RE: https://mathstodon.xyz/@gregeganSF/117121709868547118

I think these "AI overviews" are a big reason why so many people underestimate the capabilities of modern LLMs. It's a worst-case scenario for errors and hallucinations: one-shot, small models, tight latency constraints, no oracle to validate output against. It's a total waste of time and energy. And if you're an AI refusenik, you never see them under more favourable conditions!

[I am not saying you should *like* LLMs, but I'm begging you to *take them seriously*.]