Sandhini Agarwal: We now have numerous subsequent steps. I undoubtedly suppose how viral ChatGPT has gotten has made numerous points that we knew existed actually bubble up and grow to be crucial—issues we wish to remedy as quickly as potential. Like, we all know the mannequin remains to be very biased. And sure, ChatGPT is excellent at refusing unhealthy requests, however it’s additionally fairly simple to write down prompts that make it not refuse what we wished it to refuse.
Liam Fedus: It’s been thrilling to look at the varied and inventive functions from customers, however we’re all the time targeted on areas to enhance upon. We expect that via an iterative course of the place we deploy, get suggestions, and refine, we will produce essentially the most aligned and succesful expertise. As our expertise evolves, new points inevitably emerge.
Sandhini Agarwal: Within the weeks after launch, we checked out a number of the most horrible examples that folks had discovered, the worst issues individuals have been seeing within the wild. We form of assessed every of them and talked about how we must always repair it.
Jan Leike: Typically it’s one thing that’s gone viral on Twitter, however we now have some individuals who really attain out quietly.
Sandhini Agarwal: Quite a lot of issues that we discovered have been jailbreaks, which is certainly an issue we have to repair. However as a result of customers must strive these convoluted strategies to get the mannequin to say one thing unhealthy, it isn’t like this was one thing that we fully missed, or one thing that was very shocking for us. Nonetheless, that’s one thing we’re actively engaged on proper now. Once we discover jailbreaks, we add them to our coaching and testing knowledge. The entire knowledge that we’re seeing feeds right into a future mannequin.
Jan Leike: Each time we now have a greater mannequin, we wish to put it out and check it. We’re very optimistic that some focused adversarial coaching can enhance the state of affairs with jailbreaking rather a lot. It’s not clear whether or not these issues will go away completely, however we expect we will make numerous the jailbreaking much more tough. Once more, it’s not like we didn’t know that jailbreaking was potential earlier than the discharge. I feel it’s very tough to essentially anticipate what the true security issues are going to be with these techniques when you’ve deployed them. So we’re placing numerous emphasis on monitoring what persons are utilizing the system for, seeing what occurs, after which reacting to that. This isn’t to say that we shouldn’t proactively mitigate security issues after we do anticipate them. However yeah, it is rather onerous to foresee the whole lot that may really occur when a system hits the true world.
In January, Microsoft revealed Bing Chat, a search chatbot that many assume to be a model of OpenAI’s formally unannounced GPT-4. (OpenAI says: “Bing is powered by one among our next-generation fashions that Microsoft personalized particularly for search. It incorporates developments from ChatGPT and GPT-3.5.”) The usage of chatbots by tech giants with multibillion-dollar reputations to guard creates new challenges for these tasked with constructing the underlying fashions.
