Why are LLMs producing this super tense text more often now? Is it because they are being optimized to use fewer tokens?

I think they're overcorrecting for being too long-winded in previous generations (and still, in some cases). I guess this is a hard balance to get right.

I'm pretty sure the other answers are wrong and it's a side effect of RL (see thinking machines post about inkling training). It's also exacerbated in fable and sol--I think it's token efficiency effect--bc it's about to reason with fewer tokens the density of the token information goes up.

I figure they are being optimized to write code/functions and not prose so all text is getting more code like.

ChatGPT is still plenty verbose by default IMX.