New Model GPT-4-o e GPT-4 Turbo are returning weird results

I have an application running on GPT-4-0125 which was returning good results for extraction information ina PDF. However, Since GPT-4-Turbo e GPT-4-O cannot reproduce de some results and sometimes delivery weird results. Some fields which I use in the prompt to analyze some situations is not working in currrent versions. Can I open a support ticket to solve this ?

1 Like

You can press help in the account platform, and send a message through the choice path API → feedback. It will likely be answered by a contractor and not go any higher.

OpenAI damaging the quality of existing models has been an ongoing modus operandi for a year.

1 Like

Perhaps you meant GPT-4-Turbo-0125 instead of GPT-4-0125?

If so, could the issue be due to using a different variant of GPT-4-Turbo, which is why the results are inconsistent?

It seems to me on my side things have improved, I was also surprised when I was getting some off-track responses that were not spot on like previously, i am happy now. Still getting to see the difference between GPTPlus4 and GPT-4o


When using ChatGPT, it has been noted that a “4o” label momentarily flashes next to the star icon, indicating which model is being used. Previously, this “4o” label appeared only for text generated by GPT-4o as a cautionary note, but now it seems to appear for both “4” and “4o”.

This appears to be encouraging users to switch between the two models and compare the differences. It gives the impression that they want users to experience firsthand that the gap between GPT-4o and ChatGPT-4 is narrowing, or even that GPT-4o might be superior in some cases.

Following this strategy, I’ve been generating text with ChatGPT-4 and then switching to GPT-4o to regenerate the text and compare the outputs of the two models.

As a result, I am noticing that GPT-4o actively uses the browsing feature and the memory feature more frequently. This doesn’t necessarily mean the model is superior; it could simply be adjusted to behave that way. Still, I feel that the performance gap between the two models is closing.

Additionally, once multimodality becomes available, the advantage of GPT-4o might become even more pronounced.

Happy cake day!

1 Like

Well i just created a Personal Assistant GPT, and i asked it to summarise my day’s activities since this morning, instead it went on to write a business plan for one of my institutions

1 Like

imagine this response to my short specific promt

1 Like

That’s a completely useless summary unless you’re the head of the institution🤣

1 Like

GPT-4o seemed to pretty much nail my day…

This image depicts a screenshot of a text conversation where a user asks for a summary of their day and receives a sarcastic and negative response from an assistant characterized as an "angry AI." (Captioned by AI)

1 Like

It’s funny, but I don’t know what to say :rofl:

Well, if GPT-4-o isn’t fitting your needs, you can always use Turbo or the original GPT-4. Just use whatever works best for you.