Ted Werbel

OpenAI has probably implemented automatic routing to domain-specific, lower parameter count o1-mini models. Each model is trained on chains of thought that are highly specialized for a given problem space. This aligns with what qwen is doing with their latest models too. https://t.co/8sQQGeJCp7

Quick clarification - what he actually meant was that 5min of inference time compute could = reasoning capabilities as strong as models 1000x bigger than GPT4 and trained for 1000x longer! My apologies - bad interpretation on my part 😖