This was very insightful, thanks. Only the conclusion is a bit problematic: Hitting the limit of what it can do or of text it can ingest does not mean it is AGI. Could be stuck way below that.
The plan earlier makes more sense. Having LLM do what it does best, function as a language between tools, is likely the most robust way forward, and is also programmable and understandable.
This was very insightful, thanks. Only the conclusion is a bit problematic: Hitting the limit of what it can do or of text it can ingest does not mean it is AGI. Could be stuck way below that.
The plan earlier makes more sense. Having LLM do what it does best, function as a language between tools, is likely the most robust way forward, and is also programmable and understandable.