so the last days i tried a lot to see how far i can come with the preview mini and full model.
So here is my feedback to the mini model and what for issues needet to be adressed
Long Codes and Text result mostly in the warning that it is against the rules to request such things. This is really annoying
The mini model is ignoring most of the time all requests and commands and doing his own thing ignoring mostly complete the user wishes, like display me only the one file.
Coding is hard, the Model is not focussed of one thing by time, it starts to overthink all scripts and tries to make changes to them all also if you told them not to do so and stay focused. Resulting following it in mostlye total destroyed code every time.
The Mini model always tries to suggest composer for php
The Mini model often losses memorys about the request and thinks about old things and code as it would be a new issue completely ignoring the input (like some kind of delay) (but already solved)
The mini model has a really bad memory of context
A uploader of text and code files is really needed
The Mini model doesn’t care about the general user settings how he should behave to the user
The mini model has often Time-outs
The Model is “to fast”, it is so fast that you can*t see often what he is doing as dev, a bit better output and a bit slower would be good.
I agree with your points on several aspects of O1-Mini. The model often tends to ignore commands, even for simple tasks like displaying a single file. This failure to follow instructions systematically leads to unwanted modifications, making the code unusable.
Another important point to highlight is that even for programming tasks, O1-Mini does not execute these actions correctly. Text formatting errors or forgotten variable allocations are frequent, which is unacceptable for a model supposed to excel in this area.
Regarding the loss of context memory, I completely agree. The model regularly forgets previous instructions, forcing me to reopen several sessions as it completely lost track of the conversation. Moreover, it follows a predefined response template without genuinely considering the initial question, further complicating the interaction.
Another issue I frequently face is technical errors in the chat. I often receive error messages that force me to regenerate the response. This is not acceptable, especially since the number of regenerations is limited, and these errors seem directly related to the model itself.
As for speed, even though some responses are generated quickly, this does not compensate for the many errors that need fixing. The time saved by its speed is canceled out by the necessary corrections.
In conclusion, in its current state, O1-Mini is of no use to a developer. Its frequent errors, poor handling of context, and poor performance in programming tasks make it unusable. On the other hand, O1-Preview proves to be much more suitable and reliable for complex tasks.