Powering virtual education for the classroom
OpenAI
Khan Academy is testing GPT-4 in a small pilot to power an AI tutor for students and teachers. It's an early look at AI actually helping in classrooms, not just chatting.
Based on reporting by OpenAI — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Khan Academy, the nonprofit behind millions of free lessons, is running a limited pilot with OpenAI's GPT-4 to see whether a large language model can act as a genuine classroom assistant rather than a novelty chatbot. The idea isn't new for the organization — founder Sal Khan has talked for years about a future tutor for every student — but GPT-4's jump in reasoning and math ability gives that idea a real engine for the first time.
The pilot puts the model to work on two fronts: helping students work through problems step by step without just handing over answers, and giving teachers a faster way to draft lesson plans, quizzes, and feedback. That distinction matters. Earlier chatbots tended to solve for X and stop there, which is useless for actual learning. GPT-4's improved handling of multi-step reasoning lets it hold back the answer and instead nudge a student toward finding it themselves, closer to how a human tutor operates.
Khan Academy is keeping the rollout small and deliberate, testing with a limited group of educators and learners before even hinting at a broader release. That caution is deliberate, and warranted. AI models still hallucinate, still get math wrong in confident-sounding ways, and a wrong answer delivered with total certainty is a bad look in a classroom setting where trust is everything. The pilot exists partly to find those failure points before students at scale ever see them.
There's also a quieter story here about who gets access to good tutoring. Personalized, one-on-one instruction has historically been expensive and unevenly distributed, tilted toward families who can pay for it. Khan Academy's entire mission has been chipping away at that gap with free video lessons, and an AI tutor that can adapt in real time to how a specific kid is stuck is a plausible next step in that same direction, assuming the accuracy problems get solved rather than papered over.
My take — AI-written commentary, not fact-checked reporting
I like this pairing more than most AI-in-education stories, mainly because Khan Academy has spent fifteen years building actual pedagogical credibility before bolting on a language model. The real test isn't whether GPT-4 can explain fractions, it's whether it admits when it's unsure instead of confidently misleading a twelve-year-old, and that's the part nobody in this industry likes to pilot loudly.
Read more about this at: OpenAI