I have noticed over the past 15 days a significant regression in the usage of O1-Preview. Unlike before, it is no longer able to generate some codes in their entirety. Instead, it provides scattered lines of code to be replaced, making it highly fragmented, and generates some mistakes deleting completly part of code. It gets stuck when searching for solutions. For me, both O1 models have become now irrelevant even if at the begining were greats. Note : I would prefer to receive 50 accurate responses, even if limited to that number, than to get 100 flawed responses. It has been impossible to achieve a proper analysis for the past 15 days. From my point of view, it is a major regression in the model.
3 Likes
Fully agree.
While I did find some specific use cases that the o1 models were a benefit, I did find that for a majority of my needs the o1 model was overkill. I quickly learned to not use the o1 model unless a lower model failed horribly, meaning I did not consider any change to the prompt to result in a success and that COT (Chain Of Thought) or thinking was needed.
It would be nice to know if certain users are just using o1 models for everything because they can and to make it more available they are degrading it to improve the volume of use. 
2 Likes