Comment by fwipsy

I assumed the "thinking" output from Gemini was the result of a smaller model summarizing because it contains no actual reasoning. Perhaps they did this to prevent competitors training off it?