Hacker News Reader

Comment by ad404b8a372f2b9 | Hacker News Reader

I've seen ChatGPT draw logical conclusions from (most-likely) previously unseen axioms. I've seen it invent ideas that (most-likely) were not in his training set. I've seen it fail too, but when it succeeds I can not substantially distinguish it from the novelty or logical reasoning of a human.

On niche subjects, it fails often. But you can get it to correct itself by making it think logically by spelling out axioms, which seems to fit your definition of novelty.

All the arguments I've read about it not being intelligent seem to stem from some form of essentialism. Right now it's dumb intelligent but I think giving it an adjustable recursive depth, similarly to what BulgarianIdiot suggests, will go a long way to improving it given its base capabilities.

View on HN

> I've seen it invent ideas that (most-likely) were not in his training set.

How can you possibly know what's in its training set? Its training set is so large that any "guess" about what is or is not in it is a fool's errand. Given that we know how the technology works, there's no reason to claim emergent behavior when Occam's razor would say the items are likely in the training set.

It's not a fools errand, you can't know definitively but you can have a level of confidence.

It's quite easy to invent a sentence that's never been uttered before, it's also fairly easy to hit the limits of the internet's knowledge in pretty much any complex discipline that's not software engineering.

And it's apparent when you speak to ChatGPT, he will hallucinate a lot to compensate for niche topics it doesn't know much about.

Occam's razor wouldn't point to ChatGPT having omniscient knowledge of all things ever thought of and that will ever be invented.

This kind of stuff clearly shows that GPT can generalize concepts:

https://twitter.com/emollick/status/1626084142239649792

And this kind of stuff shows that it clearly cannot:

https://acoup.blog/2023/02/17/collections-on-chatgpt/

> If you know nothing about either book, this answer almost sounds useful (it isn’t). Now this is a trivial research task; simply typing ‘the limits of empire review’ into Google and then clicking on the very first non-paywalled result (this review of the book by David Potter from 1990) and reading the first paragraph makes almost immediately clear the correct answer is that Isaac’s book is an intentional and explicit rebuttal of Luttwak’s book, or as Potter puts it, “Ben Isaac’s The Limits of Empire offers a new and formidable challenge to Luttwack.”

> A human being who understands the words and what they mean could immediately answer the question, but ChatGPT which doesn’t, cannot: it can only BS around the answer by describing both books and then lamely saying they “intersect in some ways.”

> The information ChatGPT needed was clearly in its training materials (or it wouldn’t have a description of either book to make a lossy copy of), but it lacks the capacity to understand that information as information (rather than as a statistically correlated sequence of words). Consequently it cannot draw the right conclusion and so talks around the question in a convincing, but erronous way.

That appears to show that it can't answer some obscure trivia question to the standards of the blog post author. Nothing to do with generalizing concepts.

It's not an obscure trivia question. It's a question that anyone with an understanding of the material in question should be able to answer, with something vastly better than a superficial 'the two books are similar, but, like, different and talk about different stuff'.

It immediately cuts to the heart of the matter - an LLM does not understand things. It does not know anything but which word follows another. It can't do anything but regurgitate things it has read in an incredibly lossy manner.

But it's a neat party trick that seems to fool a lot of people into thinking that there's something legitimately useful, here.

Large language models are not search engines, even though a lot of people seem determined to treat them as such.

I don't think that blog post's excercise is a particularly useful demonstration of anything.

Indeed. And telling it to spell out the steps in math solution is essentially (ab)using the chat log as the memory for an iteration loop.

This demonstrates then when you instruct it "break down the problem and solve it iteratively" it comes out with superior solutions. More accurate, more novel. Ergo, then if it was allowed to iterate in this way internally, it'd produce better answers by default.

This thread sums up the same idea: https://twitter.com/ItIsFinch/status/1634730897520951296

Yeah, I think it's going to be a promising area of research. It's known as Chain-Of-Thought (CoT) in the literature and there are a few papers already out:

https://openreview.net/forum?id=_VjQlMeSB_J

https://openreview.net/forum?id=5NTt8GFjUHkr

https://openreview.net/forum?id=qFVVBzXxR2V

I'm also looking forward to explicit long-term memory mechanisms that don't rely on the main architecture's weights nor on CoT.

There's no memory or iteration log. You are just priming the (more or less deterministic even...) predictor with some history. You are doing a lot of work yourself, setting a factually correct context (and this has severe limits...).

As for "we just need a way to include iterations": isn't this what attention is supposed to do (kind of dynamically updating weights). The usual way to really update weights are variations on gradient descent. Can you link a paper outlining how we integrate your proposal into the current framework (which took around 20years to mature)? Otherwise your statements are Sci-Fi (at the flying cars-level).

Oh so there's no memory, just recallable history.

Oh so it's not learning by example, it's just primed by a sample.

Oh so there's no iteration of thought, just a loop of prediction.

OK.

Also, the papers you asked about: https://news.ycombinator.com/item?id=35115563

BTW, people are already implementing recursive/iterative queries on the ChatGPT API, and getting promising results.

The langchain project is an example of the iterative queries approach. It comes with constructs for working memory, factual lookup / calculation agents, etc.

https://github.com/hwchase17/langchain

The general (non-technical) guideline is that the LLMs can "answer" anything you just gave them the answer for. So you give it a problem, ask it how to solve it, tell it to use that method and explain the data it needs, give it that data, and then show it everything at once: "With this data you requested and summarized, use this technique to answer this question".

Are you on other social media BTW? Thx.

Afraid not, only here. You can at me at scrap.titanium@fastmail.com though.

Reply on news.ycombinator.com