After seeing LangChain and the reasoning paper I think it's fairly obvious that we're just starting to scratch the surface of AI architectures, dictated largely

After seeing LangChain and the reasoning paper I think it's fairly obvious that we're just starting to scratch the surface of AI architectures, dictated largely by scalability of GPU resources. The LLMs we're playing with are at best the proof of concept of what will eventually be the message passing for the next generation of models.