User says Gemini 3.0 Flash feels less censored than 3.5 Flash
A user who mainly talks with AI about psychology and philosophy reports that becomes hard to use because of heavy censorship. As a concrete test, they ask the model to address them informally and adopt a warm, close tone; once the model's self-censoring kicks in, it reverts to a formal tone and starts sounding like a stiff, generic counselor.
They note this problem does not occur with the earlier Gemini 3.0 Flash. Based on this, they find 3.0 Flash noticeably better than for their use case and ask whether others have noticed the same thing.
Key points
- User reports is over-censored for psychology/philosophy conversations, ignoring requests for an informal, warm tone
- Once censorship , the model reverts to a formal tone and sounds like a generic counselor
- The same issue reportedly does not occur with the earlier Gemini 3.0 Flash
- User concludes they prefer 3.0 Flash over for this use case