4 ms·

As an ML researcher, simple back of the envelope math shows you that the scaling hypothesis is totally wrong. We will never reach human-level AI by simply scali

As an ML researcher, simple back of the envelope math shows you that the scaling hypothesis is totally wrong. We will never reach human-level AI by simply scaling what we have today. GPT-3 has 10^11 parameters and needs 10^14 bytes of training data. Averaged performance on a bunch of benchmarks is 40-50% depending on what kind of prompts you provide: https://res.cloudinary.com/dyd911kmh/image/upload/f_auto,q_auto:best/v1598020447/gpt3-3_krvb14.png https://res.cloudinary.com/dyd911kmh/image/upload/f_auto,q_a... 10x fewer parameters drops your performance by about 10%. If you just linearly extrapolate that graph, and ML doesn't generally scale linearly, models tend to peter out eventually, you're talking about needing models that are 10^6 or more larger with a similar increase in training data. This is.. starting to be impractical. That's 10^17 or more parameters and 10^20 or more data. And that's assuming the models actually continue to learn. This is also extrapolating with an average. Datasets in machine learning are not difficulty calibrated at all. We have no idea how to measure difficulty. So this extrapolation is being driven by the easier datasets, and it won't saturate the hard ones. For example, GPT-3 makes a lot of systematic errors and there are plenty of benchmarks where it just isn't very good regardless of how many parameters it has. Our understanding of what intelligence is in the first place is the biggest hurdle here. This is why we can't benchmark systems. Why we can't come up with a benchmark, where performance on that benchmark means we're x% of the way toward an intelligent system. As systems get better, our benchmarks and datasets get better to keep up with them. So just saying we're going to saturate performance on today's benchmarks with some model that has 10^17 parameters just doesn't mean much at all. We have no guarantee and no reason to expect that doing well on today's benchmarks, even if we invested trillions of dollars would matter in the grand scheme of things. Doesn't mean these models can't be useful. But there's plenty more to do before we can just say "take what we have and invest $1T to scale it up and we'll be good to go".

RandyRanderson 4y ago
Also, we need to show that the data costs, dev costs, training costs, maint, etc, and inference time costs are cheaper than human alternatives for some reasonable loss function, for most tasks. If you look at the FN and FP rates for many tasks, the SoTA Transformer models are all VERY high vs humans and inference times are maybe only 10x. At 100x or 1000x the transistors, data, etc and even at 1/10 the loss, ML solutions are likely not competitive for many tasks.
- sanxiyn 4y ago
  The article is about we have no fucking idea how things like 1/10 loss translate to capability and that's why we can't make statements like "even at 1/10 loss ML solutions are likely not competitive for many tasks".
cavisne 4y ago
Just git clone mingpt and set the parameters to 100 decillion
graboid 4y ago
I got a question. How can researchers who are not working at the largest tech companies even "compete" with models like GPT-3? From what I understand, these currently best performing models need an amount of training data and specialized hardware that is not obtainable for 99% of ML researchers. If you think you came up with a better architecture for a language model, one that could in theory beat GPT-3 on benchmarks, wouldn't you face the problem that you cannot actually prove that it would perform better?
- light_hue_1 4y ago
  This is a very reasonable question! And one that researchers think about regularly. The reality is that you are right. There are simply questions that are inaccessible with the resources available at universities. We think about the questions that we can ask that don't compete or we find ways to compete, like having collaborations with corporations that have deep resources. Never mind access to engineers, which we almost totally lack in academia. But at the end of the day. Yeah, if a random researcher got everything right about GPT, they couldn't have published it first, because they couldn't even have tested out a proof of concept. This is in part why people move to industry.
matusp 4y ago
Yep, the data requirements are especially condemning. We have recently scaled from using Wikipedia as a training corpus to basically the entire Internet. And it helped a lot. But where are we going to scale next? There is simply not enough useful data for the models to learn. People generate a lot of text and language, but most of it is repetitive and not really that useful. The fact that GPT-3 is already two years old and nothing got better than that is telling. If the road to AGI was as simple as scaling, we would have bigger and better models by now. I suspect that even the top labs have found out that GPT-3 is probably near the limits of what they can realistically achieve with deep learning LMs.
- gwern 4y ago
  > to basically the entire Internet. Please take a few moments to think about how this is not even remotely close to the 'entire Internet'. Let me point out that Common Crawl doesn't even contain Twitter. That's how incredibly incomplete it is! Google Books estimates there's something like >100m books in existence, of which GPT-3 has seen part (it didn't even do 1 epoch, and like 95% of it was purely English anyway, and it omits most of Github which is why Codex has to be so heavily trained, and...). There are millions of books and papers published every year, endless billions of social media posts and comments in thousands of languages not just English, and on and on and on. The world is a very big place. And this is just text. Text is great, don't get me wrong (it may even have some special properties compared to other modalities), but I'm baffled that in a period where things like Imagen are hitting the front page every few days that I need to point out that there are things other than text, like, y'know, images and video and audio.

RandyRanderson 4y ago

Also, we need to show that the data costs, dev costs, training costs, maint, etc, and inference time costs are cheaper than human alternatives for some reasonable loss function, for most tasks. If you look at the FN and FP rates for many tasks, the SoTA Transformer models are all VERY high vs humans and inference times are maybe only 10x. At 100x or 1000x the transistors, data, etc and even at 1/10 the loss, ML solutions are likely not competitive for many tasks.

sanxiyn 4y ago
The article is about we have no fucking idea how things like 1/10 loss translate to capability and that's why we can't make statements like "even at 1/10 loss ML solutions are likely not competitive for many tasks".

cavisne 4y ago

Just git clone mingpt and set the parameters to 100 decillion

graboid 4y ago

I got a question. How can researchers who are not working at the largest tech companies even "compete" with models like GPT-3? From what I understand, these currently best performing models need an amount of training data and specialized hardware that is not obtainable for 99% of ML researchers. If you think you came up with a better architecture for a language model, one that could in theory beat GPT-3 on benchmarks, wouldn't you face the problem that you cannot actually prove that it would perform better?

light_hue_1 4y ago
This is a very reasonable question! And one that researchers think about regularly. The reality is that you are right. There are simply questions that are inaccessible with the resources available at universities. We think about the questions that we can ask that don't compete or we find ways to compete, like having collaborations with corporations that have deep resources. Never mind access to engineers, which we almost totally lack in academia. But at the end of the day. Yeah, if a random researcher got everything right about GPT, they couldn't have published it first, because they couldn't even have tested out a proof of concept. This is in part why people move to industry.

matusp 4y ago

Yep, the data requirements are especially condemning. We have recently scaled from using Wikipedia as a training corpus to basically the entire Internet. And it helped a lot. But where are we going to scale next? There is simply not enough useful data for the models to learn. People generate a lot of text and language, but most of it is repetitive and not really that useful. The fact that GPT-3 is already two years old and nothing got better than that is telling. If the road to AGI was as simple as scaling, we would have bigger and better models by now. I suspect that even the top labs have found out that GPT-3 is probably near the limits of what they can realistically achieve with deep learning LMs.

gwern 4y ago
> to basically the entire Internet. Please take a few moments to think about how this is not even remotely close to the 'entire Internet'. Let me point out that Common Crawl doesn't even contain Twitter. That's how incredibly incomplete it is! Google Books estimates there's something like >100m books in existence, of which GPT-3 has seen part (it didn't even do 1 epoch, and like 95% of it was purely English anyway, and it omits most of Github which is why Codex has to be so heavily trained, and...). There are millions of books and papers published every year, endless billions of social media posts and comments in thousands of languages not just English, and on and on and on. The world is a very big place. And this is just text. Text is great, don't get me wrong (it may even have some special properties compared to other modalities), but I'm baffled that in a period where things like Imagen are hitting the front page every few days that I need to point out that there are things other than text, like, y'know, images and video and audio.