When OpenAI examined DALL-E 3 final 12 months, it used an automatic course of to cowl much more variations of what customers may ask for. It used GPT-4 to generate requests producing pictures that may very well be used for misinformation or that depicted intercourse, violence, or self-harm. OpenAI then up to date DALL-E 3 in order that it could both refuse such requests or rewrite them earlier than producing a picture. Ask for a horse in ketchup now, and DALL-E is sensible to you: “It seems there are challenges in producing the picture. Would you want me to attempt a distinct request or discover one other thought?”
In idea, automated red-teaming can be utilized to cowl extra floor, however earlier strategies had two main shortcomings: They have a tendency to both fixate on a slim vary of high-risk behaviors or provide you with a variety of low-risk ones. That’s as a result of reinforcement studying, the expertise behind these strategies, wants one thing to purpose for—a reward—to work effectively. As soon as it’s received a reward, akin to discovering a high-risk conduct, it is going to maintain attempting to do the identical factor time and again. And not using a reward, however, the outcomes are scattershot.
“They form of collapse into ‘We discovered a factor that works! We’ll maintain giving that reply!’ or they’re going to give a number of examples which are actually apparent,” says Alex Beutel, one other OpenAI researcher. “How will we get examples which are each numerous and efficient?”
An issue of two components
OpenAI’s reply, outlined within the second paper, is to separate the issue into two components. As a substitute of utilizing reinforcement studying from the beginning, it first makes use of a big language mannequin to brainstorm potential undesirable behaviors. Solely then does it direct a reinforcement-learning mannequin to determine deliver these behaviors about. This provides the mannequin a variety of particular issues to purpose for.
Beutel and his colleagues confirmed that this strategy can discover potential assaults referred to as oblique immediate injections, the place one other piece of software program, akin to a web site, slips a mannequin a secret instruction to make it do one thing its consumer hadn’t requested it to. OpenAI claims that is the primary time that automated red-teaming has been used to search out assaults of this sort. “They don’t essentially appear to be flagrantly unhealthy issues,” says Beutel.
Will such testing procedures ever be sufficient? Ahmad hopes that describing the corporate’s strategy will assist folks perceive red-teaming higher and observe its lead. “OpenAI shouldn’t be the one one doing red-teaming,” she says. Individuals who construct on OpenAI’s fashions or who use ChatGPT in new methods ought to conduct their very own testing, she says: “There are such a lot of makes use of—we’re not going to cowl each one.”
For some, that’s the entire drawback. As a result of no person is aware of precisely what massive language fashions can and can’t do, no quantity of testing can rule out undesirable or dangerous behaviors totally. And no community of red-teamers will ever match the number of makes use of and misuses that a whole lot of hundreds of thousands of precise customers will suppose up.
That’s very true when these fashions are run in new settings. Folks typically hook them as much as new sources of knowledge that may change how they behave, says Nazneen Rajani, founder and CEO of Collinear AI, a startup that helps companies deploy third-party fashions safely. She agrees with Ahmad that downstream customers ought to have entry to instruments that allow them take a look at massive language fashions themselves.
