[pautasso]

The problem is when students just blindly copy and paste from the chatbot and submit it as their own answer without even reading it.

They should be encouraged to read and review the LLM output so they can critically understand it and take ownership of it.

[dyauspitr]

In my experience, reading a solution and even understanding it doesn’t go very far in teaching you how to do something. I can look at calculus solutions all day but only when I actually try to solve them myself do I run into all kinds of roadblocks which is where the real learning happens.

[9rx]

You're right, but learning can take place when you need it. There is no real advantage to learning something ahead of time. The bottleneck is having awareness of what is out there to learn. You can't learn about what you don't know exists. Looking at calculus solutions all day should give a sense of what calculus can be used for, so that it is in your back pocket when the time you need it comes.

Well, at least it used to be the bottleneck. Nowadays you can just ask an LLM. For all their faults, they are really good at letting you know about what tools exist in out there in the world, surfacing more than you could ever come to know about even if all you did was read about what exists all day, every day.

[dyauspitr]

I believe to count as an expert on something you need to have a ready compendium of knowledge ready to go. It becomes very hard to tackle problems or gain deep insights if you don’t already have knowledgeable people that have thought deeply about a particular space. Maybe when we have supremely reliable LLMs that can replace humans we might not but we’re not there yet.

[9rx]

> I believe to count as an expert on something you need to have a ready compendium of knowledge ready to go.

You are certainly headed in the right direction, but not quite. To be seen as an expert in the eyes of others you need to have had a vision for something and to have successfully executed on it. If the vision was dependent on calculus, then you will have reached a point where you had to learn something about calculus, of course...

But that's different to having a taskmaster tell you to learn calculus for no apparent reason. Even if you follow through and built up a huge wealth of knowledge from it, you would still not be deemed an expert by others. You're no different than an encyclopedia, which isn't an expert either. It is being able to see things others can't and the ability to act upon it that makes an expert.

Learning taking place when you need it isn't the same as never.

> Maybe when we have supremely reliable LLMs that can replace humans we might not but we’re not there yet.

Frankly, even Page Rank already replaced humans for this. But LLMs are even better at it. Humans are just that poorly performing. Like I said before, even someone doing nothing in life but looking for what exists in the world could not take in as much as databases that have indexed every written thing.

[dyauspitr]

Calculus might not be the best analogy for my point since it’s pretty fundamental. When I think of an expert I think of an accomplished mathematician or a chemist, someone that can build on existing knowledge to provide new breakthroughs. You can ask an LLM for a particular formulation but you cannot make wide spanning connections to come up with something novel until you have a good understanding of a given space. Not all progress is a series of iterative problems and tasks that need to be solved. In fact for a lot of breakthroughs it’s making disparate connections.

[9rx]

> When I think of an expert I think of an accomplished mathematician or a chemist, someone that can build on existing knowledge to provide new breakthroughs.

I think we're on the same page here. Experts have both vision and execution. Someone who has simply learned a bunch of things, or a lot about one thing, is not what we consider an expert.

> You can ask an LLM for a particular formulation but you cannot make wide spanning connections to come up with something novel until you have a good understanding of a given space.

I don't get where you are trying to go with this. Using an LLM (or Page Rank for that matter) to search for tools that have been created/discovered necessary to fulfill execution of your vision seems to have nothing to do with what you are trying to say. Nobody would ask an LLM to do what you are suggesting, if I am understanding you correctly. LLMs are most definitely not good at that. That is AGI territory.

[themafia]

They should be encouraged to not turn in casual plagiarism as their own work.

I believe there is a mechanism for this already.