Short summary: This is about a recent MIT/IBM Watson collaboration paper I read today presenting a new framework for multi-agent cooperation where at least one agent is an LLM (GPT-4). I found the idea fascinating to build a collective intelligence of artificial intelligence units gradually - and as the units grow in individual intelligence (GPT-4), hopefully, the collective intelligence system will grow as well.
Another fascinating idea was to use the English language as common ground for agent cooperation (enabled by GPT-4) rather than a more symbolic (programming or low-level) language. No 0s and 1s!
