GPT4 much lazier than GPT 3.5

Talking about the most weird case. When I ask ChatGPT to rewrite code and fix some issues, I notice that the GPT-3.5 model can quickly rewrite a large amount of code without skipping anything. However, the GPT-4 model tends to make mistakes, skip important parts, and leave comments for me to address. This is quite frustrating. I tested this on multiple custom bots, on my own bots, I added tons of instructions for not skipping any important part and still GPT 3.5 does it much better.


GPT4 was (!) better than 3.5. A lot better. But it has become worse since autumn and is still detoriating. OpenAI apparently ran into a problem that they can’t fix anymore. I think we will have to wait for GPT 5 for this to become better. Which will probably take a long time.