I've read here that some of you are using "socratic skills" where you give the LLM some content and then you get a series of questions in order to learn but I don't think it's the only approach to this.
I'll start tackling the "Attention is all you need" paper and then I want to go deeper.
What's your approach?
I've noticed that for me it's much faster to learn by doing exercises targeting the exact thing you need than just trying to apply that to a much bigger chunk of work directly, especially for someone that's a bit of a failing perfectionist.
I was just looking for that and I saw https://github.com/THU-MAIC/OpenMAIC. I haven't tried it as I'm trying a few small experiments to see if I can create a curriculum for a framework with no docs.
The first naive one failed for multiple reasons but mostly because many guardrails are needed and we need a proper process that tracks too many things.
It’s also good as a fuzzy dictionary/thesaurus/expression finder. It’s good for refining my writing skills.
I use it to learn German by writing the texts myself then comparing it with the AI’s version. It highlights subtle mistakes I make.
And the speed of learning is still limited by the individual's constraints: ability to focus for an extended time and chance to apply the new knowledge so the learning is reinforced.
Even before LLM's there was no shortage of knowledge and also the organization of them.
So I resort to same method I've been using ever since: reading books, writing code by hand, and sometimes, writing a blog post so I could refer things later.
Actually, I've added a further constraint on myself once I realised the above: to never learn anything from videos. Youtube is simply too distracting. Books are great, and coding by hand the stuff I'm trying to learn... I feel I need that experience to imprint things in my brain.
The surprise was a) it has been much harder than I expected b) I ended up learning a lot. I thought it would may be pick up a bit of programming. But it was mostly everything around programming: how OAuth works, why my sign in kept expiring after an hour (a code deletion took out the token refresh route with it), how Stripe webhooks fail silently, why emails land in the Promotions tab, how Google decides what to crawl. Each broken thing forced me to understand a system I didn't know existed the week before. Learning to work with the AI model has also been a huge lesson on its own:)
1. Ask AI for the 10 most commonly used verbs in ordinary daily speech in language X.
2. Prompt AI: generate 4 short sentences for each of the 10 verbs at the A1 (beginner) level. Provide the sentences in .csv format compatible with the Anki flash card program.
3. Import .csv into Anki. Study your 40 cards.
As you advance you can do much more like having conversations with AI avatars, transcribing videos, translating text, generating audio clips, and so on.
Dr Taylor Jones, a PhD linguist, goes into greater detail here: https://youtu.be/xQXiSGDXknA?
That being said, they absolutely can spin out a topic list and help you find gaps in your knowledge in order to create a self-study plan. As a over-engineered librarian, they are terrific.
You trust some and distrust others at face value based on prior experience and what you want to believe. Sometimes you check after the fact and sometimes you don’t.
It’s the same exact thing with LLMs.
Watching Youtube.
Same as before.