Sandhini Agarwal: We have quite a lot of subsequent steps. I positively suppose how viral ChatGPT has gotten has made quite a lot of points that we knew existed actually bubble up and turn out to be essential—issues we wish to clear up as quickly as doable. Like, we all know the mannequin remains to be very biased. And sure, ChatGPT is superb at refusing dangerous requests, but it surely’s additionally fairly simple to write down prompts that make it not refuse what we needed it to refuse.
Liam Fedus: It’s been thrilling to observe the varied and inventive functions from customers, however we’re at all times centered on areas to enhance upon. We suppose that by an iterative course of the place we deploy, get suggestions, and refine, we will produce probably the most aligned and succesful expertise. As our expertise evolves, new points inevitably emerge.
Sandhini Agarwal: In the weeks after launch, we checked out a number of the most horrible examples that folks had discovered, the worst issues folks had been seeing within the wild. We type of assessed every of them and talked about how we must always repair it.
Jan Leike: Sometimes it’s one thing that’s gone viral on Twitter, however now we have some individuals who really attain out quietly.
Sandhini Agarwal: A number of issues that we discovered had been jailbreaks, which is unquestionably an issue we have to repair. But as a result of customers should strive these convoluted strategies to get the mannequin to say one thing dangerous, it isn’t like this was one thing that we fully missed, or one thing that was very shocking for us. Still, that’s one thing we’re actively engaged on proper now. When we discover jailbreaks, we add them to our coaching and testing knowledge. All of the info that we’re seeing feeds right into a future mannequin.
Jan Leike: Every time now we have a greater mannequin, we wish to put it out and take a look at it. We’re very optimistic that some focused adversarial coaching can enhance the scenario with jailbreaking quite a bit. It’s not clear whether or not these issues will go away totally, however we predict we will make quite a lot of the jailbreaking much more tough. Again, it’s not like we didn’t know that jailbreaking was doable 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 quite a lot of emphasis on monitoring what individuals are utilizing the system for, seeing what occurs, after which reacting to that. This is to not say that we shouldn’t proactively mitigate security issues after we do anticipate them. But yeah, it is rather exhausting to foresee all the things that can 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 of our next-generation models that Microsoft customized specifically for search. It incorporates advancements from ChatGPT and GPT-3.5.”) The use of chatbots by tech giants with multibillion-dollar reputations to guard creates new challenges for these tasked with constructing the underlying fashions.