OpenAI should open source GPT3.5 turbo

I partially agree. I agree on the part that gpt3.5 is surprising only in chess (and what have you if there are more of such surprising benchmark). But having models that apparently can solve many difficult problems without that much scaffolding; models that apparently can replace most of the white collar workers soon that then got pummeled by gpt3.5 with some fine tuning is interesting. I mean, I know, ad hoc chess engines could easily defeat them all; but within the realm of LLM and the fact that the fine tuning of gpt3.5 IIRC wasn't even that massive, it is surprising to me that very large models or powerful reasoning models get defeated so easily. This beside GPT4.5 that could be simply so massive that it includes most gpt3.5 fine tuning anyway. Would you expect a SOTA reasoning model to play decently at chess? I don't mean that well, but like someone that plays in a chess club for a year? (thus not someone totally new, they know the rules and some intermediate concepts, but they aren't that strong) I would, given the claims many do on SOTA models. Well they can't (so far). Gpt3.5 is apparently still very valid in this case.