MIT Breakthrough Enables AI Models to Permanently Learn New Information Using Self-Generated Study Sheets

MIT researchers unveil SEAL, a groundbreaking technique that enables AI models to permanently learn new information by creating their own study sheets and updating internal weights, boosting accuracy by 15% on question-answering and over 50% on skill-learning tasks.

• MIT researchers develop SEAL, a new method that enables large language models to permanently internalize new information by generating their own study sheets and updating their internal weights
• The technique allows LLMs to create multiple synthetic data versions of new input, then use trial-and-error reinforcement learning to determine which version improves performance most
• SEAL improves model accuracy by nearly 15 percent on question-answering tasks and boosts success rates by over 50 percent on skill-learning tasks, though catastrophic forgetting remains a limitation