Why, with the release of version o1-preview, did model 4o suddenly become an idiot? It completely forgets instructions given within the same conversation. You ask it to translate a text, and the idiot either answers questions or translates only half and responds to the rest. On top of that, it started making mistakes in basic messages and incorrectly interpreting commands from images. This never used to happen before, and 1o isn’t exactly brilliant either. When model 4o still worked properly, it was performing better than him.
1 Like
This might answer your question:
Maybe gone too cheap from all sorts of similar reports.
You can specify gpt-4o-2024-05-13
2 Likes
I’d like to see the conversation to be able to diagnose the problem. GPT-4o has worked with me in the most nebulous of contexts with total comprehension of my prompts (with the occasional realignment ofc) and has only gotten better. Without seeing, perhaps your issue is the phrasing of the prompt that was intended to translate, did you clearly define the objective of your prompt?, did you phrase them similarly (down to the precise wording) when prompting again from later in the conversation? These are all very important in guiding the LLM to your intentions.