OpenAILeah Belsky on how AI is transforming education — the OpenAI Podcast Ep. 4
CHAPTERS
- 0:00 – 0:36
AI in education: brain rot, cheating, or a new learning companion?
Andrew Mayne frames the episode’s central tensions: whether ChatGPT harms learning, enables cheating, or becomes a powerful educational tool. The guest lineup is introduced—Leah Belsky (Head of Education at OpenAI) plus two student power users—to explore real classroom and life impacts.
- •Core questions: “brain rot,” cheating, and learning outcomes
- •ChatGPT positioned as both tool and platform affecting education
- •Set-up for perspectives from leadership and students
- •Promise of practical examples (Study Mode, day-to-day use)
- 0:36 – 1:56
Leah Belsky’s journey: from global education access to OpenAI’s “moonshot”
Leah explains her background (World Bank, Coursera) and why OpenAI’s mission pulled her in. She shares how leadership framed her role as pursuing the biggest possible transformation: an AI tutor/companion available to everyone.
- •15 years in education focused on accessibility
- •OpenAI’s mission and people as key motivators
- •“Go after the moonshot”: lifelong tutor/companion vision
- •Equity emphasis: making the product available worldwide
- 1:56 – 3:48
ChatGPT as the world’s largest learning platform—and why countries are leaning in
Leah argues learning is now one of ChatGPT’s top use cases at massive scale. She describes growing global demand, including government-level interest in deploying AI as educational infrastructure tied to economic competitiveness.
- •Learning is a top ChatGPT use case at scale (hundreds of millions of users)
- •Teachers adopt it for admin relief and classroom support
- •OpenAI for Countries program sparks ministry outreach
- •Countries see AI in schools as part of AI-economy readiness
- 3:48 – 5:27
Universities: equal access, student trust, and the post-COVID edtech hangover
The discussion shifts to higher education adoption barriers and successes. Leah highlights institutions’ push for equitable access and collaboration, but notes students’ privacy concerns—especially shaped by surveillance-like remote learning experiences during COVID.
- •Campus AI as “core infrastructure” to avoid pay-to-play inequality
- •Faculty want shared best practices across institutions
- •Students hesitate if they think school AI is monitored
- •COVID-era edtech experiences increased mistrust and fatigue
- 5:27 – 6:48
From AI detectors to better policy: redesigning assessment instead of policing
Andrew and Leah critique early institutional reactions—especially unreliable AI detectors that damaged trust. Leah advocates moving beyond enforcement toward clear usage policies and rethinking homework and assessment design.
- •AI detectors can falsely accuse and erode trust
- •Early response: hiding/banning/policing rather than adapting
- •Need explicit guidelines: when AI use is appropriate vs not
- •Reinvent assessment and assignments for an AI-available world
- 6:48 – 7:35
Study Mode: how it works and why OpenAI built it
Leah introduces Study Mode as a shift from answer-delivery to guided learning. She describes Socratic dialogue, personalization, quizzes, and scaffolding as early steps toward ChatGPT as a true tutor.
- •Study Mode aims to guide students to answers, not just provide them
- •Socratic questioning, follow-ups, and level-appropriate responses
- •Built-in prompts: quizzes and deeper exploration
- •Designed so students don’t need prompt-engineering to learn well
- 7:35 – 10:04
Study Mode’s origin story: India, tutoring economics, and learning-science “golden examples”
Leah explains that Study Mode was inspired by the high cost of tutoring and strong learner demand in India. She details a learning-science-driven development process: designing response schemas, collecting expert examples, and iteratively training toward better pedagogy.
- •High household spend on tutors motivated scalable alternatives
- •Product built with learning science and pedagogical experts
- •“Golden examples” used to train ideal tutoring interactions
- •Positioned as a beginning toward richer multimodal tutoring
- 10:04 – 11:34
AI tutoring as a confidence engine—and an accessibility unlock
Leah emphasizes AI’s biggest near-term educational impact outside the classroom: providing “adult support” where teachers, tutors, or parental help are unavailable. She shares how students report increased confidence and tells a personal story about her dyslexic daughter using Voice Mode to access news and learning.
- •AI can equalize access to support and feedback outside school
- •Confidence and persistence increase when students can ask freely
- •Infinite patience and non-judgment lower barriers to learning
- •Accessibility story: Voice Mode helps a dyslexic child engage with the world
- 11:34 – 13:56
Workforce readiness: AI literacy, productivity, and coding as a new core skill
The conversation turns to employability: AI usage correlates with productivity and hiring preference. Leah and Andrew argue graduates need practical AI fluency, and Leah adds that basic coding (and debugging/understanding code) becomes more important as creation gets easier.
- •AI users show significant productivity gains
- •Employers increasingly prefer AI-skilled candidates
- •Institutions deploy AI to ensure workforce-relevant skills
- •Coding becomes a core literacy even with AI-assisted development
- 13:56 – 19:30
The “brain rot” debate: struggle, critical thinking, and using AI the right way
Leah responds to concerns that AI weakens learning by stressing that outcomes depend on usage patterns. She uses analogies (long division vs calculators; marathon training vs a scooter) to argue learning still requires productive struggle—something tools like Study Mode can encourage.
- •AI as a tool: benefits or harms depend on how it’s used
- •If used as an answer machine, learning declines
- •Learning requires struggle, processing, and feedback loops
- •Study Mode exists to nudge students toward deeper learning behaviors
- 19:30 – 21:25
Meet the students: what they study and how they think about learning
Andrew introduces Yabsera and Alaap, who describe their academic paths and motivations. Their backgrounds set up contrasting but complementary perspectives: communication-to-analytics and EECS hands-on building.
- •Yabsera: communication undergrad, business analytics master’s track
- •Exploration via GEs; creative + analytical skill blend
- •Alaap: EECS at Berkeley; maker mindset (circuits/code/building)
- •Positioning as engaged, reflective student users of AI
- 21:25 – 25:00
First AI experiences: from To Kill a Mockingbird essays to Shrek fan fiction
The students recount their “aha” moments with ChatGPT—initially as novelty, then as utility. Alaap describes seeing it write an entire essay (and resisting using it), while Yabsera shares playful early prompts that later evolved into academic use.
- •Early surprise: interactive AI that can generate full outputs
- •Immediate temptation and the ethics of use in schoolwork
- •Non-academic entry points can spark broader adoption
- •Discussion of why creating vs consuming still matters for learning
- 25:00 – 28:31
How professors are adapting: harder projects, reflection, and AI vs non-AI tracks
Both students describe shifts in instruction and assessment. They observe fewer rote definitions, more application-based questions, and in CS, optional AI-enabled tracks with tougher scopes and reflective writing to ensure learning.
- •Assignments shift from memorization to application and meaning
- •More open-format assessments in some classes
- •AI vs non-AI project tracks with increased difficulty for AI users
- •Reflections used to surface learning, process, and understanding
- 28:31 – 32:43
Trying Study Mode: side-by-side tests, research workflows, and learning by dialogue
The students evaluate Study Mode against regular chat, highlighting its questioning-first approach and built-in recall checks. They also share practical research habits—like constraining outputs to provided sources—and how Study Mode can reduce the need for heavy upfront parameter-setting.
- •Study Mode asks clarifying questions before teaching
- •Breaks topics down and periodically checks understanding
- •Research strategy: paste sources, constrain the model to them
- •Learning feels more rigorous via dialogue vs long answer dumps
- 32:43 – 41:03
ChatGPT vs social media: attention, intentional learning, and deep research quality
They compare AI use to social platforms, arguing ChatGPT enables more intentional, targeted exploration while social media encourages passive consumption. The discussion includes deep research as a step-change in information quality and academic usefulness.
- •Stepping back from TikTok/scrolling to regain agency and focus
- •ChatGPT used for specific questions; social media reserved for leisure
- •Deep research improves sourcing and depth vs general web browsing
- •Practical tip: request academic/merit-backed sources and citations
- 41:03 – 46:02
Cheating, fears, and advice: redefining integrity, avoiding dependency, and echo chambers
The group tackles academic integrity and the ambiguity of “cheating” in an AI era. They offer advice for students to avoid overreliance, and they surface broader fears: bypassing foundational learning and centralizing “truth” into feedback loops akin to social media echo chambers.
- •Cheating is being redefined as tools become ubiquitous
- •Risk: using AI to bypass fundamentals and then failing real evaluation
- •Advice: use AI to understand, not as a crutch for answers
- •Fear: centralized knowledge and ideological chatbot echo chambers
- 46:02 – 59:38
The future of learning with AI: hybrid teaching, mentorship, and agentic workplaces
The episode closes by exploring long-term trajectories: AI-delivered personalized content, with humans focused on mentorship, ethics, and social development. They anticipate more agent-like systems orchestrating complex work, while emphasizing the need for humans to stay in the loop and for education to cultivate adaptation.
- •Vision: AI may deliver much instruction via personalized, multimodal content
- •Human role persists in mentorship, ethics, and social learning
- •Agentic systems could orchestrate multiple job functions
- •“Adaptation” framed as future job security and educational goal
