[lijok]

Accelerate education at unprecedented levels (if we can figure out how to integrate LLMs into the education system).

To this day I find myself doing a double take every time I'm about to ask ChatGPT to produce a ludicrous amount of output, because it would simply not be reasonable to ask that of some random person. But with LLMs you can do that, over and over again and it will comply every time.

How great is it that if I want to learn about, say, hexagonal architecture pattern, I can ask this thing to produce 500 LOC in a language I happen to be familiar with, and then interrogate it to no end until it clicks for me?

[captainbland]

On "accelerate education" - it's often a bad teacher especially on higher level subjects . I've asked it to summarise research for me using the Bing GPT4 model and it would frequently come to conclusions that couldn't be corroborated in the source material, that were even contradictory across different chat sessions and even generate citation links that were totally irrelevant and incorrect, and then try to tell me that it was of a totally different subject to what it had actually pointed to.

Regular chatGPT is even more dangerous because you have no idea what it's referencing most of the time. Yet it lowers the bar to this poor information to such a degree that people will be incentivised to use it regardless.

[lijok]

That has not been my experience with ChatGPT 4. If I were to ask it about some specialized info about quantum field theory or superconductor tech, I'm sure it would send back nonsense fairly frequently. But if I ask it to explain the difference between median, mean and average, or ask for examples on an architectural pattern, it's seldom incorrect.

[captainbland]

If you prompt directly about the subject it can be fine but it veers into BS territory more often if you try to get it to synthesise information meaning it's less good with applied examples that are given to it by the user. Not always but the error potential is definitely higher.

The problem then I think is that this means that students will have to prompt "on rails" to get reliable answers. The most curious students who want to stretch their knowledge are likely to cause it to generate falsehoods which are presented confidently and convincingly.

[spion]

I'm not sure the error potential is much higher than with average teachers, but I guess it depends on the domain. For coding the error rate is lower than a typical teacher right now.

I suspect we'll soon figure out an effective way for LLMs to look-up reliable references to confirm their answers, which should improve the situation drastically. As they are now most LLMs are very barebones. An educational LLM could e.g. connect to the university textbook library and use that to verify its answers.

[captainbland]

I think the main issue today is its inability to just say "I don't know" when appropriate. A teacher can do that and sometimes that's a better answer than a fabrication. A teacher can also use less certain language when their confidence level is low as appropriate.

I think this is probably a technically solvable problem but doing so has potential 'optics'/marketing issues in terms of reducing the level of confidence it projects which a lot of people associate with competence.

[eletrotupi]

>Accelerate education at unprecedented levels (if we can figure out how to integrate LLMs into the education system).

More likely we'll replace poor people's teachers with ChatGPT and whoever has more cash affords actual teachers. This is so real that we're experiencing this >today<, in a different scale with private schools and distance learning in countries such as Brasil where there are projects to reduce schools and/or move some of them to use distance learning.

[spion]

For most people today, learning / teachers can be far, far worse that even what ChatGPT 3.5 can provide.

[layer8]

I don’t know, but anyone who has the motivation to “interrogate it to no end until it clicks for [them]” already didn’t have trouble to become well-educated before. AI can speed things up for those people, but I can imagine much less impact on general education.

[spion]

The speed difference is out of this world. I have very long sessions with ChatGPT on the Rust ownership and borrowing system. Its sometimes wrong (would say about 1 in 30 messages), but when it is it becomes clear quite quickly and then I search the internet armed with the new knowledge of terminology and relations within that terminology

[lijok]

I agree on your first statement, and raise a challenge on the second.

We're smart, lets figure out how to create the same impact on general education.

Could we wire up ChatGPT so it asks us questions and leads the conversation instead of us? Could we also make it present the questions in ways that make educational youtube channels like Kurzgesagt so easy to engage with?

We could start with private tutoring. LLMs tailored to educate kids of rich parents. Easy market fit as they're already looking for private tutoring. We prove out the need then quickly move onto schools.

Imagine if we could show that engaging with an LLM for an hour produces as much learning as sitting in class for 4. Could we cut school days in half, set LLM interaction as homework for an hour (teachers would get reports from the LLM on how the child is doing)?

[ac2u]

>already didn’t have trouble to become well-educated before

I agree that I didn't have trouble interrogating a subject until it clicked before if I was sufficiently motivated. However ChatGPT reduced the time needed a lot, and thus motivated me to take on more learning than otherwise.

Of course, I take "facts" it outputs with a pinch of salt until I double check things with my own research, but it's great at uncovering unknown unknowns and is like an infinitely patient tutor that will let me throw analogies at it to test if my understanding is correct or needs adjustment.