As AI becomes ever more ubiquitous, John DeNero Ph.D. ‘10 addressed students, parents and alums at UC Berkeley Homecoming, sharing his experiences in designing curriculum and educational support tools for computer and data science students.
DeNero is teaching professor in the Department of Electrical Engineering and Computer Sciences and faculty director of pedagogy for the Data Science Undergraduate Studies program.
He received the Fiat Lux Faculty Award earlier this year in recognition of his contributions to developing world-class educational programs – including co-founding the top-ranked data science undergraduate major, currently the largest major at Berkeley in terms of student enrollment.
Building human competence in computing
“Things are going to start out a little bit gloomy because the reality of how AI is being used in education is not all positive at the moment,” DeNero said during the Oct. 9 event. “Then I’ll walk you through how I think Berkeley is leading the way in making it a lot better.”
DeNero mentioned recent news that OpenAI claimed to have solved hundreds of math problems, leading mathematicians and computer scientists to consider the role of humans in their fields. “The rate of progress and the problems that people are working on is just changing really fast.”
“So can’t AI just do this stuff for us now?” he asked. “The answer right now is definitely not. There’s a ton of value in building up human competence and human understanding for computing.”
DeNero cited a 2026 Anthropic study finding that success rates of Claude Code sessions varied from 39% for novice users to 65% for expert users. “Human competence is still a huge advantage, and we’ve also observed that students use AI differently depending on how much they know.”
“Students that have built software themselves tend to focus on scrutinizing the AI's plan, reading the code that comes out of the AI, and making highly specific requests about how it should change – that’s what leads to the higher success rate,” he said.
Human evaluation of AI is still necessary, DeNero said. Every company building AI “is trying to understand whether some AI tool is actually solving the problem it's supposed to solve, and that’s very much a human activity. There are lots of attempts to have AI evaluate itself, and they don’t work yet.”
Learning computer science takes practice
Transitioning from why humans should be trained in computing and AI, DeNero next considered how students should learn computer science.
“They should do their homework because that’s how you get to practice actually using what you learn from reading the textbook or going to lecture, and it's really important,” he said. “You have to practice the piano in order to learn piano. You can’t just watch videos of people playing piano. Computer science is like that. You need to do the work in order to get good at it.”
Homework serves several purposes, according to DeNero, including practice in applying techniques and concepts, discovering new ways that techniques and concepts can apply, and helping students assess their capabilities.
“When students have AI explain to them how to do their homework – and that usually includes the AI telling them the answers to their homework – they may feel like they're learning because now they know the answers, and they have an explanation,” said DeNero. “It feels very productive, but I don't think it works. I don't think it actually gives them practice. I don't think it helps them discover how things apply, and it certainly doesn't help them assess their capabilities.”
Developing better AI support tools
In considering how AI tools can support learning, DeNero mentioned a few early examples of AI-enabled course tutoring tools developed in 2023 separately by UC Berkeley, Harvard University, and Khan Academy.
“The idea was to build an AI that wouldn’t give the answers away, but would guide students to their mistakes,” he said. “You still need humans to write the problems, but AI can provide individualized feedback to the students when they’re doing the problems that the professor assigned.”
As one example, DeNero summarized findings from 61A Bot, an interactive AI assistant developed by a Berkeley team and deployed in the CS 61A course as a personalized homework tutor.
“The AI system described why their answer was wrong, not just telling them what was wrong, but conceptually what they did wrong, and what they needed to do next to solve the problem,” he said.
He described several other subsequent interventions that Berkeley computer science education researchers are investigating as relates to the use of AI tools to enhance learning both inside and outside the classroom. For example, students are now interviewed about their answer every time they complete a homework problem correctly.
At the end of the talk, DeNero responded to a few questions from audience members. The last person to speak was Max Nguyen, a student majoring in electrical and computer engineering who’s taking the CS 61A course this semester. He thanked DeNero and told the audience he was able to replicate the AI tutoring approach from that course to help him in an applied math class.
“Since using the ‘interviewing assistant’ approach in my applied math class, my understanding of the problems has gone up,” Nguyen said.
DeNero responded with a fist pump and an audible “yes!” from the podium.