3 ms·
LLMs effectively disambiguate in an arguably similar way humans would: look at the same term in multiple contexts to reduce confidence in incorrect interpretati
by naasking 3y ago
LLMs effectively disambiguate in an arguably similar way humans would: look at the same term in multiple contexts to reduce confidence in incorrect interpretations, leaving only correct interpretations.
- mistermann 3y ago> leaving only correct interpretations Unlike humans, LLM's can consistently realize (or at least claim to realize) when they've made a mistake (like assuming one's beliefs are necessarily correct, [because some persuasive story]).