TE
TechEcho
Home24h TopNewestBestAskShowJobs
GitHubTwitter
Home

TechEcho

A tech news platform built with Next.js, providing global tech news and discussions.

GitHubTwitter

Home

HomeNewestBestAskShowJobs

Resources

HackerNews APIOriginal HackerNewsNext.js

© 2025 TechEcho. All rights reserved.

Reality Check on Deep Research by Ben Evans

1 pointsby alexdong3 months ago

1 comment

roosgit3 months ago
I&#x27;ve known about this issue since Lllama 1. Tried it with Llama 2 and Mistral when those models were released. LLMs are not databases.<p>The test I ran was to ask the LLM about an expired domain of a doctor (obstetrician). I no longer remember the exact domain, but it was similar to annasmithmd.com. One LLM would tell me it used to belong to a doctor named Megan Smith. Another got the name right, Anna Smith, but when I asked it what kind of a doctor, which specialty, it answered pediatrician.<p>So the LLM had no clue, but from the name of the domain it could infer (I guess that&#x27;s why they call it inference) that the &quot;md&quot; part was associated with doctors.<p>By the way, newer LLMs are very good at making domains more human readable by splitting them into words.