←back to thread

359 points FromTheArchives | 1 comments | | HN request time: 0.213s | source
Show context
oceanhaiyang ◴[] No.45293700[source]
No one who understands ai can rely on it to help us learn. I provided one with 100 citations I wanted to standardize and it deleted 10 and made up 10 to replace them. Can’t imagine this being used to replace a textbook or even explain a textbook.
replies(3): >>45293802 #>>45294049 #>>45296961 #
criddell ◴[] No.45293802[source]
> explain a textbook

I've had very good luck using LLMs to do this. I paste the part of the book that I don't understand and ask questions about it.

replies(3): >>45293911 #>>45293968 #>>45294577 #
bigfishrunning ◴[] No.45293911[source]
But the problem is, you don't understand the passage, so therefore how will you vet the answers? Seems like hallucinations would be very very damaging in this use-case
replies(4): >>45294006 #>>45294052 #>>45294616 #>>45295154 #
lacy_tinpot ◴[] No.45294052[source]
If you can't discern what good answers look like to the questions you're asking, you're not asking the right kind of questions.

Asking the right kind of questions is a genuine skill.

It applies to every domain of life where you are at the mercy of a "professional" or at the mercy of some knowledge differential. So you need to be a good judge of whether the answers you're getting are good answers or bad answers.

replies(2): >>45295186 #>>45295214 #
squigz ◴[] No.45295214[source]
> If you can't discern what good answers look like to the questions you're asking, you're not asking the right kind of questions.

Whaaaaat? How does this work? If you're trying to learn a new topic, how are you supposed to recognize a good (and truthful) answer, whether it's from an LLM or instructor?

replies(1): >>45295299 #
lacy_tinpot ◴[] No.45295299[source]
I would argue you're doing it right now.

By being skeptical of the answers, testing the answers, corroborating with other sources, etc.

This isn't new. This is literally how we've been exploring this knowledge game for thousands of years.

replies(1): >>45295376 #
squigz ◴[] No.45295376[source]
I would argue this isn't a fair comparison. There's a big difference between a fairly open-ended discussion about a topic both parties are at least somewhat familiar with, and someone trying to learn a new subject.
replies(1): >>45295473 #
lacy_tinpot ◴[] No.45295473[source]
All knowledge is open ended...

I bet when you're learning a new subject you do the same exact thing.

replies(1): >>45295501 #
squigz ◴[] No.45295501[source]
When learning from a source like a textbook, docs, or being instructed by a person, I do not expect the source of truth to lie to me, and verify everything they tell me.
replies(2): >>45296037 #>>45296588 #
lacy_tinpot ◴[] No.45296588[source]
Did we all just collectively forgot basic literacy?

You as the reader when you're reading anything are supposed to verify claims the author is making.

You never expect anything to be sources of truth.

That's why every textbooks either cites the sources or proves their work.

Very rarely do you have any textbook that's just a list of facts out of thin air. I don't think I've seen a single textbook, even bad ones, do this. They always cite their claims, or they show the logical steps to prove or justify a claim. Good textbooks make it easy to follow and clearly show their steps for the convenience of their readers.

Any good textbook seriously considers both the historic literature on their subject, presents the context of that literature, and shows some kind of proof of work that synthesizes all of that to support their claim.

This is always the case. This is how basic academic writing is done.

And it is the job of the reader to follow those citations, and to verify the claims. That's literally how our academic system works.

It's basic literacy.

replies(2): >>45296798 #>>45296979 #
1. squigz ◴[] No.45296798[source]
> And it is the job of the reader to follow those citations, and to verify the claims.

How do you verify the claims? Replicate every piece of research cited?