Terence Tao on Nostr: nprofile1q…8g23r Currently, I rate the best-case outcome of LLM queries to be ...

nprofile1qy2hwumn8ghj7un9d3shjtnddaehgu3wwp6kyqpqm3fp0m7fjdsd58dhs39sfvz4d45zsxtp4c362avjw4a869kr453qx8g23r (nprofile…g23r) Currently, I rate the best-case outcome of LLM queries to be superior to web search due to the speed and appropriateness of the response; it requires less mental effort to create a prompt than to create a search query, scan through a half-dozen hits to determine which ones are the most promising, and then mentally translate the answer in the best hit into the notation and context one actually wants for one's application. However, in the worst case, these queries require more mental bandwidth than a traditional search. So right now the two options are roughly comparable in utility. But I expect the quality of LLM queries to continue to improve, whereas for web search it seems that the main potential sources for further possible improvement also come from AI integration.