I appreciate his emphasis on "always" here. Though given our current rise of artificial intelligence and ChatGPT, this is obviously a problem which people are attempting to overcome.

Sadly, AI seem to be designed for the commercial masses in the same way that Google Search is (cross reference: https://hypothes.is/a/jx6MYvETEe6Ip2OnCJnJbg), so without a large enough model of your own interests, can AI solve your personal problems? And if this is the case, how much data will it really need? To solve this problem, you need your own storehouse of personally curated data to teach an AI. Even if you have such a store for an AI, will the AI still proceed in the direction you would in reality or will it represent some stochastic or random process from the point it leaves your personal data set?

How do we get around the chicken-and-egg problem here? What else might the solution space look like outside of this sketch?

I could totally see this UI with a video generated version of Niklas Luhmann answering questions using the training set of notes in his online zettelkasten at https://niklas-luhmann-archiv.de/bestand/zettelkasten/tutorial

syndication link: https://bavatuesdays.com/ai106-long-live-the-new-flesh/comment-page-1/#comment-388943

https://ideaflow.app/

Audio transcription notes with AI

2023-12-12 Released on PruductHunt https://www.producthunt.com/posts/ideaflow

https://a16z.simplecast.com/episodes/a-true-second-brain-xrODaBD2

Recommended by Michael Grossman