I’m surprised Karpathy thinks this is a viable solution for a quasi-continual learning system. Yes, it’s cool to experiment with these sort of ‘intermediate knowledge systems’, but the real goal within this current LLM paradigm remains clear: new information within a knowledge system should be manifest through updating the weights! Many efforts are taking a crack at this, which this really helpful talk [0] by Jack Morris goes into. I’m in full agreement with the other comments here: this ‘LLM Wiki’ merely results in “2nd-order information” that will only muddy the picture.
Furthermore, I’m left wondering why Karpathy jumped to a wiki-style approach versus something more akin to “version control but for LLM notebooks”.
I have no doubt that using LLMs as a notebook, which many of us have been doing now (and the best end-user application of which is Google’s NotebookLM), are a viable path forward for knowledge management. I find myself going back to certain LLM conversations as a ‘running log’ of a given project (akin to notebook-style thinking/building).
But what about merging concepts from the version control world (Git/SVN), rather than the wiki world? Karpathy should do more explanation about what took him down the wiki route vs that, in these early days of using LLMs as notebooks.
[0] “Stuffing Context is not Memory, Updating Weights Is": https://www.youtube.com/watch?v=Jty4s9-Jb78