How MIT’s New Training Technique Could Make LLMs Masters of Complex Reasoning
Imagine asking a state-of-the-art AI, like OpenAI’s GPT-4, to not just summarize a financial report, but to anticipate market swings, strategize business growth, or deduce the culprit in a fraud investigation. As powerful as today’s large language models (LLMs) are, they still stumble when confronted with truly complex reasoning tasks—especially ones they haven’t seen before….