he’s been salty about this for years now and frustrated at companies throwing training and compute scaling at LLMs hoping for another emergent breakthrough like GPT-3. i believe he’s the one that really tried to push the Llama models toward multimodality