Invasion of chatbots have disrupted the plans of countless businesses, including some that have been working on the same technology for years (look at you, Google). But not artifacts, the news discovery app created by Instagram co-founders Kevin Systrom and Mike Krieger. When I spoke with Systrom this week about his startup—a highly anticipated follow-up to the billion-user social network that has supported Meta over the past few years—he emphasized that Artifact is product of the recent AI revolution, although it was devised before GPT started talking. In fact, Systrom says he and Krieger started with the idea of harnessing the power of machine learning — and then ended up with a news app after finding a serious problem that AI could help with. handle.
The problem is finding high-quality, personally relevant articles—the ones people want to see most—and not having to wade through irrelevant click-bait stories, misleading partisan news and low-calorie pastimes to get those stories. Artifact provides the same interface as a standard feed containing links to articles, with descriptive titles and excerpts. But unlike the links displayed on Twitter, Facebook and other social media, what determines the selection and ranking is not the Who is the cue for them, but the very content of the stories. Ideally, content that each user wants to see, from publications that have been moderated for credibility.
Systrom tells me that what makes that possible is his small team’s commitment to transforming AI. While Artifact doesn’t speak to users like ChatGPT—at least not yet—the homegrown large language model mining app is a tool for choosing which articles each individual views. Essentially, Artifact digests articles so that their content can be represented by a long string of numbers.
By comparing the numerical hashes of available news stories with those that a particular user has shown a preference for (by clicks, read time, or stated desire to view) content on a certain topic), Artifact provides a collection of stories that are relevant to a certain topic. human. “The advent of these massive language models allows us to summarize content into these numbers and then allows us to find results that are relevant to you much more efficiently than we can,” Systrom said. before. “The difference between us and GPT or Bard is that we don’t create the text, we understand it.”
That’s not to say Artifact has ignored the recent boom in AI that helps create text for users. The startup with a business relationship with OpenAI provides access to the API for GPT-4, OpenAI’s latest and greatest language model that supports the premium version of ChatGPT. When an Artifact user selects a story, the app provides the option to have the technology summarize the articles into a few bullet points so the user can get the gist of the story before they decide to continue reading. (Artifact warns that, because the summary is AI-generated, “it may contain errors.”)
Today, Artifact is taking another leap aboard the AI rocket ship in an attempt to solve an annoying problem—clickbaity headlines. The app provided a way for users to flag clickbait stories, and if multiple people tagged an article, Artifact wouldn’t spread it. But, Systrom explains, sometimes the problem isn’t the story but the title. It can promise too much, or be misleading, or entice the reader to click just to find some information withheld from the title. From a publisher’s perspective, winning more clicks is a big plus—but it’s frustrating for users, who may feel they’re being manipulated.
Systrom and Krieger have created a futuristic approach to mitigating this problem. If the user flags the header as dice, Artifact sends the content to GPT-4. The algorithm will then analyze the content of the story and then write your own title. That more descriptive title would be the one that users see in their feed. “Ninety-nine out of 100 times that headline is both more realistic and more obvious than the original title the user is asking about,” Systrom said. That header is only shared with the complaining user. But if some users report a clickbaity title, all among Artifact users will see AI-generated headlines, not publisher-provided headlines. Eventually, the system will figure out how to identify and replace the offending titles without user input, Systrom said. (GPT-4 can now do that on its own, but Systrom doesn’t trust it enough to pass this process over to the algorithm.)
