Our fault lines paper has really divided the web.
It’s amazing how many people seem to believe that humans are just large language models.
Yes, humans hallucinate. We actually hallucinate a lot. We even have collective hallucinations, such as religions. Human hallucinations are fascinating precisely because they reveal deep desires and fears, like the fear of death.
LLM hallucinations are different. They are just statistical glitches. They do not reveal hidden motives, values, or existential concerns. LLMs will never have collective hallucinations, because they don’t have shared desires and fears.
Despite this, many people are conflating statistical accuracy with epistemic accuracy. Because LLM outputs are often aligned with human outputs, many people infer that LLMs must have gone through the same epistemic stages humans do. This new condition is called Epistemia.
If you haven’t read the paper yet, you’ll find the link in the first comment.
W/
@Walter4C &
@matjazperc