Comment by Grimblewald

I maintian a log of tasks, prompts, related information etc. So i can repeat past workflows verbatim, and I can qualitatively say each model beyond 4.5 has been a regression, and it would not surprise me 4.8 continues the trend. Each iteration has failed at more tasks previously completed succesfully. Right now it flat out refuses to answer many benign chemistry questions, or leans into shilling to hard and ignores non industry funded studies on certain topics. I'm transitioning to deepseek as a reuslt. Cheaper by far and at this stage not strictly speaking less capable.