In subjects like engineering, chemistry and statistics, which drive a lot of traffic to Chegg but often involve diagrams, it was felt that relying too much on AI to parse visual information was unreasonable, the former employees say. So the ethic of unleashing an imperfect product made Chegg pause. “We knew that generative was lurking,” says a former director. “Text analytics was easy to embrace in the short term.”
In 2020, OpenAI’s GPT-3 model was released, making text generation much better. Some machine learning leaders at Chegg wanted to get their hands on it, but one source says executives weren’t aggressive about securing access to the technology, which OpenAI didn’t make available as open source. Early this year, GPT-3’s successor was added to ChatGPT, and the central role of generative AI in Chegg’s future became indisputable, as it was carved into the company’s dented user growth.
fight back
Chegg is now focused on proving with its in-house bot CheggMate that it’s possible to outdo ChatGPT when it storms your territory. “We happened to be one of the industries that was first involved with it, and that gives us a great opportunity to understand it deeper and faster and get to the other side of it with unique and value-creating products for our consumers,” says Schultz, the COO.
The company has put all extra hands into CheggMate and AI development, including reallocating teams that worked to collect more data from users to personalize services in more traditional ways. Brown, the CFO, told investors last month that the company’s summer interns will focus entirely on CheggMate. But Chegg doesn’t have the best track record of developing products from scratch and has previously relied on acquisitions, leading some former executives to follow CheggMate closely, unsure of the outlook.
The new service doesn’t exactly lighten the ethical implications, either. Chegg has long faced accusations from colleges and universities that it facilitates cheating while students secretly use their tools to complete homework and exams. Officially, Chegg prohibits unfair use and conducts and supports integrity investigations, said Nina Huntemann, the company’s chief academic officer. But former Chegg data scientist Eric Wang worries that CheggMate and similar applications could spread the habit of cheating. Students feel overwhelmed and short of time, and feel like they’re competing for scarce opportunities, he says. “All of these forces push students who know better to make decisions that aren’t great in hindsight,” says Wang, suggesting there could be better ways to support students and teachers.
Select users, along with Chegg’s subject matter experts and academic advisors, have begun testing CheggMate in recent weeks, but it’s not expected to be publicly launched until next year. That means it won’t be ready before the fall semester in the US, when Chegg typically generates the biggest sales. Schultz says he is proud of the company’s response to ChatGPT’s arrival. “We didn’t plan to react overnight and just throw something on the site,” he says. “We have a responsibility to be considerate.”
When a user types a question to CheggMate, it first tries to categorize whether it’s for help understanding a concept, solving a particular problem, or topic, says Schultz. The system then tries to route the question to the best source, with options such as asking GPT-4, an answer from a human expert, or rebroadcasting an old answer from Chegg’s database. CheggMate is designed to keep users engaged through positive reinforcement and pushing related content. “We could say, ‘Why don’t you try this similar problem? Why don’t you guess a step?'” says Huntemann, the chief academic officer. “Conversation allows us to expand the experience.”
Chegg executives hope that tailoring their chatbot to education in this way will make ChatGPT look less attractive as a homework helper. The price for CheggMate has not been set; using generative models is expensive, and those costs increase with use. But two former employees say it costs about $2 to have a human expert answer a question. Generating a similar response via GPT-4 might cost half a dollar, and with expert editing it could cost a total of $1, they say, suggesting the economy could work for Chegg.
At the same time, competition is likely to increase from ChatGPT itself, Microsoft and Google’s generative AI-powered search, or rivals developing their own AI teachers using OpenAI technology like Quizlet, Brainly, and Khan Academy. That could force Chegg to spend more on marketing to stay relevant. Silber, the equity analyst, expects Chegg’s operating profit margins to suffer for some time.
The recent drive has made Rosensweig, who has led Chegg since 2010, and his friend of 20 years, OpenAI’s Altman, competitors and perhaps enemies. They both now have a hand in shaping the next chapter of education.
All the people WIRED spoke to described Rosensweig, whose mother was a public school teacher, as someone who wants people to have the opportunity to lift themselves up through educational opportunities. He hosts an online show called Go broke, in which he and a financial strategist help people with major money problems get back on their feet. If Rosensweig were to rebuild Chegg for the generative AI era, ChatGPT suggests show could be called Chegg Resurgence: triumph over turbulence.
