There's lots of ways in which blockchains and LLMs are eerily similar, but there's one I've not seen mentioned: how every single use-case for both can be categorised into one of two buckets: "can be done better by something else" and "nobody actually wants to do that".