Interactive atlas of research papers

  • Thread starter Thread starter Borg
  • Start date Start date
Join the discussion
Registration is free. Ask a follow-up in this thread, or start your own.
3 replies · 626 views
Science Advisor
Gold Member
Messages
2,394
Reaction score
5,225
Reply
  • Like
  • Care
Likes   Reactions: fluidistic, Greg Bernhardt and FactChecker
Physics news on Phys.org
Greg Bernhardt said:
Where is the toggle to filter out papers with hallucinated citations?
Last year I asked an AI for a list of papers on a niche topic. I identified ten papers that I wanted to take a look at. Each of the papers was a hallucination. Knowing that the papers were hallucinations, it was still difficult to get the AI to admit that they were hallucinations. It would generate hallucinated table of contents for specific issues of journals.
 
Reply
  • Wow
Likes   Reactions: BillTre, FactChecker and Greg Bernhardt
Frabjous said:
Each of the papers was a hallucination.
People really need to treat LLMs like they have dementia that slowly goes away over time. What may have been hallucinated last year may not be the case now. A year is a really long time in the LLM world.

Last year, mainstream LLMs were just over 2 years old. Open AI's ChatGPT will have its 4th anniversary in November and now it's generating proofs for math theories that nobody had been able to solve. I still remember how many on this forum in 2023 were so convinced that AI was garbage because it couldn't add at the time. LLMs are coming out of their toddler phase and headed for serious capabilities before they're even in their teenage years. Focusing on what they were like a year ago is like looking at the dust of a dragster that just blew past you.