The Professor Who Required His Students Use ChatGPT In Class — With Ethan Mollick

Episode Summary

Executive Summary: Ethan Mollick argues AI is a general-purpose technology that is already reshaping education and work, even if consumer interest in ChatGPT appears to ebb. He says the biggest near-term shift is not chatbot novelty but how AI changes homework, tutoring, classroom design, task execution, and workplace incentives, forcing schools and companies to rethink cheating, expertise, and the meaning of work itself.

Main Topics: AI as a general-purpose technology (Priority: 5/5): Mollick frames AI as a once-in-a-generation general-purpose technology like the internet or steam power, with broad productivity effects beyond consumer app usage. Education and the 'homework apocalypse' (Priority: 5/5): He argues AI makes traditional homework, essays, and problem sets obsolete or easily gamed, requiring schools to shift toward active learning, flipped classrooms, and in-class practice. AI as tutor and teaching tool (Priority: 5/5): He describes AI as a universal tutor that can fill learning gaps, generate exercises, critique work, and help students do 'impossible' tasks, making one-on-one tutoring more scalable. Workplace adoption is bottom-up (Priority: 4/5): AI use in companies is often driven by individual employees experimenting privately rather than formal management rollout, because the personal productivity upside is immediate and easy to test. Future of jobs, tasks, and meaning (Priority: 4/5): Mollick distinguishes tasks from jobs and warns AI will reshape job contracts, reduce drudgery, and force organizations to rebuild systems that depend on human slowness and signaling. Model landscape, tools, and hallucinations (Priority: 3/5): He compares models like Claude, Pi, and GPT, praises code interpreter/tool use as a major capability boost, and says hallucinations are falling but still matter for expert review.

Key Arguments: Consumer download trends are less important than real-world adoption in work and education, because AI’s impact comes from task-level integration and API use. Cheating is real but is not the central story; the larger change is that homework and essays no longer measure learning well, so education must be redesigned. Traditional teaching methods overemphasize lectures and rote work; AI can enable more active learning, better feedback, and tutoring at scale. Students still need facts and deliberate practice because expertise cannot be shortcut; AI increases the value of deep subject knowledge rather than replacing it. Companies will adopt AI faster than schools because individual workers have strong incentives to use tools that save time and increase output, even if firms ban them. AI will automate drudgery first, but that creates a meaning problem: many workplace rituals exist as signals of effort and control, not because they produce value. The biggest long-term question is whether AI keeps improving rapidly enough to fundamentally restructure work, inequality, and the social contract.

Data Points: ChatGPT adoption trend: Interest appears to be dropping off - Discussed as possibly due to school being out, competition from Bing/Claude/Bard, or users bouncing off the product Workplace AI disclosure: 50% said no - Mollick referenced a poll where half of respondents said they do not tell anyone if they use AI at work AI productivity gains in tasks: 30% to 80% - He said task performance improvements from AI could be very large in some workflows Homework effectiveness study: 86% to 45% - One study of 11 years of college courses found homework improved test grades for 86% of students in 2008 but only 45% in 2017 Internet-enabled homework checking: Half of students - The drop in homework effectiveness was attributed to roughly half of students looking up homework answers online in 2017 Full-time essay writing market: 20,000 people - He cited a figure that, as of last year before ChatGPT, about 20,000 people in Kenya wrote essays full-time Personal research capital raised: About $2 billion - He said entrepreneurship students in his classes have raised about $2 billion over the last decade or so Claude context window: 100,000 tokens - He noted Claude can hold a short book in memory and is strong for document work Model hallucination rates: 44%, 22%, 2% - He cited a board-exam study in which Bard hallucinated 44% of answers, ChatGPT 3.5 hallucinated 22%, and GPT-4 only 2% Global availability: 169 countries - He said Bing/GPT-4 access is available in 169 countries Top-percentile threshold: Top 10% / 8% / 7% - He suggested that if AI keeps improving, only the top slice of workers may remain competitive in certain tasks

Pivotal Quotes: "We’re facing what I call the homework apocalypse." — Ethan Mollick: On how generative AI threatens traditional homework, essays, and problem sets "In some ways, the AI requires us to actually double down on the basic knowledge because we can only build the advanced knowledge from the basic knowledge." — Ethan Mollick: On why students still need facts and deep expertise even with AI tools "The companies that figure this out first are going to win." — Ethan Mollick: On organizational redesign as AI changes work and coordination

Implications: Schools must redesign assessment around AI’s presence, shifting from take-home rote work to tutoring, practice, and in-class application. Firms should encourage safe experimentation, reward productivity gains, and rethink job design before AI reshapes incentives for them.