Hacker News Reader

Comment by hangsi | Hacker News Reader

Arguably it is succeeding on this front - LLMs have been much more helpful for learning about mathematics topics than Wikipedia for a while now. Admittedly a somewhat low bar, but they are genuinely helpful. In many other fields the models can still only pull up equal (in my opinion).

View on HN

It's the 'can answer arbitrary questions' part for me.

You could give me the most well crafted, comprehensive document ever and I would still misread bits of it, or miss some crucial bit of information that contextualises the rest or whatever.

With LLMs I can ask endless dumb questions in a way that just would not be practical otherwise. The quality of the responses doesn't even have to be particularly good for it to be extremely useful - it's basically a turbo charged rubber duck for learning new concepts.

I want the multimedia version of this that can animate pictures and play sounds, to help explain things.

More like The Young Lady's Illustrated Primer [0] than our current chatbot paradigm.

I expect there are 48956789560984 startups trying for this now, though, so... soon?

[0] https://en.wikipedia.org/wiki/The_Diamond_Age

Agreed. An always-on agent focused solely on improving your knowledge/skills could ask you questions daily, build a picture of the gaps in your understanding and learn where to push you next.

I'm experimenting with this at https://miletus.app

I'm also curious about training models with that singular goal. If anyone is exploring similar ideas, I'd love to chat.

How do you handle the 'people often learn through repetition' problem?

How can you source or generate a curriculum graph (similar to what math academy does)?

It seems to me that it would be cool if that could be solved automatically for any learning goal/target

I started with a curated backbone across several fields, drafted with frontier models and reviewed by hand. The automatic (and fun) part happens when you pick a goal and a quest gets generated with its own prerequisites, mapped onto that backbone.

It's def not as robust as Math Academy's graph yet, they spent years hand-building and testing it on real students. But I bet a few more iterations of frontier models get there, especially once there's learner data to correct it.

> LLMs have been much more helpful for learning about mathematics topics than Wikipedia for a while now.

I can't imagine anyone but someone extremely gifted at first parse understanding learning mathematics from wikipedia. No effort is made to make it understandable for the average reader.

Reply on news.ycombinator.com