The inside story of how ChatGPT was built from the people who made it
Sandhini Agarwal: We have a lot of next steps. I definitely think how viral ChatGPT has gotten has made a lot of issues that we knew existed really bubble up and become critical—things we want to solve as soon as possible. Like, we know the model is still very biased. And yes, ChatGPT is very good at refusing bad requests, but it’s also quite easy to write prompts that make it not refuse what we wanted it to refuse.
Liam Fedus: It’s been thrilling to watch the diverse and creative applications from users, but we’re always focused on areas to improve upon. We think that through an iterative process where we deploy, get feedback, and refine, we can produce the most aligned and capable technology. As our technology evolves, new issues inevitably emerge.
Sandhini Agarwal: In the weeks after launch, we looked at some of the most terrible examples that people had found, the worst things people were seeing in the wild. We kind of assessed each of them and talked about how we should fix it.
Jan Leike: Sometimes it’s something that’s gone viral on Twitter, but we have some people who actually reach out quietly.
Sandhini Agarwal: A lot of things that we found were jailbreaks, which is definitely a problem we need to fix. But because users have to try these convoluted methods to get the model to say something bad, it isn’t like this was something that we completely missed, or something that was very surprising for us. Still, that’s something we’re actively working on right now. When we find jailbreaks, we add them to our training and testing data. All of the data that we’re seeing feeds into a future model.
Jan Leike: Every time we have a better model, we want to put it out and test it. We’re very optimistic that some targeted adversarial training can improve the situation with jailbreaking a lot. It’s not clear whether these problems will go away entirely, but we think we can make a lot of the jailbreaking a lot more difficult. Again, it’s not like we didn’t know that jailbreaking was possible before the release. I think it’s very difficult to really anticipate what the real safety problems are going to be with these systems once you’ve deployed them. So we are putting a lot of emphasis on monitoring what people are using the system for, seeing what happens, and then reacting to that. This is not to say that we shouldn’t proactively mitigate safety problems when we do anticipate them. But yeah, it is very hard to foresee everything that will actually happen when a system hits the real world.
In January, Microsoft revealed Bing Chat, a search chatbot that many assume to be a version of OpenAI’s officially 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 protect creates new challenges for those tasked with building the underlying models.
The Download: AI films, and the threat of microplastics
The Frost nails its uncanny, disconcerting vibe in its first few shots. Vast icy mountains, a makeshift camp of military-style tents, a group of people huddled around a fire, barking dogs. It’s familiar stuff, yet weird enough to plant a growing seed of dread. There’s something wrong here.
Welcome to the unsettling world of AI moviemaking. The Frost is a 12-minute movie from Detroit-based video creation company Waymark in which every shot is generated by an image-making AI. It’s one of the most impressive—and bizarre—examples yet of this strange new genre. Read the full story, and take an exclusive look at the movie.
—Will Douglas Heaven
Microplastics are everywhere. What does that mean for our immune systems?
Microplastics are pretty much everywhere you look. These tiny pieces of plastic pollution, less than five millimeters across, have been found in human blood, breast milk, and placentas. They’re even in our drinking water and the air we breathe.
Given their ubiquity, it’s worth considering what we know about microplastics. What are they doing to us?
The short answer is: we don’t really know. But scientists have begun to build a picture of their potential effects from early studies in animals and clumps of cells, and new research suggests that they could affect not only the health of our body tissues, but our immune systems more generally. Read the full story.
Microplastics are everywhere. What does that mean for our immune systems?
Here, bits of plastic can end up collecting various types of bacteria, which cling to their surfaces. Seabirds that ingest them not only end up with a stomach full of plastic—which can end up starving them—but also get introduced to types of bacteria that they wouldn’t encounter otherwise. It seems to disturb their gut microbiomes.
There are similar concerns for humans. These tiny bits of plastic, floating and flying all over the world, could act as a “Trojan horse,” introducing harmful drug-resistant bacteria and their genes, as some researchers put it.
It’s a deeply unsettling thought. As research plows on, hopefully we’ll learn not only what microplastics are doing to us, but how we might tackle the problem.
Read more from Tech Review’s archive
It is too simplistic to say we should ban all plastic. But we could do with revolutionizing the way we recycle it, as my colleague Casey Crownhart pointed out in an article published last year.
We can use sewage to track the rise of antimicrobial-resistant bacteria, as I wrote in a previous edition of the Checkup. At this point, we need all the help we can get …
… which is partly why scientists are also exploring the possibility of using tiny viruses to treat drug-resistant bacterial infections. Phages were discovered around 100 years ago and are due a comeback!
Our immune systems are incredibly complicated. And sex matters: there are important differences between the immune systems of men and women, as Sandeep Ravindran wrote in this feature, which ran in our magazine issue on gender.
Welcome to the new surreal. How AI-generated video is changing film.
Fast and cheap
Artists are often the first to experiment with new technology. But the immediate future of generative video is being shaped by the advertising industry. Waymark made The Frost to explore how generative AI could be built into its products. The company makes video creation tools for businesses looking for a fast and cheap way to make commercials. Waymark is one of several startups, alongside firms such as Softcube and Vedia AI, that offer bespoke video ads for clients with just a few clicks.
Waymark’s current tech, launched at the start of the year, pulls together several different AI techniques, including large language models, image recognition, and speech synthesis, to generate a video ad on the fly. Waymark also drew on its large data set of non-AI-generated commercials created for previous customers. “We have hundreds of thousands of videos,” says CEO Alex Persky-Stern. “We’ve pulled the best of those and trained it on what a good video looks like.”
To use Waymark’s tool, which it offers as part of a tiered subscription service starting at $25 a month, users supply the web address or social media accounts for their business, and it goes off and gathers all the text and images it can find. It then uses that data to generate a commercial, using OpenAI’s GPT-3 to write a script that is read aloud by a synthesized voice over selected images that highlight the business. A slick minute-long commercial can be generated in seconds. Users can edit the result if they wish, tweaking the script, editing images, choosing a different voice, and so on. Waymark says that more than 100,000 people have used its tool so far.
The trouble is that not every business has a website or images to draw from, says Parker. “An accountant or a therapist might have no assets at all,” he says.
Waymark’s next idea is to use generative AI to create images and video for businesses that don’t yet have any—or don’t want to use the ones they have. “That’s the thrust behind making The Frost,” says Parker. “Create a world, a vibe.”
The Frost has a vibe, for sure. But it is also janky. “It’s not a perfect medium yet by any means,” says Rubin. “It was a bit of a struggle to get certain things from DALL-E, like emotional responses in faces. But at other times, it delighted us. We’d be like, ‘Oh my God, this is magic happening before our eyes.’”
This hit-and-miss process will improve as the technology gets better. DALL-E 2, which Waymark used to make The Frost, was released just a year ago. Video generation tools that generate short clips have only been around for a few months.
The most revolutionary aspect of the technology is being able to generate new shots whenever you want them, says Rubin: “With 15 minutes of trial and error, you get that shot you wanted that fits perfectly into a sequence.” He remembers cutting the film together and needing particular shots, like a close-up of a boot on a mountainside. With DALL-E, he could just call it up. “It’s mind-blowing,” he says. “That’s when it started to be a real eye-opening experience as a filmmaker.”