I’ll admit, I’m more of a layman when it comes to Gemini than other users and didn’t really notice the ways that transparent reasoning benefitted output quality, but now that it’s gone I undoubtedly took notice of the diminished quality. To mirror what I see others commenting here, I would truly appreciate if you could reverse this aspect of the update.
Then again, if what someone speculated is true, that this is a move designed to stifle competition then perhaps it shouldn’t surprise me. Disappointing, but not surprising
Edit: Oops, I posted this in API feedback mistakenly.To qualify, I’m a google ai studio user
2 Likes
Welcome to the forum.
Quality CoT traces are valuable, in that they can be used to upgrade smaller models. Replacing the quality CoT with an abridged, unspecific, opaque commentary devalues the CoT trace for that use (=allows Google to maintain a competitive advantage) and at the same time it devalues the model response for human users. The speculation is, the first effect is what Google wanted and the second is an unfortunate side effect.
I recently read in the technical report on Phi-4 mini reasoning that Microsoft used CoT traces from DeepSeek R1 to train that model. Quality CoT traces are still available.
Hope that helps.