I’m a university instructor and I definitely use AI to help learn material. But I write all the student-facing material by hand. And I use LLMs to check my work for errors, directing it to never offer a rewrite.
If I found out I’m paying thousands of dollars in tuition just for my instructors to be learning the very subject matter they’re teaching from LLMs I would be so upset that I might drop out and seek refunds.
Look, if I had a magic button that eliminated LLMs from the planet, I’d have already pressed it. About the best I can do is keep losing Anthropic money with every prompt I give. Hopefully they crash out and we can all run local models once RAM gets cheap enough to buy.
But the fact is that LLMs are actually useful for getting to accurate answers more quickly than slogging through SEO crap, much of which was written by LLMs anyway.
Noooooooo. Current models are completely incapable of fact checking, despite any claims they may make. They get extremely basic shit wrong. For example, one popular model was recently caught telling people the wrong day for Election Day. Which is just so incredibly basic but they can’t manage it.
Current models are completely incapable of fact checking, despite any claims they may make.
This is demonstrably false. Ask it to find the bugs in a chunk of code, for example. You can try that right now.
You always have to back up what it says with an authoritative source, but that’s not particularly tough to do once you have a candidate answer. And if it’s code, you compile it.
It’s arguably far easier than trying to sift through the mountains of SEO shit yourself.
I’m not sure whom I’d trust more than yourself for what your definition of a fact is. It’s clear “this program has bugs” is not a fact by your definition. Are you talking about things like “the number of souls on the Titanic”?
Yep, you get it! And thats the kind of thing LLMs are weak at fundamentally. It’s terrifying that an educator is relying on them for that because they simply don’t work that way. They are much better suited to tasks like the example you gave.
I’m a university instructor and I definitely use AI to help learn material. But I write all the student-facing material by hand. And I use LLMs to check my work for errors, directing it to never offer a rewrite.
If I found out I’m paying thousands of dollars in tuition just for my instructors to be learning the very subject matter they’re teaching from LLMs I would be so upset that I might drop out and seek refunds.
I expect better from a university instructor.
Well, it’s hardly the only place I learn things. But if you can verify a fact that an LLM gives you, it’s still a fact.
Another fact: there’s a pile of human shit out there, too. So you get the truth by hook or by crook.
if I can verify a fact an LLM gave me I’m still:
Sure, humans still lie and manipulate and fuck up all the time. That doesn’t excuse or explain the many flaws of LLMs.
Look, if I had a magic button that eliminated LLMs from the planet, I’d have already pressed it. About the best I can do is keep losing Anthropic money with every prompt I give. Hopefully they crash out and we can all run local models once RAM gets cheap enough to buy.
But the fact is that LLMs are actually useful for getting to accurate answers more quickly than slogging through SEO crap, much of which was written by LLMs anyway.
Noooooooo. Current models are completely incapable of fact checking, despite any claims they may make. They get extremely basic shit wrong. For example, one popular model was recently caught telling people the wrong day for Election Day. Which is just so incredibly basic but they can’t manage it.
Soulds deliberate to me.
Unless getting the election day wrong was just one of millions of other things it got wrong.
If this LLM gets 99.99% of things right, but it got the election day wrong, that’s intentional.
This is demonstrably false. Ask it to find the bugs in a chunk of code, for example. You can try that right now.
You always have to back up what it says with an authoritative source, but that’s not particularly tough to do once you have a candidate answer. And if it’s code, you compile it.
It’s arguably far easier than trying to sift through the mountains of SEO shit yourself.
That is not fact checking. Ask your slop machine to explain to you why that isn’t fact checking.
Why not so being obtuse and just tell me what you mean by “fact” and I’ll tell you if I use LLMs for that or not?
Because I want you to hear it from somewhere you actually trust.
I’m not sure whom I’d trust more than yourself for what your definition of a fact is. It’s clear “this program has bugs” is not a fact by your definition. Are you talking about things like “the number of souls on the Titanic”?
Yep, you get it! And thats the kind of thing LLMs are weak at fundamentally. It’s terrifying that an educator is relying on them for that because they simply don’t work that way. They are much better suited to tasks like the example you gave.
Except I don’t rely on them for those kinds of facts, so… I guess we’re good here. :)