Ryan Lufkin

At least understand how these tools work.

Then you understand what they're capable of.

And then you can actually make the more informed decision about how deeply do you want to use these tools and put you in a better position to help your students understand how to use them ethically, how to use them effectively, things like that.

I will point to what was called the strawberry conundrum.

ChachiBT 3.2, if you ask it how many R's were in the word strawberry, it would tell you two.

And you'd say, go back and maybe look that again and tell me how many are there.

And it would say that there are two R's in strawberry.

Of course, there's three R's in strawberry.

And they could not figure out because essentially large language models are a black box.

You don't know what's going on inside there.

They could not figure out why it hung up on that.

And it wasn't until the next model, 3.3, came out, they fixed that.

That's the reason that we all need to be experts.

ChatGPT has a high propensity for what we would call hallucination.

More often, it's confidently incorrect.

What it's trying to do is give you what you've asked it for.

Some cases, it makes things up.

It'll make up links, sources.

It'll make up whole sets of information because it's just trying to give you what you want.

And if it doesn't find it or doesn't spend the time to find it, it just kind of makes it up.