James Gleick on Nostr: npub1w6d0a…cytvx npub1j5a6t…hp848 My brief statement was admittedly a kind of ...

npub1w6d0ahgsr65xeeucfadr99xh65fq7d7hyrs3vj49s2vvcpsu9p9q0cytvx (npub1w6d…ytvx) npub1j5a6tz6a2ttelsdzdl3lnpqq2q8hyfluz0c7kpgntt548q7f5snsrhp848 (npub1j5a…p848) My brief statement was admittedly a kind of shorthand, oversimplified. Of course they can answer questions that haven’t been asked before, as long as the questions are LIKE familiar questions and as long as “answer” means “produce plausible answers that often happen to be correct.”

We know computers can be programmed to reason in various ways. The LLMs’ special skill is generating text probabilistically based on pre-existing text.