4 ms·

I think the author does a decent job laying out good ways of using the LLMs. If you’re gonna use them, this is probably the way. But he acknowledges the ethica

I think the author does a decent job laying out good ways of using the LLMs. If you’re gonna use them, this is probably the way. But he acknowledges the ethical social issues (also misses the environmental issues https://disconnect.blog/generative-ai-is-a-climate-disaster/ https://disconnect.blog/generative-ai-is-a-climate-disaster/) and then continues to use them anyway. For me the ickiness factor is too much, the benefit isn’t worth it.

j45 2y ago
Efficency in models and specialized hardware just for the computation will likely level things out. Compute Power per watt might be different using say something on a large scale Apple Silicon, compared to the cards.
- myaccountonhn 2y ago
  Very often that just increased efficiency just lead to increased demand. I’m skeptical.
  - j45 2y ago
    You’re welcome to be skeptical. If it’s ok I’d like to both share how I’m navigating my skepticism and also being mindful of the need to keep in perspective other people’s skepticism if it doesn’t offer anything to compare. Why? I have friends who can border on veiled cynicism without outlining what might be in the consideration of skepticism. The only things being looked at are why something is not possible, not a combination. Both can result in a similar outcome. Not having enough time to look into intent enough, it just invalidates the persons skepticism until they look into it more themselves. Otherwise used as a mechanism to try and trigger the world to expend mental labour for free on your behalf. It’s important to ask one’s self if there may be partially relevant facts to determine what kind of skepticism may apply: - Generally, is there a provenance of efficiency improvement both in the world of large scale software and algorithmic optimizations? - Have LLMs become more optimized in the past year or two? (Can someone M1 Max Studio run more and more models that are smaller and better to do the same) - Generally and historically is there provenance in compute hardware optimizations, for LLm type or LLM calculations outright? - Are LLMs using a great deal more resources on average than new technologies preceding it? - Are LLMs using a massive amount of resources in the start similar to servers that used to take up entire rooms compared to today?
AndyNemmity 2y ago
In a just society where private corporations didn't attempt to own everything in existence, there are no ethical social issues in my mind. LLMs just use the commons, and should only be able to be owned by everyone in society. The problem comes in with unaccountable private totalitarian institutions. But that doesn't mean the technology is an ethical social issue, it's the corporations who try to own common things like the means of production that is the problem. Yes, there's the pragmatic view of the society we live in, and the issues that it contains, but that's the ethical issue that we need to address. Not that we can as a society create LLMs based on the work of society.
- RodgerTheGreat 2y ago
  LLMs do not simply use the commons, they are a vehicle for polluting the commons on an industrial scale. If, hypothetically, the ethical problems with plagiarizing creative work to create these models were a non-issue, there would still be massive ethical problems with allowing their outputs to be re-incorporated into the web, drowning useful information in a haze of superficially plausible misinformation.
  - visarga 2y ago
    I don't think you are right. If you test LLM text and random internet text for inaccuracies and utility, you'd probably find more luck with LLM text. For example, if you use a LLM to summarize this whole debate, you would get a decent balanced report, incorporating many points of view. Many times the article generated from the chat thread is better than the original one. Certainly better grounded in the community of readers, debunks claims, represents many perspectives. (https://pastebin.com/raw/karBY0zD https://pastebin.com/raw/karBY0zD)
    - RodgerTheGreat 2y ago
      I am not going to fact-check your sludge for you.
      - 2y ago

j45 2y ago

Efficency in models and specialized hardware just for the computation will likely level things out. Compute Power per watt might be different using say something on a large scale Apple Silicon, compared to the cards.

myaccountonhn 2y ago
Very often that just increased efficiency just lead to increased demand. I’m skeptical.
- j45 2y ago
  You’re welcome to be skeptical. If it’s ok I’d like to both share how I’m navigating my skepticism and also being mindful of the need to keep in perspective other people’s skepticism if it doesn’t offer anything to compare. Why? I have friends who can border on veiled cynicism without outlining what might be in the consideration of skepticism. The only things being looked at are why something is not possible, not a combination. Both can result in a similar outcome. Not having enough time to look into intent enough, it just invalidates the persons skepticism until they look into it more themselves. Otherwise used as a mechanism to try and trigger the world to expend mental labour for free on your behalf. It’s important to ask one’s self if there may be partially relevant facts to determine what kind of skepticism may apply: - Generally, is there a provenance of efficiency improvement both in the world of large scale software and algorithmic optimizations? - Have LLMs become more optimized in the past year or two? (Can someone M1 Max Studio run more and more models that are smaller and better to do the same) - Generally and historically is there provenance in compute hardware optimizations, for LLm type or LLM calculations outright? - Are LLMs using a great deal more resources on average than new technologies preceding it? - Are LLMs using a massive amount of resources in the start similar to servers that used to take up entire rooms compared to today?

AndyNemmity 2y ago

In a just society where private corporations didn't attempt to own everything in existence, there are no ethical social issues in my mind. LLMs just use the commons, and should only be able to be owned by everyone in society. The problem comes in with unaccountable private totalitarian institutions. But that doesn't mean the technology is an ethical social issue, it's the corporations who try to own common things like the means of production that is the problem. Yes, there's the pragmatic view of the society we live in, and the issues that it contains, but that's the ethical issue that we need to address. Not that we can as a society create LLMs based on the work of society.

RodgerTheGreat 2y ago
LLMs do not simply use the commons, they are a vehicle for polluting the commons on an industrial scale. If, hypothetically, the ethical problems with plagiarizing creative work to create these models were a non-issue, there would still be massive ethical problems with allowing their outputs to be re-incorporated into the web, drowning useful information in a haze of superficially plausible misinformation.
- visarga 2y ago
  I don't think you are right. If you test LLM text and random internet text for inaccuracies and utility, you'd probably find more luck with LLM text. For example, if you use a LLM to summarize this whole debate, you would get a decent balanced report, incorporating many points of view. Many times the article generated from the chat thread is better than the original one. Certainly better grounded in the community of readers, debunks claims, represents many perspectives. (https://pastebin.com/raw/karBY0zD https://pastebin.com/raw/karBY0zD)
  - RodgerTheGreat 2y ago
    I am not going to fact-check your sludge for you.
    - 2y ago