5 ms·
I did not read the paper (just like most people here), but by the title — “On the Dangers of Stochastic Parrots: Can Language Models Be Too Big?” — it does not
by hezag 6y ago
I did not read the paper (just like most people here), but by the title — “On the Dangers of Stochastic Parrots: Can Language Models Be Too Big?” — it does not look like the CO2 emissions thing is the main topic of this research.
BTW, "Stochastic Parrots" is a very descriptive name for the problem
> Moreover, because the training datasets are so large, it’s hard to audit them to check for these embedded biases. “A methodology that relies on datasets too large to document is therefore inherently risky,” the researchers conclude. “While documentation allows for potential accountability, [...] undocumented training data perpetuates harm without recourse.”
Since these models are being applied in a lot of fields that directly affects the life of millions of people, this is a very important and underdiscussed problem.
I really want to read the paper.
- sanxiyn 6y ago> Since these models are being applied in a lot of fields that directly affects the life of millions of people In particular, it is being applied right now to rank Google search results, and probably responsible for lots and lots of Google's profit. You should be skeptical of Google's appraisal of the paper that is material to Google's profit.
- visarga 6y agoI bet you don't need to evaluate the NLP model when you have direct access to the end result, Google Search.
- visarga 6y ago> BTW, "Stochastic Parrots" is a very descriptive name for the problem Meh, both parrots and language models are inferior to humans in producing language, but one is more useful than the other. And real parrots are also stochastic, like all living things.
- basicneo 6y agoAnalogies aren't meant to withhold careful scrutiny. They're suggestive.
- zepto 6y agoReal parrots are clearly more useful. After the stochastic parrots have caused the collapse of civilization, we can eat the real parrots.
- YeGoblynQueenne 6y ago>> And real parrots are also stochastic, like all living things. I don't understand how living things are "stochasic". Can you please elaborate on the matter?
- thu2111 6y agoWhy? It sounds like ideologically driven circular reasoning. If you train an AI on the largest dataset it's possible to obtain then you have, almost by definition, done the most you can to avoid bias of any sort: the model will learn the most accurate representation of reality it can given the data available. Gebru is the type of person who defines "bias" as anything that isn't sufficiently positive towards people who look like herself, not the usual definition of a deviation from reality as exists. Having encountered AI "fairness" and "bias" papers (words quoted because the words aren't used with their dictionary definitions), it's not even clear to me they should count as research at all, let alone be worth reading. They take as the starting premise that anything a model learns about the world that is politically incorrect is a bug, and go downhill from there.
- kevinventullo 6y agoIf you train an AI on the largest dataset it's possible to obtain then you have, almost by definition, done the most you can to avoid bias of any sort All politics aside, this is not even true for toy ML problems. If I’m trying to do digit recognition and “all the data I can find” is a billion hand-written 0’s and a million hand-written 1’s through 9’s, naively training on that data will yield a model that’s pretty close to guessing 0 every time.
- thu2111 6y agoWe're talking about tech firms that have access to the entire internet and use it. Hypothetical examples involving imaginary datasets that nobody would use don't prove anything relevant. And note that my argument is not about whether you actually avoid bias, it's about whether you've done the most you can do to avoid it. If you used all the data you've got, then you've done the most you can, even if for some reason the data you've got isn't any good.
- d_e_solomon 6y ago> We're talking about tech firms that have access to the entire internet and use it. I think you're trying to say that a large enough dataset will be free of bias. I don't see how that follows. If I train a model on home mortgage decisions, I will replicate the bias on that currently exists on that dataset - https://news.northwestern.edu/stories/2020/01/racial-discrimination-in-mortgage-market-persistent-over-last-four-decades/ https://news.northwestern.edu/stories/2020/01/racial-discrim... - unless there are conscientious choices to reduce that bias. Researchers in ethics in ML are specifically trying to enable tech companies to do a better job of not replicating bias and justifiably point out where that is occurring. Third, I would argue that applying an ML model to do something faster if it replicates the bias of a previously human decision is even worse. The bias has taken the human element completely out and systematized the bias and made it possible with even less friction.
- Veedrac 6y agoThe first page was leaked. The environmental angle was a significant part of it, particularly the claim that environmental and financial costs ‘doubly punishes marginalized communities’.
- throwaway_hare 6y agoThe paper appears to be here: https://gofile.io/d/WfcxoF https://gofile.io/d/WfcxoF (source: https://www.reddit.com/r/MachineLearning/comments/k77sxz/d_timnit_gebru_and_google_megathread/gepcliq/ https://www.reddit.com/r/MachineLearning/comments/k77sxz/d_t...)