5 ms·

Can we actually expect OpenAI to remain "open" with investments like this getting dumped into the project? I'm still waiting for the 1.5G GPT-2 set to get rele

by RIMR 7y ago

Can we actually expect OpenAI to remain "open" with investments like this getting dumped into the project? I'm still waiting for the 1.5G GPT-2 set to get released, but they're still going with that "too dangerous for society" BS that they're using to get journalists' attention...

anchpop 7y ago
They need to prove it's safe for society before releasing it. If they don't believe it is then not releasing it is of course the smart thing to do. Furthermore, anyone else in the future who creates something they think might be dangerous now has a better argument, because they can point to OpenAI playing it safe and say "I'm just doing the same thing"
- RIMR 7y ago
  I think you're making a silly mistake actually giving this "too dangerous for society" line any validity. I understand that this tech could be used for nefarious purposes, but this isn't world-ending tech. This is just hard to differentiate from human writing... The choice to keep their "too powerful model" unreleased is more an attempt to stoke sensationalism out of journalists eager to report on "The AI too dangerous to release" than it is actually an earnest attempt at protecting society. The dangerous rogue-AI is a Hollywood trope. We don't live in the Ghost In the Shell universe, we live in reality, and a text-generating algorithm isn't particularly dangerous when you think about it.
  - anchpop 7y ago
    > I understand that this tech could be used for nefarious purposes, but this isn't world-ending tech. This is just hard to differentiate from human writing... When, in any OpenAI communication, have they ever implied GPT2 is world ending tech? > We don't live in the Ghost In the Shell universe, we live in reality, and a text-generating algorithm isn't particularly dangerous when you think about it. Are you sure it won't be used to automatically post fake news or create artificial group cohesion by bad state actors? We can already do stuff like that, but this allows you to do it faster and cheaper. Don't you think that's at least a little scary?
    - RIMR 7y ago
      Again, I already made it clear that this has nefarious purposes. But there is zero utility in keeping the larger dataset secret. Society is not safer as a result. The only reason they ever framed it the way they did was to get media attention. "This AI is so dangerous, the creators are refusing to release it to the public!" is going to get way more clicks than "This AI writes English sentences that almost appear coherent.". If you want more funding and attention towards your project, you're going to say stuff that gets journalists' attention.

by RIMR 7y ago

anchpop 7y ago

They need to prove it's safe for society before releasing it. If they don't believe it is then not releasing it is of course the smart thing to do. Furthermore, anyone else in the future who creates something they think might be dangerous now has a better argument, because they can point to OpenAI playing it safe and say "I'm just doing the same thing"

RIMR 7y ago
I think you're making a silly mistake actually giving this "too dangerous for society" line any validity. I understand that this tech could be used for nefarious purposes, but this isn't world-ending tech. This is just hard to differentiate from human writing... The choice to keep their "too powerful model" unreleased is more an attempt to stoke sensationalism out of journalists eager to report on "The AI too dangerous to release" than it is actually an earnest attempt at protecting society. The dangerous rogue-AI is a Hollywood trope. We don't live in the Ghost In the Shell universe, we live in reality, and a text-generating algorithm isn't particularly dangerous when you think about it.
- anchpop 7y ago
  > I understand that this tech could be used for nefarious purposes, but this isn't world-ending tech. This is just hard to differentiate from human writing... When, in any OpenAI communication, have they ever implied GPT2 is world ending tech? > We don't live in the Ghost In the Shell universe, we live in reality, and a text-generating algorithm isn't particularly dangerous when you think about it. Are you sure it won't be used to automatically post fake news or create artificial group cohesion by bad state actors? We can already do stuff like that, but this allows you to do it faster and cheaper. Don't you think that's at least a little scary?
  - RIMR 7y ago
    Again, I already made it clear that this has nefarious purposes. But there is zero utility in keeping the larger dataset secret. Society is not safer as a result. The only reason they ever framed it the way they did was to get media attention. "This AI is so dangerous, the creators are refusing to release it to the public!" is going to get way more clicks than "This AI writes English sentences that almost appear coherent.". If you want more funding and attention towards your project, you're going to say stuff that gets journalists' attention.