This idea stuck in my head yesterday as I was reading about Chauffeur Knowledge in “the art of thinking clearly” - AI is Chauffeur Knowledge. I was definitely not the first to make such a connection, and I did a quick Google. Most of the articles, however, only make the association and mostly go off on points that I don’t think are as pertinent, mainly that what capability you outsource, you lose (i.e. general thinking).
I think what I would have liked to see is: 1) people explaining why AI is chauffeur knowledge - fundamentally, how current AI (like ChatGPT) works is finding and matching similar information most relevant to the user provided question, and then predict the next word in a sequence to generate natural and relevant text. Now - here I wanted to say there is no ‘thinking’, but then decided to Google some more and it turns out, it’s not a simple answer! Read here for more: https://www.anthropic.com/research/tracing-thoughts-language-model Now, after this I’m not so sure AI is chauffeur knowledge anymore (not Claude at least LOL)…
2) what to do about it. This article I thought was pretty good: https://www.trenegy.com/publications/planck-vs-chauffeur-knowledge-dont-get-chauffeured-by-ai It gives concrete suggestions on how to work with AI (i.e. give context rich prompts and ask follow-ups).
3) future - how about AI agents? https://www.anthropic.com/engineering/building-effective-agents
That was quite fun - I went down a couple of rabbit-holes but it was interesting! Maybe the next time I’ll do a brain dump before I start Googling and sidetrack my own thoughts…