3 ms·

The Register is so irritating to read. I get that the tone is a shtick, but really? > But AI companies have insisted they’re working to suppress harmful conten

The Register is so irritating to read. I get that the tone is a shtick, but really? > But AI companies have insisted they’re working to suppress harmful content buried within AI training data so things like recipes for explosives don’t appear. Why the implication that this is not true? I know it’s house style to imply wrongdoing whenever possible, but damn it gets old to read. And then! To not even cover the difference between models and prompts. Does this jailbreak only work on models where the safety instructions are purely at the prompt level? Is that why GPT-4 resisted, because it has actual trained weights for safety? Does it even have those? Who knows? Certainly not the article’s author, who has likely focused on searching and replacing “says” with “claims” or “admits” rather than anything relevant to the subject. /rant

bitmasher9 2y ago
* This is the register, I saw a deeper technical explanation than I was expecting, even if it is too shallow for those an intermediate understanding of these models. * I think releasing these models with this weakness is a level of wrongdoing, and while they are working on a fix someone could be learning how to harm other people. I’m skeptical all knowledge leaks will be patched in a decade.
- brookst 2y ago
  But “releasing these models” is the whole problem. Models and prompts are different things. Prompts evolve daily. And any safety at the prompt level is literally non-existent when developers access the model itself without the prompt used in a commercial product. OpenAI playground versus ChetGPT, for example. The article he’s negative informational value. I wouldn’t mind all of the energy Register puts into snark if there was a real effort to be informative, but it’s irritating to read something where far more energy went into tone than content.

bitmasher9 2y ago

* This is the register, I saw a deeper technical explanation than I was expecting, even if it is too shallow for those an intermediate understanding of these models. * I think releasing these models with this weakness is a level of wrongdoing, and while they are working on a fix someone could be learning how to harm other people. I’m skeptical all knowledge leaks will be patched in a decade.

brookst 2y ago
But “releasing these models” is the whole problem. Models and prompts are different things. Prompts evolve daily. And any safety at the prompt level is literally non-existent when developers access the model itself without the prompt used in a commercial product. OpenAI playground versus ChetGPT, for example. The article he’s negative informational value. I wouldn’t mind all of the energy Register puts into snark if there was a real effort to be informative, but it’s irritating to read something where far more energy went into tone than content.