The LM Brief: The Syntax Illusion
This episode of Summation with Auren Hoffman discusses MIT research on how large language models can prioritize sentence structure over factual accuracy. This preference for grammar can lead models to produce misleading content or bypass safety protocols.