3 ms·

I’ve seen it do this even for things that are so niche that it couldn’t possibly have been fine-tuned manually (it was unrelated to anything political). It must

I’ve seen it do this even for things that are so niche that it couldn’t possibly have been fine-tuned manually (it was unrelated to anything political). It must develop some sort of sense of what’s a reasonable/wise perspective and then stick to that. Like what a human would do. But I think it’s a bias to safety and politeness. …which is pretty reasonable? “Hey, this is how to make a gun!” …is a fast track to getting sued.

dragonwriter 4y ago
> I’ve seen it do this even for things that are so niche that it couldn’t possibly have been fine-tuned manually (it was unrelated to anything political). It is worth noting that one of OpenAI’s public product is a moderation classifier running its own (continuously updated) model and providing scores in various “objectionable content” categories; it’s possible they are using something like a more advanced version of this to determine whether, and what kind, of “I won’t respond because…” answer to use, rather than something relying only on manual identification of particularly narrow content.
- Robotbeat 4y ago
  That sounds exactly like what is going on. Are there specific ways of prompting it that can let you manipulate that moderation classifier? (I almost think half of the point of opening up CharGPT like this is to flex the fact that they can prevent it from being Milkshake Ducked by the usual level of cherrypicking and malevolent use.)
itsyaboi 4y ago
Why would gun making instructions get you sued?
- Robotbeat 4y ago
  They want to deploy this tech to, like, high schools.
  - josephcsible 4y ago
    Does Google censor searches for gun making from high schools? If not, have they gotten sued?
    - Robotbeat 4y ago
      I can assure you that “SafeSearch” is enabled on school computers.
      - josephcsible 4y ago
        Does SafeSearch block results about making guns, though? I just did a quick test of it and it didn't seem to.

dragonwriter 4y ago

> I’ve seen it do this even for things that are so niche that it couldn’t possibly have been fine-tuned manually (it was unrelated to anything political). It is worth noting that one of OpenAI’s public product is a moderation classifier running its own (continuously updated) model and providing scores in various “objectionable content” categories; it’s possible they are using something like a more advanced version of this to determine whether, and what kind, of “I won’t respond because…” answer to use, rather than something relying only on manual identification of particularly narrow content.

Robotbeat 4y ago
That sounds exactly like what is going on. Are there specific ways of prompting it that can let you manipulate that moderation classifier? (I almost think half of the point of opening up CharGPT like this is to flex the fact that they can prevent it from being Milkshake Ducked by the usual level of cherrypicking and malevolent use.)

itsyaboi 4y ago

Why would gun making instructions get you sued?

Robotbeat 4y ago
They want to deploy this tech to, like, high schools.
- josephcsible 4y ago
  Does Google censor searches for gun making from high schools? If not, have they gotten sued?
  - Robotbeat 4y ago
    I can assure you that “SafeSearch” is enabled on school computers.
    - josephcsible 4y ago
      Does SafeSearch block results about making guns, though? I just did a quick test of it and it didn't seem to.