Yeah but that test is for humans, its like saying a tiger is a pretty good wrestler because it did well in the wrestling ring, instead of just understanding that a tiger is big and brutal, which just happens to coincide with the sport. The tiger needs its own sport/test.

I'm not sure what your analogy proves, I wouldn't want to fight a tiger with bare hands. I agree, that humans and machines are not the same, but when it comes to intelligence, we are are in the same sport: answering questions in plain text.

The exam was a reply for this part:

> isn't even close to understanding scientific concepts and knowing how to connect them

That test in my book definitely proves that it can understand (or let's say process and create abstractions) scientific concepts and can place them on a knowledge graph.

I'm just saying its not the same test for the model vs the human. The human has to remember stuff, understand stuff in their limited human way of understanding, which the model does not. So what is really the utility of "testing" it? You don't take tests to become a better test taker, to prove you can pass the test. A test is more a tool to use to try and gauge our understanding of things, its not something itself. You're logic here is extremely flawed.

We obviously work in different ways, but how it comes up with the answer doesn't really matter.

> So what is really the utility of "testing" it?

Eventually these models will be integrated into everyday products so we need to learn how trustworthy they are. These tests are written in natural language, the same way everybody interacts with them.

It absolutely matters if you're going to say it passing a human test speaks to anything at all! If humans had a hardcoded multiplication table in their brain, it would be silly to "test" them on their multiplication ability, it would be pointless. Test's are useful precisely to gauge human retention/understanding, but they are even then a crude tool as I think anyone whose taken a test would agree. It borders on absurd to take it seriously for a model. Its like applauding the calculator for doing math.

If you really do think this stuff is important, show it on its own terms. Make tests suited for the model with rigorous expectations tuned for a computer, not a human. Then you can start showing some impressive things. I just urge you to see the basic flaw in your conceptions here. I don't want to rob you of your enthusiasm, just better direct it.

If you wanted to hire people to multiply numbers, it would be very worthwhile to test the functioning of the hardcoded multiplication table, or whatever other implementation they had.

Nobody is saying that GPT-3 is smart. But it is surprisingly capable.

Right, and again, if there was some virtue to passing a test in itself, I would be on board. But a test for a human is meant to show something about the human that isnt just their test taking abilities. But with GPT we can only ever say that it has good human test taking abilities. Again, I know people are excited, and they have reason to be, but you gotta adjust your mindset here a little or you will end up in a weird B.F. Skinner logic hole.

In elementary school they hardcode the multiplication table and they do test for them, at least it was my experience. For some weird reason we were taught to remember 8x125 is 1000. It came handy maybe 3 times in the past 35 years, but in those cases people looked at me like a wizard :)

> Test's are useful precisely to gauge human retention/understanding

Indeed, and these models are also compressing information they were trained on. It makes sense to test them how well can we access and query that knowledge.

I'm not saying at all, that passing tests make them human or anything like us, but it speaks a lot about the quality of the model. It's not the ultimate quality check of course, but it's an important one, because you can assign a score to the results of the test.

> Its like applauding the calculator for doing math.

Show that calculator to Charles Babbage, he would be in awe :) This is probably the leap why most people are freaking out, it feels a significant step towards intelligent machines regardless of the system under the hood.

I actually tested ChatGPT, (like a few million other people). I'm not saying it was done with scientific rigour of course. It passed a few flawlessly and failed some other. I believe I know what I'm interacting with, I've read about how GPT works and I'm planning to learn more. Don't worry about my enthusiasm :)

from your reference video (thanks):

> 7:36 The exam is structured with multiple choice questions at the start and then some freeform short answers towards the end.

Imagine if you could look up all material in the last 10 years published about astrophysics (or insert any major topic here) near instantly, the question's details are also presented for ingestion, then make prediction what is and is not based on weighing the frequency and potentially the source of information, and then respond in relatively coherent sentences.

There is no astrophysics in this.

ChatGPT adds explanations why it picked a certain answer. But you are right in a sense, this checks for existing knowledge, It probably couldn't come up with a solution to an open problem.

Minor correction: ChatGPT does not add explanations why it picked a certain answer; ChatGPT adds explanations that it thinks fit with the answer it has already given. The what predates the why.

That's why, if you want ChatGPT to actually think about a problem, you have to get it to give the explanation first, gradually, and only then output the answer. That way, GPT can use the rationale it has given in the prediction of the answer that fits best. If it has to give the answer first, it's already committed.

If a human has to answer a question, they follow a process like this:

- parse the question

- compute a rationale, and evaluate it to gain an answer

- output the answer first, because it's the most useful

- output the cached rationale.

But GPT trains left to right, and it cannot see the first two steps as they're not in the text. From its perspective, most humans produce answers and explanations simultaneously, instantaneously and intuitively. This makes it almost impossible for GPT to learn how to think. It has to rely on the very few samples where humans give the rationale first, and then the answer. And because this is rare, you have to really push GPT to actually use its scant cognitive abilities. ("Let's think about it step by step.")