AIAhead of AIabout 1 year ago
The State of Reinforcement Learning for LLM Reasoning
40 min read
GPT-4.5 and Llama 4 were trained without explicit reinforcement learning for reasoning, drawing muted reactions. Competitors xAI and Anthropic added reasoning buttons to Grok and Claude. OpenAI's o3 reasoning model used 10x more training compute than o1, showing room for improvement via reinforcement learning methods.
Level
Hype check
Tap to vote and see what everyone thinks.