This is becoming a familiar story - users postulating “shrinkflation” in the quality of ChatGPT’s generated responses. I wonder, can we use another LLM to grade the responses in terms of insightfulness, thoroughness, accuracy, etc. so they can be measured objectively and monitored?
It stopped being useful to me a couple of months ago. My guess is that corporate figured out that they can dumb it down for the average person and rent out the better versions to the people with the deepest pockets. Most likely government/military.