I’m often reading tons of code blocks within the chat, only to find out there hasn’t been changed a single thing, even though stated otherwise. I have this issue with other models too but o1-mini does this the most.
Plus, o1-mini writes overly long explanations. Would be great if that could be optimized (without me having to explicitly prompt it every time)
o1-mini is effectively a research product - they released it, knowing it has some value, but nobody knows how to work with it properly yet - same with 01-preview. it is going to take a few more generations before people really get to grips with these new style models. I suspect one aspect of that will be routing to different models depending on the query.
For what’s worth, I have a lot the same kind of issue with codestral