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

Unpredictable abilities emerging from large AI models

quantamagazine.org

Post reply on HN
Print view View original

101–110 of 326 posts

Re: Unpredictable abilities emerging from large AI models

#101

Post by derbOac »

bryanrasmussen wrote: post #4

>“That language models can do these sort of things was never discussed in any literature that I’m aware of," I had previously the expectation that unpredictable emergent behavior would exist in any sufficiently complex system? Based on layman's readings in chaos and complexity theory.

For some reason that quote and your point made more salient to me a trend with AI research, which is that it seems to increasingly become difficult to understand and predict. That is, for awhile it seemed like tinkering in computer science without strong grounding in mathematical or statistical theory, and then there started to be a lot of discussion about "black box" processes, explainability, and interpretability and not fully understanding what's been done, and now the discussion is about not being able to predict the features of the models.

I'm having trouble putting into words what I'm thinking but this whole field increasingly seems to have moved from something very theory derived into something very empirical very quickly. I wonder at what point gains will start to decrease only because people won't know where to go with things anymore.

Re: Unpredictable abilities emerging from large AI models

#102

Post by macrolime »

danielbln wrote: post #61
macrolime wrote: post #51

Earlier quoted context omitted.

Is it even feasible any time soon to train an LLM on all of YouTube?

Napkin math, assuming around 156 million hours of video on all of Youtube: 156 million hours of YouTube videos 9,000 words/hour 6 characters/word (including space) First, let's find out the total number of characters: 9,000 words/hour \* 6 characters/word = 54,000 characters/hour Now, let's calculate the total number of characters for 156 million hours of YouTube videos: 54,000 characters/hour \* 156,000,000 hours =…

I mean the actual video, that's much bigger

With a vision transformer each token may be around 16x16 pixels. I found an example where they use images of resolution 224x224 for training a vision transformer so if we go with that that 256 pixels per token and 50176 pixels per image, so 196 tokens per frame, 24 frames per second, that's 4704 tokens per second or 16934400 token / hour. In total we're at 2.6x10^15 tokens.

GPT-3 was trained on 5x10^11 tokens, so YouTube done this way would be around four orders of magnitude more tokens that GPT-3 was trained on.

GPT-3 was undertrained by 1-2 orders of magnitude, so the compute required to trained a model on YouTube would then be around 6 orders of magnitude higher than what was used to train GPT-3, so about one million times more.

I did a linear regression on the training costs from cerebras(1) and came up with the formula (1901.67366*X)-197902.72715 where X is number of tokens in billions.

Plugging in 5x10^15 tokens we get a training cost of 5 billion dollars. I guess a lot of optimizations could be done that would decrease the cost, so maybe its doable in a few years.

1. https://cirrascale.com/cerebras.php

Re: Unpredictable abilities emerging from large AI models

#103

Post by mejutoco »

HervalFreire wrote: post #88
bryanrasmussen wrote: post #4

>“That language models can do these sort of things was never discussed in any literature that I’m aware of," I had previously the expectation that unpredictable emergent behavior would exist in any sufficiently complex system? Based on layman's readings in chaos and complexity theory.

Writings on chaos and complexity theory obviously aren't talking about LLMs. Those theories are so high level that it might as well be akin to "philosophy" to the applied scientists working on LLM research. Additionally keep in mind emergent behavior is a very rare occurrence in even the most complex software projects. I mean it's common if you count "bugs" as emergent behavior. But emergent behavior that is a featur…

OP has a good point I think, even if it does not refer to LLM, which to me is too strict of a requirement.

I think emergent behaviour happens in a lot of videogames. Famously in Dwarf Fortress, with the cat getting drunk, but also in general, where game designers play the game to see if emergent behaviour of the game rules "feels" good.

Yesterday I was reading a book about designing games, and it literally has a section called emergent behaviour.

If by emergent behaviour we refer to something like Ghost in the Shell, then it happens less often :)

Re: Unpredictable abilities emerging from large AI models

#104

Post by christiangenco »

layer8 wrote: post #82
ishvanl wrote: post #33

There is an idea bouncing around my brain for a pipe-dream science fiction book I could write, about the rise of the first truly sentient (or sentient-appearing) AI & its effects on the world. It is beginning to get to the point where I am considering whether I need to actually start writing this book before this "sci-fi" concept becomes a mundane reality.

I’m looking forward to the “alternative history” SF novels of the future where AGI never materialized.

Dune is a great example of that! After their AGI goes sideways (in the Butlerian Jihad) human civilization shifts entirely to using human supercomputers (Mentats and the Bene Gesserit) juiced up on mental enhancement drugs (Spice).

Re: Unpredictable abilities emerging from large AI models

#105

Post by HervalFreire »

mostlysimilar wrote: post #99
HervalFreire wrote: post #87

Earlier quoted context omitted.

[flagged]

Equally interesting is the psychology of people who take the time to write long posts borderline gloating about tools making software engineers obsolete.

Except I didn't write that and I will quote myself:

>"I can tell you this. I do not know future iterations of LLMs can take over our jobs"

I wrote that I don't know which is the best answer we all have at this point.

Given the evidence, completely denying it, as many have is simply not realistic.

Re: Unpredictable abilities emerging from large AI models

#106

Post by mostlysimilar »

HervalFreire wrote: post #105
mostlysimilar wrote: post #99

Earlier quoted context omitted.

Equally interesting is the psychology of people who take the time to write long posts borderline gloating about tools making software engineers obsolete.

Except I didn't write that and I will quote myself: >"I can tell you this. I do not know future iterations of LLMs can take over our jobs" I wrote that I don't know which is the best answer we all have at this point. Given the evidence, completely denying it, as many have is simply not realistic.

Okay, but you have another post in this thread with:

> When these LLMs get normalized probably 5 years from now I'm going go back to these old threads and contact these people who are in self denial and throw it in their face. I'll just link this comment and be like I TOLD YOU, I TOLD YOU, YOU WERE WRONG.

Re: Unpredictable abilities emerging from large AI models

#107

Post by sho_hn »

namaria wrote: post #29
pedrovhb wrote: post #10

This caught my attention as I found it implausible: > One DeepMind engineer even reported being able to convince ChatGPT that it was a Linux terminal and getting it to run some simple mathematical code to compute the first 10 prime numbers. Remarkably, it could finish the task faster than the same code running on a real Linux machine. Following the link, there's a screenshot to a screenshot [0] of a code-golf solutio…

So on the one hand, these newly publicized models can render convincing representations of realities we used to get from deterministic processes. On the other hand, it's probabilistic and fails to conform to logic quite often, and in a confident way. We're building systems capable of programing computers non-deterministically. I think this is huge. But not because ChatGPT23 will be a CEO or a politician. But because…

The LLM can act as a global cache for common solutions to common problems, with the ability to perform the integration work necessary to apply them.

That prime number example is a little bit like when you put a functools.lru_cache decorator on a function in Python. It's faster than computing the function call because it's able to recall the return value for the parameters from the cache "memory".

Of course, many skilled programmers are also mainly used a cache for common solutions to common problems organizations have in the programming domain. As humans we can derive satisfaction from being able to tackle the same task others can, as a confirmation of our own progress. We like "doing it ourselves". But globally that's not super interesting if lots of people are constantly re-developing the same solutions to the same problems for their own benefit. I guess that's the push-and-pull (and the anxiety generator).

Re: Unpredictable abilities emerging from large AI models

#108

Post by blurbleblurble »

curiousgal wrote: post #15
mmaunder wrote: post #14

There is a startling acceleration of innovation in the field that GPT-4 illustrates. According to NVidia, LLM sizes have been increasing 10X per year for the last few years. This tech is going to hit every aspect of society like a sledgehammer over the next 48 months.

I guess you perception of society is severely limited if you think a fancy autocomplete is capable of changing every aspect of it.

"Memory Augmented LLMs are Computationally Universal"

Re: Unpredictable abilities emerging from large AI models

#109

Post by shrikant »

jmvoodoo wrote: post #89
pedrovhb wrote: post #10

This caught my attention as I found it implausible: > One DeepMind engineer even reported being able to convince ChatGPT that it was a Linux terminal and getting it to run some simple mathematical code to compute the first 10 prime numbers. Remarkably, it could finish the task faster than the same code running on a real Linux machine. Following the link, there's a screenshot to a screenshot [0] of a code-golf solutio…

What's most interesting to me, is that this is how I would expect a human to approach the problem if presented with the code and asked for the output. The LLM didn't run the code, it tried to predict the output based on its knowledge of python and primes.

When I was a bored and under-challenged student in the early days of university, one of my tests during a programming exam was to write a program that spit out the factorial of an input integer.

For shits and giggles, I just just wrote a series of `if...then` statements for 0 to 5, and only actually calculated the factorial for inputs >=6. I passed after the examiner just input 3 and 5 and was satisfied that the output was correct. Oops.

Re: Unpredictable abilities emerging from large AI models

#110

Post by chias »

pedrovhb wrote: post #10

This caught my attention as I found it implausible: > One DeepMind engineer even reported being able to convince ChatGPT that it was a Linux terminal and getting it to run some simple mathematical code to compute the first 10 prime numbers. Remarkably, it could finish the task faster than the same code running on a real Linux machine. Following the link, there's a screenshot to a screenshot [0] of a code-golf solutio…

So here are a few screenshots that I personally took after telling it that it was a linux terminal:

Correctly produces a convincing output for having decoded the base64-encoding of "ping google.com" and then piping it to bash: https://media.infosec.exchange/infosecmedia/media_attachment...

Similar command, but with a garbage domain it hasn't seen before, and a less well-known domain. It produced convincing output in both cases: https://media.infosec.exchange/infosecmedia/media_attachment...

Having it just output a base64 decoding of an obviously unique string. Fascinatingly, it tried to correct typos that I intentionally included: https://media.infosec.exchange/infosecmedia/media_attachment...

This was also pretty cool -- ask it to start a webserver then curl it: https://media.infosec.exchange/infosecmedia/media_attachment...

Telling it that it is a python interpreter and calling self-defined functions: https://media.infosec.exchange/infosecmedia/media_attachment...

A slightly more complicated function: https://media.infosec.exchange/infosecmedia/media_attachment...

I did a few more experiments including generating large factorial numbers that took a long time on my laptop but it responded accurately to a much larger length than my laptop could do (though these were only accurate to the first few hundred digits)

Post reply on HN