Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Cargo Cult AI

queue.acm.org

Post reply on HN
Print view View original

161–170 of 191 posts

Re: Cargo Cult AI

#161

Post by agalunar »

sebzim4500 wrote: post #151
neatze wrote: post #112

Earlier quoted context omitted.

> clearly don’t have if you’ve looked at lots of gpt4 examples for example, can you fine tune GPT to play chess at ELO 1600 ? If you don't know answer, you are in for surprise.

Given the lc0 policy network plays at a 2000+ strength on its own, I would expect that with enough finetuning gpt4 would be able to play way above 1600 strength. It's possible that finetuning would basically be training a new network from scratch and the resulting network would forget everything apart from chess. It would be a really interesting experiment, GPT-2 is probably too small but I think llama-7B might be su…

It's worth mentioning the truly ludicrous number of positions that chess networks are trained on, usually in the billions. Which is a bit interesting, because it suggests they're not as good at generalizing from examples as human grandmasters, who might see only millions. Perhaps this isn't surprising, since LLMs are this way with language, but given the structure of chess it's be reasonable to expect they wouldn't be quite as data hungry.

My personal take is that chess isn't all that structured, really. Yes, the ruleset, pieceset, and board are small, but the tactics of the game make it chaotic – seemingly trivial differences in board state can lead to dramatically different consequences.

Re: Cargo Cult AI

#162

Post by foobarqux »

_yb2s wrote: post #79

The whole premise of this article hinges on the idea that LLMs have fundamental limitations that they clearly don’t have if you’ve looked at lots of gpt4 examples. For example, it can do scientific thinking if you specifically ask it to, and it can reason about totally new situations outside of the training data based on generalizable models of reality it creates to predict training data. If you are certain these lim…

LLMs (including GPT4) clearly do have limitations, no one disputes this. Those limitations have certain general characteristics which you might categorize informally as failures of "reasoning", "scientific thinking", etc. You could redefine or reinterpret those words to mean something else or to be used in a misleading way but it wouldn't change the fact that LLMs fail to do many things that humans can do.

Re: Cargo Cult AI

#163

Post by Freebytes »

ilaksh wrote: post #55

You can give GPT-4 some data and ask it to make a hypothesis to explain it, give it tools to test the hypothesis, and have it try to establish causation. You can literally do that now quite easily with ChatGPT abilities like plugins and it will prove the premise of the article false. An interesting contrast can be drawn between this article and a criticism like Yann Lecun's. In that his actually has substance. https:…

The definition of life involves many scenarios that do not need to be related to artificial intelligence. I think you were suggesting the term sentience. However, we should likely devote our energies to AI tools that do not attempt to mimic or achieve sentience. We can make tools that are far beyond the capability of humans while not requiring sentience as part of that equation. Machine sentience and intelligence are independent.

Re: Cargo Cult AI

#164

Post by cubefox »

albertzeyer wrote: post #82
kerkeslager wrote: post #60

Earlier quoted context omitted.

> Echoing many who now find themselves blindsided by the emergent abilities and rapid adoption of LLMs, the OP I've said this over and over again: there are no emergent abilities . Before you leap to link me this paper, I'll link it myself: https://arxiv.org/abs/2206.07682 I read that paper. Did you? Did you understand it? Because if you had, you'd have seen that early on they define what they mean when they say "eme…

> the idea that larger models can do more than smaller models, which should be a surprise to no one. Actually this was quite a surprise to a lot of people, since the whole race to scaling up began, with GPT2 or so. It was totally not obvious that you can scale up the model (and also training) and it would improve the performance. Many (most?) people thought there would be some limit, and we were close to that limit w…

The limit of LLMs becomes clear when we consider how they work. They gain their abilities from imitation learning. They imitate human text. This could get them to Wikipedia level intelligence (or knowledge), which might be somewhat superhuman, but not much further.

To really go beyond human intelligence, you'd need a system which isn't constrained by imitating text. In fact, animal brains probably work by predicting sensory experiences (predictive coding), which are a function of reality. Unlike predicting text, which is a function of human ability.

Re: Cargo Cult AI

#165

Post by bionhoward »

photochemsyn wrote: post #75

The thing about LLMs that is revolutionary is just how fast you can find a solution to a problem or expand on answers to a problem even to the point of generating a realistic computational-mathematical model of the problem. For example, the author brings up Kepler, so let's ask: > "Please explain in concise terms how Kepler used Tycho Brahe's observational data to come up with Kepler's three laws, on ellipitical orbi…

Hey, check out this paper, it’s probably got a bunch of descendants by now:

https://arxiv.org/abs/1810.10525

Also DreamCoder is relevant.

I reckon the answer to your real question is “yes”

Re: Cargo Cult AI

#166

Post by Freebytes »

tunesmith wrote: post #155
fsckboy wrote: post #64

Earlier quoted context omitted.

> There's absolutely no reason to be "brutally" honest. You mean to say there's no absolute reason to be "brutally" honest, because then you can see there's no absolute reason to be smotheringly polite either. (did you read the brief piece I linked?) There absolutely is a reason to say what springs to your mind, it's quick and efficient, and that's something that people who quickly come up with quality thoughts prize…

Yup, I read it, and I read it months ago too, it's come up a few times. What I'm saying is that it's absolutely possible to quickly and efficiently say those quality thoughts that springs to mind, in a respectful manner. And yes, it is a skill people should learn because it gives you a superset of advantages compared to if you don't. You can talk to a wider slice of people, and learn more from them as well. It's not…

You hit the nail on the head, especially about the time-consuming aspect. People confuse honesty and respect. You can always be truthful while still maintaining respect and appreciation of others.

"Did you like my singing?"

Option A: "No. You sounded like an animal being slaughtered. If this is a hobby of yours, I would recommend a different one."

Option B: "Not really."

Option C: "It is not my style, but you sounded like your were enjoying yourself."

Option D: "You really had a lot of passion. There were times you struggled, though, so you should record yourself and practice on those points. When you have everything ironed out, I would be happy to offer an honest critique again."

This is going from rude responses to respectful responses. Option B is actually more honest than Option A, but is less 'brutal'. However, if a person is truly seeking a critique, you can offer one in detail, but it is probably better to ask if they are seeking your actual preferences or your critique in that type of situation.

Re: Cargo Cult AI

#167

Post by Freebytes »

surgical_fire wrote: post #46
celestialcheese wrote: post #15

Earlier quoted context omitted.

Simple things that don't seem impressive to you, but are impressive to others, may be because you haven't experienced the first hand difficulty of doing that thing pre-transformers/GPT. For example, I get _unbelievably_ excited with knowledge extraction and question answering demos on PDFs. Why? Because i've built similar systems for over a decade and know how difficult it is to build on top of messy archival data. N…

I am kind of in the same boat as you. I sit in a weird spot of thinking the current trend of AI is really impressive, but also thinking people are massively over-hyping it (with a smaller counter wave of some people really undermining it). Having tangled with natural language processing and transformation in the past (always with dismal results), I can say it's one of the most annoying problems to tackle in computati…

I had thought about a translation system that sought to translate into a different, more coherent language, pass that into an AI model, get a response in that language, and then translate the results back into English. Early OpenAI responses did not seem very impressive to me; however, that was before ChatGPT came along and blew me away. And GPT-4 gives even better responses! I am not sure if the intermediate language solution is necessary or viable at this point because you would need a powerful LLM to perform the translation. But, there might be value in symbolic logic being part of the machine learning vernacular as part of the early tokenization and training process.

Re: Cargo Cult AI

#168

Post by cubefox »

albertzeyer wrote: post #142
kerkeslager wrote: post #93

Earlier quoted context omitted.

You didn't read the post you're responding to, and shouldn't be responding. In fact, you didn't read the part of the post you quoted, where I said it should be a surprise to no one. But, unsurprisingly, the sort of people who stop reading at the first chance they see to correct something, are easily surprised, since actually understanding LLMs would require actually doing some nuanced reading.

I think you are misunderstanding sth. I did read your post. I'm also publishing peer-reviewed research articles related to this. I think I have some good understanding on this. I was simply saying that I partly disagree with you. And I still do. It's wrong that this should be a surprise to no-one. In fact, I think it is reasonable that it is surprising. It was indeed really unexpected that scaling up such models lead…

One of the few people who publicly predicted the amazing abilities of the GPT-2 successors was Scott Alexander:

https://slatestarcodex.com/2019/02/19/gpt-2-as-step-toward-g...

There may have been others who expected similar miracles, but there certainly weren't many. We can conclude that at least GPT-3 took most by surprise. (I don't know how many people even knew about GPT before GPT-2 came out, it's probably hard to get a sample.)

Heck, even the vast majority of people here on HN wasn't much interested in GPT-3 before ChatGPT, their new cheaper API prices, and GPT-4 came out.

And when we talk about "normal people", they didn't know GPT-3 at all, they only noticed Dall-E 2, somewhat, and finally were caught by surprise when word about ChatGPT spread in a matter of weeks.

Re: Cargo Cult AI

#169

Post by ukuina »

mustacheemperor wrote: post #144
sebzim4500 wrote: post #111

Earlier quoted context omitted.

In my experience, GPT-4 is equally willing to make things up if it doesn't know something but it has so much more knowledge than GPT-3.5 that this happens less often in practice.

On that note, I’ve found that just including in the prompt a request for GPT4 to consider its confidence level in an answer and inform me of that confidence level, to reconsider its answer if its confidence is low, and that accuracy is critically important for the topic of the conversation, also can result in better steering it. I mean, kind of works with humans too. In a high pressure work or school environment, peo…

When I tried this, it made up random, non-repeatable confidence values that had little bearing to the veracity of its statements.

Re: Cargo Cult AI

#170

Post by ukuina »

CharlesW wrote: post #17
celestialcheese wrote: post #15

Earlier quoted context omitted.

Simple things that don't seem impressive to you, but are impressive to others, may be because you haven't experienced the first hand difficulty of doing that thing pre-transformers/GPT. For example, I get _unbelievably_ excited with knowledge extraction and question answering demos on PDFs. Why? Because i've built similar systems for over a decade and know how difficult it is to build on top of messy archival data. N…

Hey, would you mind sharing the tools and other resources you’ve found helpful? I’m really interested in trying this but am not sure where to start.

Here's a packaged solution!

https://github.com/imartinez/privateGPT

Post reply on HN