I have limited interest in reasoning models. I understand that LLMs get stuff wrong a lot, but at least give me that wrong answer now, so I can verify now. I don't want to wait minutes or longer for an answer and then still to go verify it. Of course, reasoning models are supposed to have lower error rates, but not so low that I no longer need to verify.
The way I see it, normal "AI assistants" are a character being acted out from a mad-libs script, and these "reasoning" models are the same but the script is film noir, so the Bot is also quietly muttering to itself.
I can see how it would be less-likely to wander, but it doesn't seem like a qualitative leap in the fictional thinking going on.
Grok is a reasoning model but is still very fast.
so it cost 100 times more to run and chatgpt was losing money. what kind of money are these other companies burning through?
I think they believe the moat, if any, is capital investment. So they found something new that works "better", but requires much more capital.