In my experience Pangram has been, if not perfect, then at least good enough for my purposes, and I'm glad that at least the big labs seem incentivised to have text outputs to be clearly identified as written by AI.

Probably doesn't help with Twitter though because individual engagement-baiting messages will probably be too short, but... makes me wonder if we're going to move to something more like an answering-machine internet where we're using LLMs as intermediaries in any and all lower-trust environments (which seems like it would be anywhere strangers can congregate?) or if it's going to be more like proof-of-humanity or just proof-of-attention.