running language models in parallel with each one focused on a sub-task, all orchestrated by a conductor language model picture something like a massive tree of GPT models working on answering a single complex prompt

→ View original post on X — @alexalbert__,