AI models can now learn from their mistakes, like humans do

New research shows how AI models can be taught to update their understanding of a situation over time. This means AI systems could better adapt to changing human intentions or unexpected events in the real world.

Most AI models are static. They make a decision based on what they see right now, without remembering past interactions or updating their 'beliefs' about a situation. This new approach gives AI a form of memory and the ability to learn from its own errors, making it more robust for tasks that unfold over time, like assisting a human in a complex environment. It moves AI closer to being able to handle dynamic, unpredictable human behavior.