Hi everyone in the forum,
I am developer in the bioinformatic field. The o4-min-high was uploaded yesterday. Although it is great to be able to communicate, it cannot remember the details of the code, and it always output fragments of code, which makes it difficult to track the modification. Also, it can input long prompt, but it always miss some of details in my prompt. I think the both o3 and o4 performance was a lot decreased compare with the ability before yesterday. Are there anyone have similar thoughts, and any adjustment I can do to my prompt?
I used to ask for “full code without changing the unrelated part” to get what I want, but now it does not understand what I say!
Thanks~
3 Likes
I have exactly the same experience unfortunately. o3-mini-high was really great. Best LLM for coding I’ve used. o4-mini-high is a very frustrating experience, because even when I give it exact function, and ask it how to do some changes, it will ALWAYS rewrite also unrelated pieces of code, often using very advanced syntax which trades readability for brevity.
It usually requires at least 2-3 additional prompts to convince it that I only want to see changes of my code related to the adjustment/improvement I asked for. It always starts with giving me back completely rewritten function with variable names changed, etc… and it requires lots of additional effort to make it give me my original code just with the localized improvement.
1 Like
Exactly. I have unsubscribed ChatGPT now and move to Claude gradually.