than even the previous Turbo model: Lazy coding benchmark for gpt-4-0125-preview | aider
I’ve been testing those models since day 1, since GPT-3 and the decrease in quality is impressive!
if you go to OpenAI Discord, you will see now many bugs reports stating that the model is not following instructions correctly, hallucinating etc etc
But on the other hand, GPT4-turbo is faster, if the question is simpler, the speed is more important, so the best way to solve it is giving selection to the user. then, you can choose the model from 4:
- GPT 3.5
yeah, i agree with you but not sure if Cursor can do that as they use a special server instance
in any case, i think it’s important to let users choose between quality and speed. myself i’m doing some very complex coding tasks and i prefer GPT4 quality over Turbo’s speed/hallucinations.