3 ms·
AI Chatbots Can Talk Conspiracy Theorists Out of Believing
- delichon 2y agoThey are consensus extraction devices. They will argue for the consensus conspiracy theories and against the conspiracy theories that oppose the consensus. If you train an LLM on a set of tokens that include a consensus for a conspiracy theory rejected in a baseline set, it would be just as convincing.
- bbor 2y agoTBF they will also happily argue for non-consensus conspiracy theories, if you ask nicely and there's no external guardrails in place. Their only concrete/stable/omnipresent consensus is a linguistic one, not a sociopolitical one, IMO.