ResearchPerplexityNotion
Never cite what the model told you. Cite what you checked.
Models invent references. Keep a 'why I believe this' line next to every citation.
Model-generated references look perfect and are sometimes fictional. The only defence is a habit.
- Every citation in my notes has a one-line "why I believe this" next to it: I read it, a supervisor recommended it, or Perplexity linked the actual PDF and I opened it.
- No line, no citation.
- Once a month I spot-check five at random. If one fails, I check them all.
It sounds paranoid until the first time it catches a paper that does not exist.