Hacker News new | ask | show | jobs
by NalNezumi 461 days ago
I think your personal take is more so that you require "interaction" in the learning process; which is extremely common and universally accepted as best way to learn stuff.

>If I don't understand I cannot move on to next step

Everytime you encounter new hard concept, you really have to actively think/interact with that information to make it yours. This is often lacking in the "lecture" style teaching. One key benefit of reading is that you can do it at your pace.

I think for audio/verbal medium, you would probably have a completely different experience if you were talking to Feynman. He have a lecture for you, and everytime you get stuck, you can interrupt him and ask follow up question or ask further. This is probably why Private tuition is so valuable (and very very often used in East Asian nation); whichever fence of preference you fall to, a someone dedicated to teaching you is an extreme advantage.

1 comments

> This is probably why Private tuition is so valuable (and very very often used in East Asian nation); whichever fence of preference you fall to, a someone dedicated to teaching you is an extreme advantage.

That's gonna be a money printer for LLMs (even if for a short while - "no moat" & stuffs). Many people are fine with LLM-assisted learning, and able to navigate the hallucination problem. Then there's the (in)famous NotebookLM, that's known for only one of its many features: the ability to blend a bunch of source documents and generate an engaging podcast out of them. And a feature they're testing is letting the listener interrupt the podcast with their own questions. Meanwhile, OpenAI's "advanced voice" has been demonstrating for close to a year now, that a spoken conversation with an LLM can be done and feels... natural.

It's not hard to see the convergence of these ideas into ad-hoc experts: blend some textbooks, replace a podcast with a conversational "advanced voice" model, and suddenly you may experience "talking to Feynman". Not a real one - nowhere near as smart or likely to be correct - but still better than anything else you or me have access to.

(Whether or not you believe LLMs are good enough for this, or are too dumb a markov chain stochastic parrot something - it still begs to be tried, and someone will try it soon.)