Why does it feel like AI suddenly got good at things?
Help us reach our goal Weâre aiming to add 3,000 financial contributions from readers by December 31 to help keep Vox free for all. Recurring monthly or annual gifts, in particular, help us plan and weather a notoriously unpredictable industry. Will you help us reach our goal by making a contribution to Vox today? [Yes, I'll give]( AI is finally good at stuff, and that's a problem A few weeks ago, Wharton professor Ethan Mollick told his MBA students to play around with [GPT](, an artificial intelligence model, and see if the technology could write an essay based on one of the topics discussed in his course. The assignment was, admittedly, mostly a gimmick meant to illustrate the power of the technology. Still, the algorithmically generated essays â although not perfect and a tad over-reliant on the passive voice â were at least reasonable, Mollick recalled. They also passed another critical test: a screening by Turnitin, a popular anti-plagiarism software. AI, it seems, had suddenly gotten pretty good. It certainly feels that way right now. Over the past week or so, screenshots of conversations with ChatGPT, the newest iteration of the AI model [developed]( by the research firm OpenAI, have gone viral on social media. People have directed the tool, which is freely available online, to make jokes, write TV episodes, compose music, and even debug computer code â all things I got the AI to do, too. [More than a million people]( have now played around with the AI, and even though it doesnât always tell the truth or make sense, itâs still a pretty good writer and an even more confident bullshitter. Along with the recent updates to [DALL-E](, OpenAIâs art-generation software, and Lensa AI, a [controversial]( [platform]( that can produce digital portraits with the help of machine learning, GPT is a stark wakeup call that artificial intelligence is starting to rival human ability, at least for some things. âI think that things have changed very dramatically,â Mollick told Recode. âAnd I think it's just a matter of time for people to notice.â If youâre not convinced, you can [try it yourself here](. The system works like any online chatbot, and you can simply type out and submit any question or prompt you want the AI to address. How does GPT even work? At its core, the technology is based on a type of artificial intelligence called a language model, a prediction system that essentially guesses what it should write, based on previous texts it has processed. GPT was built by training its AI with an extraordinarily large amount of data, much of which comes from the vast supply of data on the internet, along with billions of dollars, including [initial funding]( from several prominent tech billionaires, including Reid Hoffman and Peter Thiel. ChatGPT was also trained on examples of back-and-forth human conversation, which helps it make its dialogue sound a lot more human, as [a blog post]( published by OpenAI explains. OpenAI is trying to commercialize its technology, but this current release is supposed to allow the public to test it. The company made headlines two years ago when it released GPT-3, an iteration of the tech that could produce poems, role-play, and answer some questions. This newest version of the technology is GPT-3.5, and ChatGPT, its corresponding chatbot, is even better at text generation than its predecessor. Itâs also pretty good at following [instructions](, like, âWrite a Frog and Toad short story where Frog invests in mortgage-backed securities.â (The story ends with Toad following Frogâs advice and investing in mortgage-backed securities, concluding that âsometimes taking a little risk can pay off in the endâ). The technology certainly has its flaws. While the system is theoretically designed not to cross some [moral red lines]( â itâs adamant that Hitler [was bad]( â itâs not difficult to trick the AI into sharing advice on how to engage in all sorts of evil and nefarious activities, particularly if you tell the chatbot that itâs writing fiction. The system, like [other AI]( models, can also say [biased]( and offensive things. As my colleague Sigal Samuel has [explained](, an earlier version of GPT generated extremely Islamophobic content, and also produced some pretty concerning talking points about the treatment of Uyghur Muslims in China. Both GPTâs impressive capabilities and its limitations reflect the fact that the technology operates like a version of Googleâs smart compose writing suggestions, generating ideas based on what it has read and processed before. For this reason, the AI can sound extremely confident while not displaying a particularly deep understanding of the subject itâs writing about. This is also why itâs easier for GPT to write about commonly discussed topics, like a Shakespeare play or the importance of mitochondria. âIt wants to produce texts that it deemed to be likely, given everything that it has seen before,â explains Vincent Conitzer, a computer science professor at Carnegie Mellon. âMaybe it sounds a little bit generic at times, but it writes very clearly. It will probably rehash points that have often been made on that particular topic because it has, in effect, learned what kinds of things people say.â So for now, weâre not dealing with an all-knowing bot. Answers provided by the AI were recently banned from the coding feedback platform StackOverflow because they were very likely to be incorrect. The chatbot is also easily tripped up by riddles (though its attempts to answer [are extremely funny](). Overall, the system is perfectly comfortable making stuff up, which obviously makes no sense upon human scrutiny. These limitations might be comforting to people worried that the [AI could take their jobs](, or eventually pose [a safety threat to humans](. But AI is getting better and better, and even this current version of GPT can already do extremely well at certain tasks. Consider Mollickâs assignment. While the system certainly wasnât good enough to earn an A, it still did pretty well. One Twitter user said that, on a mock SAT exam, ChatGPT [scored]( around the 52 percentile of test takers. Kris Jordan, a computer science professor at UNC, told Recode that when he assigned GPT his final exam, the chatbot received a perfect grade, far better than the median score for the humans taking his course. And yes, even before ChatGPT went live, students were using all sorts of artificial intelligence, including earlier [versions]( of GPT, to complete their assignments. And theyâre probably not getting flagged for cheating. (Turnitin, the anti-plagiarism software maker, did not respond to multiple requests for comment). Right now, itâs not clear how many enterprising students might start using GPT, or if teachers and professors will figure out a way to catch them. Still, these forms of AI are already forcing us to wrestle with what kinds of things we want humans to continue to do, and what weâd prefer to have technology figure out instead. âMy eighth grade math teacher told me not to rely on a calculator since I wonât have one in my pocket all the time when I grow up,â Phillip Dawson, an expert who studies exam cheating at Deakin University, told Recode. âWe all know how that turned out. âRebecca Heilweil, reporter [The Twitter logo in a cloudy spotlight.]( Christina Animashaun/Vox [We donât need another Twitter]( [Twitter never fulfilled its promise. Donât expect its current crop of replacements to, either.]( [Elon Musk dressed in a gladiator-style costume, holding his arms up and out in a âlook at meâ gesture.]( Taylor Hill/Getty Images [The Twitter power users who canât just quit]( [Under Elon Muskâs leadership, journalists have even more of a love-hate relationship with the platform.]( [Attendees of Web Summit 2022 mingle as they arrive at Centre Stage at the Altice Arena in Lisbon, Portugal, on November 1. ]( Lukas Schulze/Sportsfile for Web Summit via Getty Images [Conferences want to cure the work-from-home blues]( [Sunshine and wine caves: How conferences are trying to win over a remote-work world.](
Â
[Learn more about RevenueStripe...]( [Photo of Reed Hastings, seated and with his arms crossed, looking to the side in a dark room.]( Wolter Peeters/Fairfax Media via Getty Images [Netflix CEO Reed Hastings likes Elon Musk, video games, and bingeing]( [âElon Musk is the bravest, most creative person on the planet.â]( [A photo of a man in a workerâs uniform wheeling office chairs, on top of which plants perch, through an empty-looking office setting.]( David Oliver/Getty Images [Hereâs what employers are cutting instead of your job]( [You may get to keep your job, but you might lose your Zoom and your desk.]( Help us reach our goal We aim to understand technology and the companies behind it shaping our world. Support our mission by making a gift during our end-of-year campaign. [Yes, I'll make a gift]( [Listen To This] [Listen to This]( [How Eventbrite survived the pandemic]( Selling tickets for live events can be⦠tricky, as Ticketmaster has proven yet again. But Eventbrite CEO Julia Hartz runs a much smaller ticketing business and says sheâs learned valuable lessons after the pandemic brought her revenue to zero. She talks to Recodeâs Peter Kafka about growing once again â and how sheâs preparing for a potential recession. [Listen on Apple Podcasts.]( [This is cool] [Stylists arrive in the metaverse](
Â
[Learn more about RevenueStripe...]( [Vox Logo]( [Facebook]( [Twitter]( [YouTube]( This email was sent to {EMAIL}. Manage yourâ¯[email preferences]( orâ¯[unsubscribe](param=recode). View our [Privacy Notice]( and our [Terms of Service](. Vox Media, 1201 Connecticut Ave. NW, Floor 12, Washington, DC 20036. Copyright © 2022. All rights reserved.