The whole mess is a good example why benchmark-driven-development has negative consequences.

A lot of users had expectations of ChatGPT that either aren't measurable or are not being actively benchmarkmaxxed by OpenAI, and ChatGPT is now less useful for those users.

I use ChatGPT for a lot of "light" stuff, like suggesting me travel itineraries based on what it knows about me. I don't care about this version being 8.243% more precise, but I do miss the warmer tone of 4o.