And now we need to do a contest and compare how humans will answer on this questions, and whose answer is closer to the truth. Take my bet - humans will lose or will be on the same level of guessing.
I asked GPT-NeoX-20B a hundred arithmetic questions
21–30 of 235 posts
Re: I asked GPT-NeoX-20B a hundred arithmetic questions
#22Re: I asked GPT-NeoX-20B a hundred arithmetic questions
#23Earlier quoted context omitted.
Poor performance is more likely due to how transformer neural networks view numbers. It memorises them like words instead of modeling their numerical structure. Thus even if it’s seen the number 3456 and 3458, it knows nothing of 3457. Totally different embedding. It’s like a kid memorising a multiplication table instead of learning the more general principle of multiplication (related: this illusion is why big model…
Take a look at this paper: Deep Symbolic Regression for Recurrent Sequences https://arxiv.org/abs/2201.04600 If you look at embedding visualization it is very clear that the model learns order of numbers. (Interactive demo: http://recur-env.eba-rm3fchmn.us-east-2.elasticbeanstalk.com... ) There is also: Grokking: Generalization Beyond Overfitting on Small Algorithmic Datasets https://arxiv.org/abs/2201.02177 Again, l…
Modulo 97 (the arxiv paper). That is what they do.
It is quite easy to grok operations modulo 97.
Re: I asked GPT-NeoX-20B a hundred arithmetic questions
#24What? You think this is poor performance?
This totally blows my mind. I would never have guessed that GPT could get ANY of these right.
I mean, is there a data point in the dataset used to train where you can read 2241 + 19873 = 22114? Quite unlikely...
And those multiplications. It's consistently getting the number of digits right and the first two numbers correct. How the hell does this happen?
Sure, it's sometimes way off. But generally it is in the right ballpark.
I certainly think people should look into what's happening inside the model.
Re: I asked GPT-NeoX-20B a hundred arithmetic questions
#25What? You think this is poor performance? This totally blows my mind. I would never have guessed that GPT could get ANY of these right. I mean, is there a data point in the dataset used to train where you can read 2241 + 19873 = 22114? Quite unlikely... And those multiplications. It's consistently getting the number of digits right and the first two numbers correct. How the hell does this happen? Sure, it's sometimes…
It's unlikely that "2241 + 19873 = 22114" specifically is in the dataset, but very likely that there are many expressions equivalent to that expression in the dataset, and we've just picked one of those.
Imagine someone watching every lottery draw and after each draw going "Wow! the chances of those exact numbers coming up in that order are atronomical!"
Re: I asked GPT-NeoX-20B a hundred arithmetic questions
#26Can we get a title edit? GPT3 is over 8x the parameter count.
Re: I asked GPT-NeoX-20B a hundred arithmetic questions
#27This just shows that this model did not learn anything. Humans do not see billions of examples to add numbers. We see just few and can apply learned notation and procedures to infinity with 100% precision. GPT-3 learned mathematical intuition. Humans can hardly learn multiplication table over months and repetitions of same examples, and that table hardly matters at all. GPT-3 is just plainly wrong objective they tryi…
The vast majority of humans don't just see a few examples and figure it out. They're taught an algorithm. Eventually they may also come up with another algorithm, but they're taught one first.
They also don't have "100% precision". Many, many humans are incredibly bad at math, and even the ones that are good at it often make mistakes.
Re: I asked GPT-NeoX-20B a hundred arithmetic questions
#28Thank you for this. Technically it's not GPT-3, but GPT-NeoX-20B, although they are based on a similar architecture. The poor performance is most likely due to not having a large database of math problems to draw from. Github, for example, is part of the dataset that is used to train both GPT-3 and GPT-Neo variants, which is partly why they can generate meaningful code (sometimes). I wonder how a model finetuned for…
I went and checked, it turns out for this version Eleuther-AI has in fact included math problems [1]. So my earlier comment is partly incorrect. [1] http://eaidata.bmk.sh/data/GPT_NeoX_20B.pdf
And isn't it trivial to generate lots of correct sample data ? :-)
Re: I asked GPT-NeoX-20B a hundred arithmetic questions
#29What? You think this is poor performance? This totally blows my mind. I would never have guessed that GPT could get ANY of these right. I mean, is there a data point in the dataset used to train where you can read 2241 + 19873 = 22114? Quite unlikely... And those multiplications. It's consistently getting the number of digits right and the first two numbers correct. How the hell does this happen? Sure, it's sometimes…
Maybe this is an example of where you need an "extra specialized skill"(arithmetic) vs the general and semi-ambiguous-skill of language+conversation.
GPT-3 is "good with conversation (language)"
GPT-3 now needs a "sub-nn-model" to do the very 'specialized skill called math'
*GPT-3 Should 'learn' to recognize which questions should be delicate to a submodel.
Re: I asked GPT-NeoX-20B a hundred arithmetic questions
#30What? You think this is poor performance? This totally blows my mind. I would never have guessed that GPT could get ANY of these right. I mean, is there a data point in the dataset used to train where you can read 2241 + 19873 = 22114? Quite unlikely... And those multiplications. It's consistently getting the number of digits right and the first two numbers correct. How the hell does this happen? Sure, it's sometimes…
It's unlikely that "2241 + 19873 = 22114" specifically is in the dataset, but very likely that there are many expressions equivalent to that expression in the dataset, and we've just picked one of those. Imagine someone watching every lottery draw and after each draw going "Wow! the chances of those exact numbers coming up in that order are atronomical!"
> there are many expressions equivalent to that expression in the dataset
What do you mean by this?