Language models can now 'feel' their memories to make better choices

Researchers found a way to give language models 'emotional memories' alongside factual ones. This makes the models much better at making good decisions, especially when they also have factual knowledge.

Humans use gut feelings, or 'somatic markers,' to make decisions. This paper shows that language models can do something similar. It means AI could move beyond pure logic in complex situations, making decisions that require a sense of risk or safety. This could open up new uses for AI in areas where human-like judgment is critical.