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

Talking About Large Language Models

arxiv.org

Post reply on HN
Print view View original

51–60 of 158 posts

Re: Talking About Large Language Models

#51

Post by marcinzm »

radford-neal wrote: post #44
marcinzm wrote: post #34

Earlier quoted context omitted.

> If your criteria for whether a model is good is "it fooled a 9th grader" well, I don't know what to tell you. Fooling a 9th grader is amazing. That's a pretty well formed human being right there except with less life experience. Fundamentally no different from you in general reasoning terms except on a smaller set of information. So fooling you is merely a question of model size.

"Fool" is the operative word here. ChatGPT is quite capable of producing very plausible sounding text about biology that is totally incorrect. See, for example, the example in my comment at https://www.lesswrong.com/posts/28XBkxauWQAMZeXiF/?commentId...

You're basically complaining that a single model doesn't have full knowledge of every single area of all of human knowledge. It's got decent knowledge of most areas including programming with probably better overall knowledge than a high school student. That's downright amazing and probably more knowledge than any single human actually has. The rest is likely a matter of improvement along the same lines versus some radical redesign.

Re: Talking About Large Language Models

#52

Post by anyonecancode »

RosanaAnaDana wrote: post #46
anyonecancode wrote: post #29

Earlier quoted context omitted.

Increasingly I don't think the question of "what is intelligence" is so useful or relevant here. It feels a bit like arguing over whether the "artificial horse" that started appearing at the end of the 19th/beginning of the 20th C were actually horses. Cars weren't, and still aren't, but that misses the point. AI isn't intelligent, and never will be, and I don't think that matters all that much.

I think I agree in sentiment, and I'm wondering what your ake is on the article/ current discussions article. I guess my premise is that I don't think we have a useful enough definition of intelligence because the ones I see people writing articles on seem to be dependent or defined by agency, and specifically humanish forms of agency. So I guess your point would be "these systems aren't intelligent, but that's not r…

I think the question of "what is intelligence" is an interesting one, and technology (especially computer technology) gives us some interesting angles to look at it, but I think it dominates the conversation disproportionately to its importance. Things like ChatGPT, and the technologies they presage, will absolutely have a significant impact on society, economics, etc, but getting tangled up in questions of "what is intelligence" impede rather than help us to think through these implications and prepare for them.

Put another way -- I do not believe the future holds Blade Runner replicants. If we're not careful, though, it does hold Blade Runner corporations. While, philosophically, it's interesting to ask if androids dream of electric sheep, that question isn't very helpful in trying to nudge the future in a more utopic rather than dystopic direction.

Re: Talking About Large Language Models

#53

Post by TeMPOraL »

b3morales wrote: post #42
TeMPOraL wrote: post #24

Earlier quoted context omitted.

Sure, when the topic matters or I need to study it in depth, I can still go to Wikipedia or PubMed or Arxiv. But there are plenty of searches one does that are trivial, or serve to illuminate the problem space, and cover topics that in which I can rely on common sense to correct wrong advice. And the issue with non-technical topics, the kind applicable to mass audience - like e.g. cooking or parenting or hygiene - ar…

But if so this isn't because of its nature (the fact that it's an LLM), but because of its inputs. An LLM fed the same bullshit pseudo articles you refer to would likewise spit out more bullshit. If ChatGPT works it's because its sources have been carefully curated.

Fair. But the practical reality right now is that ChatGPT delivers useful results without the noise, whereas normal web search does not. It blows the web out of the water when it comes to value to effort ratio of generic web searches. It won't last forever, but I'm enjoying it for as long as I can.

Re: Talking About Large Language Models

#54

Post by radford-neal »

marcinzm wrote: post #51
radford-neal wrote: post #44

Earlier quoted context omitted.

"Fool" is the operative word here. ChatGPT is quite capable of producing very plausible sounding text about biology that is totally incorrect. See, for example, the example in my comment at https://www.lesswrong.com/posts/28XBkxauWQAMZeXiF/?commentId...

You're basically complaining that a single model doesn't have full knowledge of every single area of all of human knowledge. It's got decent knowledge of most areas including programming with probably better overall knowledge than a high school student. That's downright amazing and probably more knowledge than any single human actually has. The rest is likely a matter of improvement along the same lines versus some r…

Well, I agree that it's amazing - it almost always produces grammatical output, for instance. But it's not a reliable way of obtaining knowledge. One should not, in particular, try to learn about biology by asking ChatGPT questions. It often produces made-up stuff that is just wrong. And it's very confidently wrong, with the output often coming across like someone barely concealing their contempt that you might doubt them.

It may or may not be fixable without radical redesign. The underlying training objective of mimicking what humans might say may be too at variance with an objective of producing true statements.

Re: Talking About Large Language Models

#56

Post by visarga »

nathan_compton wrote: post #10
axg11 wrote: post #9

Earlier quoted context omitted.

I disagree that LLMs are overhyped, but it's very subjective. Are current LLMs a few steps from AGI? No. Will LLMs change the computing landscape? Yes, I believe they will. ChatGPT, without any major changes, is already the best tool out there for answering programming questions. Nothing else comes close. I can ask it to provide code for combining two APIs and it will give useful and clean output. No need to trudge t…

It can reproduce a statistically plausible paragraph, certainly. But there is a great deal more to research than producing statistically plausible paragraphs. It doesn't _understand_ anything! I've actually worked on a project where there have been attempts to use GPT like models to summarize scientific results and the problem is it gets shit wrong all the time! You have to be an expert to separate the wheat from the…

The problem is that we need to pair generative models with verification systems. We have the models, but no verification yet. Fortunately code and math are easier to verify. Some things require simulation. In other cases you can substitute an ensemble of solutions & picking the most frequent answer as consistency based verification. But for each domain we need to create verifiers and that will take some time.

The good thing is that we'll be able to generate training data with our models by filtering the junk with the verifiers. Then we can retrain the models. It's important because we are getting to the limit of available training data. We need to generate more data, but it's worthless unless we verify it. If we succeed we can train GPT-5. Human data will be just 1%, the race is on to generate the master dataset of the future. I read in a recent paper that such a method was used to improve text captions in the LAION dataset. https://laion.ai/blog/laion-5b/

Re: Talking About Large Language Models

#57

Post by lachlan_gray »

gamegoblin wrote: post #26
lachlan_gray wrote: post #21

Earlier quoted context omitted.

This makes me wonder if GPT could be any good at defining its own control flow. E.g. asking it to to write a python script that uses control structures along with calls to GPT to synthesize coherent content. Maybe it could give itself a kind of working memory.

Libraries such as https://github.com/hwchase17/langchain allow for easy programmatic pipelines of GPT "programs". So you could imagine taking a few hundred of these programs written by humans for various tasks, as are sure to come into existence in the next year or two, then adding those programs to the training data and training a new GPT that knows how to write programs that call itself.

Wow. Thank you for sharing. I had no idea there was a scene for this.

Re: Talking About Large Language Models

#58

Post by plutonorm »

RosanaAnaDana wrote: post #11

I like the discussion, but this article 'feels' like more Luddite goalpost moving, and is reflective of a continuous sentiment I feel strains so much of the conversation around intelligence, agentism, and ai going on today. I think that because we lack a coherent understanding of what it means to be intelligent at an individual level, as well as what it means to be an individual, we're missing much of the point of wh…

I 100% agree. I would also add that most of the arguments are driven by emotion. The truth is that we dont know what intelligence means and we dont know what kinds of system have intelligence. The only tools we have to measure intelligence are those designed for humans. When we test the machines they do better than terribly and they are improving very quickly. There is no possible logical argument you can put forward against their intelligence in the face of this evidence from these human tests - because we cannot define intelligence in any other way than these tests. Claims against intelligent machines always boil down to 'obviously they aren't' and the arguments have have to be this shallow simply because they have no firm footing from which to base their argument.

Re: Talking About Large Language Models

#59

Post by gillesjacobs »

I am NLP researcher who volunteers for peer review often and the anthropomorphisms in papers are indeed very common and very wrong. I have to ask authors to not ascribe cognition to their deep learning approaches in about a third of the papers I review.

People do this because mirroring cognition to machine learning lends credence that their specific modeling mechanism mimicks human understanding and so is closer "to the real thing". Obviously this is almost never the case, unless they explicitly use biomimetic methods in which case they are often outperformed by non-biomimetic state-of-the-art approaches.

Thanks OP for giving me citation ammo to refer to in my obligatory "don't humanise AI" section of reviews. (It is so common I copy paste this section from a template).

Re: Talking About Large Language Models

#60

Post by gillesjacobs »

gillesjacobs wrote: post #59

I am NLP researcher who volunteers for peer review often and the anthropomorphisms in papers are indeed very common and very wrong. I have to ask authors to not ascribe cognition to their deep learning approaches in about a third of the papers I review. People do this because mirroring cognition to machine learning lends credence that their specific modeling mechanism mimicks human understanding and so is closer "to…

Not to shoot across the bow of CS Engineers but the trend I spot (tentatively) is that it is pure computer science folk that most often do this. In NLP you have a mix of people coming from pure CS and signal processing (the latter esp. in speech processing) and others who come from linguistics or other humanities.

The CS people seem all too happy to humanise computation, probably because they had less direct teaching regarding the cognitive mechanisms behind cognition and language production.

Post reply on HN