Andrei Karpathy, a former head of AI at Tesla and a researcher at OpenAI, is launching Eureka Labs, an “AI-native” education platform. In tech terms, that usually means built from the ground up with AI at its core. While Eureka Labs' ambitions for AI are lofty, the company is starting with a more traditional education approach.
San Francisco-based Eureka Labs, which Karpathy registered as a limited liability company in Delaware on June 21, aims to leverage recent advances in generative AI to develop AI teaching assistants that can guide students through course material.
According to Karpathy, who posted the news on X, Eureka Labs envisions an AI assistant or personality working alongside human teachers to help “anyone learn anything.” Teachers would still design course materials, but the AI assistant would provide support. The startup doesn't appear to have yet built or tested the effectiveness of incorporating an AI assistant in the classroom. At least one study from Georgia State University found that AI teaching assistants helped some students improve their grades.
Karpathy's post hints at a future where assistants are based on actual humans, like Meta's quirky celebrity chatbot or Character AI's character chatbots. X's post, which is also reflected in Eureka's bare-bones new website, doesn't offer much information about the new startup, including whether it's simply a MOOC with a chatbot or a product Karpathy hopes to implement in places like high schools.
Karpathy did not respond to TechCrunch's request for more information.
Alongside X's post announcing the news, Karpathy also posted a supposedly AI-generated image of a futuristic school, with a spaceship-like building, solar panels everywhere (even on the floor) and a smiling girl with… are those three hands?
While saying that Eureka Labs aims to build an AI teaching assistant, Karpathy said the new venture's first product will be an AI course, LLM101n, an undergraduate-level class that will help students train their own AI. Karpathy said this mini-me is like a smaller version of the AI teaching assistants that Eureka Labs wants to build and scale. On X and Eureka's stripped-down new website, the AI pioneer wrote that the course materials will be available online and the startup will run both digital and physical cohorts that learn the materials together.
The link to this AI course leads to a GitHub repository that suggests it's a different kind of course than the one Eureka Labs is advertising. Rather than “How to build an AI assistant,” the link leads to how to build a “Storyteller AI Large Language Model (LLM).”
“If we hold hands [to] “Craft, refine, and illustrate small stories with AI,” reads the page copy. The class promises to teach eager AI students how to “build everything from the basics to a functional web app similar to ChatGPT, end-to-end, from scratch in Python, C, and CUDA, with minimal computer science prerequisites.”
It's unclear which course Eureka Labs plans to introduce first, but neither appears to be finished: a note posted to its GitHub page said that building the courses will take time, and there's no specific timeline.
It's also unclear whether Karpathy bootstrapped Eureka Labs or was backed by investors, or what the startup's business model is. There were no public investment documents related to Eureka Labs. Karpathy is the only signer of the startup's limited liability company application filed with the California Secretary of State, and he did not disclose whether he is working with other prominent leaders in the AI field.
At X, Karpathy said Eureka Labs is the culmination of his passion for AI and education over the past 20 years. Karpathy taught deep learning for computer vision at Stanford University before leaving in 2015 to co-found OpenAI. Two years later, Karpathy moved to Tesla to lead the company's AI team and led the computer vision team for Tesla Autopilot. Autopilot is Tesla's advanced driver assistance system that uses cameras to capture environmental data and perform certain driving tasks, such as cruise control and automatic steering.
Karpathy left Tesla in 2022 and returned to OpenAI to lead a small team related to ChatGPT. In February, the research scientist also left his position at OpenAI. In both cases, Karpathy claimed there was no drama or fallout that led to his decision to step down.
Throughout his career at Tesla and OpenAI, Karpathy has continued to work as an educator, currently teaching an online course called “Neural Networks: Zero to Hero” that teaches students how to build neural networks from scratch in code. Karpathy also runs a YouTube channel where he posts semi-regular lectures on LLM and AI topics.