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

How does GPT obtain its ability? Tracing emergent abilities of language models

yaofu.notion.site

Post reply on HN
Print view View original

191–200 of 205 posts

Re: How does GPT obtain its ability? Tracing emergent abilities of language models

#191

Post by adamsmith143 »

nathias wrote: post #152
roywiggins wrote: post #32

Earlier quoted context omitted.

Most knowledge sources don't make up totally fictional citations to nonexistent sources. Or, if they do, nobody uses them for anything serious. Even Wikipedia citations will get removed if they point to URLs that never existed.

if we focus on the best sources, even in studies a lot of research can't be replicated, and if we focus on te most common ones like newspapers and tv, I'd say most of it is made up or might as well be

That's just a bad take and it doesn't excuse the problems with GPT.

Re: How does GPT obtain its ability? Tracing emergent abilities of language models

#192

Post by aeternum »

ajuc wrote: post #151
aeternum wrote: post #27

Earlier quoted context omitted.

Whose knowledge is trustworthy? We've somehow come to associate certain institutions or scientific authorities with truth when that is about the furthest from real science: "Have no respect whatsoever for authority; forget who said it and instead look what he starts with, where he ends up, and ask yourself, Is it reasonable?" -Richard P. Feynman "One of the great commandments of science is, "Mistrust arguments from a…

I, too, always recreate double blind experiments before I take the drugs my doctor gives me :) I also double-check the transistors in my computer work correctly before I run any code on them, and of course I re-derive the physics to be able to do that :) In practice you are an expert in a very small domain (if any) and in all the other domains you have no choice but to accept somebody's authority.

That's good, I don't go quite as far, but do try to consult multiple independent sources.

Doctors have been known to overprescribe things like Benzos, and opioids from time to time.

I also just use tools like a RAM diagnostic that can check large numbers of transistors at once. I imagine you're quite good at QM after all that practice applying the wave equation though. Impressive!

Re: How does GPT obtain its ability? Tracing emergent abilities of language models

#193

Post by anon291 »

yunwal wrote: post #123
anon291 wrote: post #111

Earlier quoted context omitted.

You're looking for the paper 'attention is all you need'. Gpt is not a bayesian next word chooser. It does something different.

Not really. Attention is all you need describes a new mechanism used in transformer networks, but the model is still a Bayesian word chooser

What does a Bayesian word chooser mean to you?

Re: How does GPT obtain its ability? Tracing emergent abilities of language models

#194

Post by anon291 »

hakuseki wrote: post #113
anon291 wrote: post #111

Earlier quoted context omitted.

You're looking for the paper 'attention is all you need'. Gpt is not a bayesian next word chooser. It does something different.

I think that's not a bad summary, though? Perhaps you would say it is a probabilistic next-token chooser, but that just seems like a very minor distinction.

Probabilistic and bayesian are not identical things. Moreover, GPT the deep-learning model is not a probabilistic next-token chooser. You can envision many different ways to choose the next word based on GPT output. OpenAI's API for GPT is a probabilistic word chooser paired along with GPT. But GPT is the model. It generates a set of probability distributions for the next word, not using a Bayesian process but something entirely different. GPT takes a vector space representation of a sentence and projects it onto some space (we'll call it GPTThink) and then re-projects that space to a new vector space. Then it uses softmax to turn that vector space into a probability distribution. That's not a Bayesian process.

Re: How does GPT obtain its ability? Tracing emergent abilities of language models

#195

Post by Digory »

anon291 wrote: post #194
hakuseki wrote: post #113

Earlier quoted context omitted.

I think that's not a bad summary, though? Perhaps you would say it is a probabilistic next-token chooser, but that just seems like a very minor distinction.

Probabilistic and bayesian are not identical things. Moreover, GPT the deep-learning model is not a probabilistic next-token chooser. You can envision many different ways to choose the next word based on GPT output. OpenAI's API for GPT is a probabilistic word chooser paired along with GPT. But GPT is the model. It generates a set of probability distributions for the next word, not using a Bayesian process but someth…

Better! The last sentence still sounds like "magic," but this is getting closer to my mental comprehension of how you get from BASIC and Python to GPT.

Re: How does GPT obtain its ability? Tracing emergent abilities of language models

#196

Post by elcomet »

PartiallyTyped wrote: post #175
elcomet wrote: post #135

Earlier quoted context omitted.

This can clearly be guessed from a search as well. Popularity can be well defined, and in the case of Obama, there is clearly one much more popular than the others.

The model still needs to infer from the sentence the entity to look it up. It is also the case that this is a relatively simple example as 'Obama' refers to a single class of entities and there is not a lot of ambiguity around resolution of class, only resolution of specific entity. Take this sentence: > When was KitKat released? I could refer to the sweet, or the Android OS. Vastly different classes, and the model h…

And that's perfectly fine. Humans have exactly the same problem. They will get this wrong, and you will reply "no, I'm talking about the android version". Language is ambiguous so we cannot expect machines to get it right all the time.

Re: How does GPT obtain its ability? Tracing emergent abilities of language models

#197

Post by moyix »

astrange wrote: post #129
typon wrote: post #119

Earlier quoted context omitted.

It's not storing any text? The weights are floating point numbers - the "text" is in some extremely high dimensional embedding space.

Of course it's storing text. GPT was trained for less than one epoch; they just continually throw new text in there and it mostly just remembers it (= learns it = compresses it). It's not simply "a high dimensional embedding" because words aren't differentiable; you'll get different words if you round off your "coordinates". If you go to https://beta.openai.com/playground/ and prompt it "Read me the book Alice in Won…

GPT's compression of text is a model of probabilities for the next token in a sequence, where a token is a bit of text from a vocabulary of ~52,000. You can definitely reduce the precision of the parameters that determine that model without hurting the model's overall accuracy much (consider truncating a probability like 98.0000001221151240690% to 98.0%).

Empirically, people have quantized the weights of language models down to INT4 with very little loss in accuracy; see GLM-130B: https://arxiv.org/abs/2210.02414

Re: How does GPT obtain its ability? Tracing emergent abilities of language models

#198

Post by snickerer »

PartiallyTyped wrote: post #175
elcomet wrote: post #135

Earlier quoted context omitted.

This can clearly be guessed from a search as well. Popularity can be well defined, and in the case of Obama, there is clearly one much more popular than the others.

The model still needs to infer from the sentence the entity to look it up. It is also the case that this is a relatively simple example as 'Obama' refers to a single class of entities and there is not a lot of ambiguity around resolution of class, only resolution of specific entity. Take this sentence: > When was KitKat released? I could refer to the sweet, or the Android OS. Vastly different classes, and the model h…

I hope nobody ever releases the famous Kitkat Club in Berlin from its chains. Because there are not so many.

My experience with ChatGPT is that it gets what I mean very well from the context.

Re: How does GPT obtain its ability? Tracing emergent abilities of language models

#199

Post by PartiallyTyped »

elcomet wrote: post #196
PartiallyTyped wrote: post #175

Earlier quoted context omitted.

The model still needs to infer from the sentence the entity to look it up. It is also the case that this is a relatively simple example as 'Obama' refers to a single class of entities and there is not a lot of ambiguity around resolution of class, only resolution of specific entity. Take this sentence: > When was KitKat released? I could refer to the sweet, or the Android OS. Vastly different classes, and the model h…

And that's perfectly fine. Humans have exactly the same problem. They will get this wrong, and you will reply "no, I'm talking about the android version". Language is ambiguous so we cannot expect machines to get it right all the time.

I do agree with you that it is fine, what I was getting at was that there needs to be a way to measure uncertainty in a manner that is robust to unbalanced distributions or context drifting.

Re: How does GPT obtain its ability? Tracing emergent abilities of language models

#200

Post by abi »

Digory wrote: post #108

This just confirms I'm not minimally competent in this conversation. Is there a "college freshman" explainer? GPT seems to be doing something incredibly different than prior AI. Is it really a Bayesian "next word" chooser at incredible scale?

I've been enjoying Andrej Karpathy's YouTube series on neural networks: https://www.youtube.com/playlist?list=PLAqhIrjkxbuWI23v9cThs...

It starts from absolute basics and goes slowly. I've only watched about half of it and it has already helped me understand a lot of AI concepts that I see frequently spoken about.

Post reply on HN