Researchers suggest a new approach is needed to build LLMs with accurate world models.
LLMs have shown that they can excel at various things – like writing, generating computer programs, and more activities.
This can make it seem like these models are learning some general truths about the world, but the study found out otherwise.
For the new study, the researchers found that a popular type of GenAI model can provide turn-by-turn driving directions in New York City with near-perfect accuracy — without having formed an accurate internal map of the city.
Leave a reply