Can Reinforcement Learning Lead to AGI? - Daniel Han, Unsloth

Can Reinforcement Learning Lead to AGI? - Daniel Han, Unsloth With the release of reasoning models like the O series of models, DeepSeek R1, Gemini and others, reinforcement learning has come back in full force. The question is whether we can scale RL to infinity in the limit to reach "AGI", and are RL algorithms actually learning new knowledge or are they simply accentuating knowledge in pretrained models? We will try to address these questions, and provide predictions on where RL is heading.