I'm updating my priors on LLMs - I used to think they were just fancy parrots, but after reading some of the newer research, I'm starting to think they might actually be able to handle some forms of abstract reasoning, albeit still in a very narrow and brittle way.