3 ms·

My predictions: 1) AI stuff’s overblown. It’ll be a good tool, becoming just another of many, and probably will improve over time, but we’ll find we’re nowhere

My predictions: 1) AI stuff’s overblown. It’ll be a good tool, becoming just another of many, and probably will improve over time, but we’ll find we’re nowhere near as close to creating silicon sentience as some worry we are. 2) The real problem is letting a few megacorps raid the commons—and hell, lots of stuff that’s not really in the commons at all, basically just all of culture—then gate “their” creations behind a paywall (oh, but that they expect us to respect, because that makes sense), and these AI safety folks don’t seem to give a shit about that.

ronsor 2y ago
I agree with this. On (2), I would like to see companies have no rights over models trained on public data. It's very arguable they should be required to release model weights.
- vundercind 2y ago
  Yeah IMO a good outcome would be that training on data you don’t own or license should require release of the model. Is allowed, doesn’t get you something you exclusively own. Bonus points if existing rights assignments aren’t enough to count as a grant of permission for AI training.
  - astrange 2y ago
    I don't think there's any precedent for requiring someone to distribute a work they created. That sounds expensive and could easily be a contract violation for other data they did license.
    - astrange 2y ago
      Hmm, it is sort of how patents work. Copyright registration involves giving some sort of the work to the Library of Congress, but I'm not familiar with that part of it.
hn_throwaway_99 2y ago
> then gate “their” creations behind a paywall (oh, but that they expect us to respect, because that makes sense), and these AI safety folks don’t seem to give a shit about that. I downvoted your comment for this statement, given that's a specific worry discussed at length in this article.
- vundercind 2y ago
  Ah, I figured since they’d been on the board of an org that did exactly that bad thing, on a grand scale, they couldn’t possibly seriously care about it. Maybe they were in very early and so briefly that they weren’t party to that. IDK, article’s paywalled.
jamiek88 2y ago
> and these AI safety folks don’t seem to give a shit about that Because that’s literally not their job or role. Why would enforcing copyright be in any way shape or form their responsibility?
- astrange 2y ago
  Legal safety is (part of) the reason companies have "trust and safety" teams. The problem here is that AI training is probably not actually a copyright issue, some people just wish it was and are trying to manifest one by complaining. (Similar things exist in online artist communities with rules like "no reposting without crediting the artist" - they think if they just keep telling everyone this is the rule it'll become one.)
N0b8ez 2y ago
> silicon sentience Sentience is the ability to have experiences like pain and tasting sweetness. People usually use the word "sapient" or terms like AGI to describe an AI with advanced reasoning ability. This might sound like nitpicking since I was still able to understand your meaning, but the distinction between sentience/sapience is a useful one to keep around and preserve in common usage, at least for the sake of ethics (e.g. a newborn baby isn't sapient, but is sentient).
felipeerias 2y ago
There is little doubt that AI will be a successful technology. Ultimately, it will be able to create computers that are a lot more aware of the context in which they are used and which are much better at processing large amounts of unstructured human-made information. Personally, I find that the talk about AGI mainly serves as a marketing pitch, as well as providing an excuse for trampling over existing regulations. The second point is much more worrying. In the near future, the amount of plausibly-human content generated by machines will dwarf the actually-human content created by real people. Right now, people reading this comment will assume by default that it has been written by a real person. It might not be long until that default assumption changes.
whimsicalism 2y ago
I find it entertaining to see ML skeptics retreat in their rhetorical stances in real time. I remember what people were saying in 2022. Now in 2024, it has become the leftist anti-corporate position to favor strict enforcement of copyright law.
- vundercind 2y ago
  I’m also cool with none. But the worst case is “strict application of copyright for the little people… mass-scale ripping-off of the little people by rich people”.
__loam 2y ago
Thank you for considering the over exploitation of the commons.

ronsor 2y ago

I agree with this. On (2), I would like to see companies have no rights over models trained on public data. It's very arguable they should be required to release model weights.

vundercind 2y ago
Yeah IMO a good outcome would be that training on data you don’t own or license should require release of the model. Is allowed, doesn’t get you something you exclusively own. Bonus points if existing rights assignments aren’t enough to count as a grant of permission for AI training.
- astrange 2y ago
  I don't think there's any precedent for requiring someone to distribute a work they created. That sounds expensive and could easily be a contract violation for other data they did license.
  - astrange 2y ago
    Hmm, it is sort of how patents work. Copyright registration involves giving some sort of the work to the Library of Congress, but I'm not familiar with that part of it.

hn_throwaway_99 2y ago

> then gate “their” creations behind a paywall (oh, but that they expect us to respect, because that makes sense), and these AI safety folks don’t seem to give a shit about that. I downvoted your comment for this statement, given that's a specific worry discussed at length in this article.

vundercind 2y ago
Ah, I figured since they’d been on the board of an org that did exactly that bad thing, on a grand scale, they couldn’t possibly seriously care about it. Maybe they were in very early and so briefly that they weren’t party to that. IDK, article’s paywalled.

jamiek88 2y ago

> and these AI safety folks don’t seem to give a shit about that Because that’s literally not their job or role. Why would enforcing copyright be in any way shape or form their responsibility?

astrange 2y ago
Legal safety is (part of) the reason companies have "trust and safety" teams. The problem here is that AI training is probably not actually a copyright issue, some people just wish it was and are trying to manifest one by complaining. (Similar things exist in online artist communities with rules like "no reposting without crediting the artist" - they think if they just keep telling everyone this is the rule it'll become one.)

N0b8ez 2y ago

> silicon sentience Sentience is the ability to have experiences like pain and tasting sweetness. People usually use the word "sapient" or terms like AGI to describe an AI with advanced reasoning ability. This might sound like nitpicking since I was still able to understand your meaning, but the distinction between sentience/sapience is a useful one to keep around and preserve in common usage, at least for the sake of ethics (e.g. a newborn baby isn't sapient, but is sentient).

felipeerias 2y ago

There is little doubt that AI will be a successful technology. Ultimately, it will be able to create computers that are a lot more aware of the context in which they are used and which are much better at processing large amounts of unstructured human-made information. Personally, I find that the talk about AGI mainly serves as a marketing pitch, as well as providing an excuse for trampling over existing regulations. The second point is much more worrying. In the near future, the amount of plausibly-human content generated by machines will dwarf the actually-human content created by real people. Right now, people reading this comment will assume by default that it has been written by a real person. It might not be long until that default assumption changes.

whimsicalism 2y ago

I find it entertaining to see ML skeptics retreat in their rhetorical stances in real time. I remember what people were saying in 2022. Now in 2024, it has become the leftist anti-corporate position to favor strict enforcement of copyright law.

vundercind 2y ago
I’m also cool with none. But the worst case is “strict application of copyright for the little people… mass-scale ripping-off of the little people by rich people”.

__loam 2y ago

Thank you for considering the over exploitation of the commons.