Eureka Labs

Eureka Labs is an artificial intelligence education company founded by AI researcher and educator Andrej Karpathy in 2024. The company was established to explore a new approach to education in which experienced human teachers work alongside AI Teaching Assistants to provide more accessible, personalised, and interactive learning experiences. Its central idea is to build an AI-native school where carefully designed course materials are supported by intelligent assistants that can guide learners, explain difficult concepts, answer questions, and provide help throughout the learning process.

The company was created around a challenge in traditional education: access to excellent teachers and subject-matter experts is limited. A skilled teacher may have deep knowledge of a subject, strong communication skills, and the patience to guide students through difficult topics, but cannot personally provide unlimited individual attention to every learner. Eureka Labs aims to use generative AI to make more of this teaching support available. Its proposed model combines human-created educational materials with AI assistance, allowing one carefully designed curriculum to support a larger number of students while giving individual learners more opportunities to ask questions and receive guidance.

Unlike a conventional online learning platform that mainly provides recorded videos, reading materials, and quizzes, Eureka Labs proposes a more interactive learning environment. The teacher remains responsible for designing the curriculum and determining what students should learn, while the AI Teaching Assistant helps students work through the material. The assistant is intended to provide explanations, clarify unfamiliar ideas, and guide learners through the learning process. This teacher-and-AI approach is designed to combine the knowledge and educational judgement of human instructors with the availability and responsiveness of AI systems.

Eureka Labs was founded by Andrej Karpathy, who is known for his work in artificial intelligence, deep learning, computer vision, and education. He was a founding member of OpenAI and later served as a senior AI leader at Tesla, where he worked on the company’s Autopilot and computer-vision technology. He also developed educational material that explains neural networks and machine learning through practical implementation. His experience in both AI research and technical education influenced the direction of Eureka Labs, particularly its interest in teaching people how modern AI systems work rather than limiting learning to the use of existing AI applications.

The company’s first announced educational product is LLM101n, an undergraduate-level course intended to teach students how to build their own language model. The course is designed around a project called a Storyteller, an AI system that can create, refine, and illustrate short stories. Rather than simply teaching learners how to interact with an existing chatbot, LLM101n is intended to guide them through the underlying concepts and practical steps involved in building a language model from the beginning.

The proposed course is particularly relevant to students and developers who want to understand how large language models work internally. Its curriculum is designed to cover the progression from simple language modelling to more advanced neural-network architectures and the development of a working AI application. The intended learning experience combines theoretical explanations with practical programming exercises, helping students connect mathematical and computational concepts with the behaviour of real AI systems.

The public LLM101n syllabus outlines 17 chapters covering important areas of machine learning and language-model development. The early chapters introduce bigram and n-gram language models, automatic differentiation, backpropagation, and multilayer perceptrons. These subjects provide a foundation for understanding how neural networks learn from data and how their parameters are adjusted during training. The course then progresses to attention mechanisms, softmax, positional encoding, and transformer architectures, which are central to modern language models.

Tokenisation is another important topic in the proposed curriculum. Students are expected to learn how text is converted into numerical representations that language models can process. The syllabus includes byte-pair encoding and other tokenisation concepts, helping learners understand how a model handles words, characters, and sequences of text. This subject is important because tokenisation influences how language models represent and generate written content.

The training section of LLM101n is intended to introduce students to model initialisation, optimisation, and training algorithms such as AdamW. It also covers the computational requirements involved in training neural networks. Topics include CPU and GPU execution, mixed-precision computation, and distributed training. These areas help explain why modern AI models require specialised hardware and how training workloads can be divided across multiple devices.

The proposed syllabus also includes dataset preparation and data loading. Students are expected to examine how training data is collected, organised, processed, and supplied to a model. Synthetic data generation is another listed topic, providing an introduction to methods for creating additional training examples. These subjects are important because the quality and structure of training data can influence a model’s performance and the kinds of outputs it produces.

Further chapters are planned around inference, quantisation, and key-value caching. These topics explain how a trained model generates responses and how its computational requirements can be managed during use. The curriculum also includes supervised fine-tuning, parameter-efficient fine-tuning, and methods such as LoRA. Students would learn how an existing model can be adapted to follow instructions or perform more specialised tasks.

The course outline also covers reinforcement learning and preference optimisation. These subjects introduce methods for improving model behaviour using feedback and reward-based training approaches. The final stages of the proposed course include deployment, API development, web application creation, and multimodal systems. The multimodal section introduces technologies involving images, audio, video, and diffusion-based models. This progression is intended to help learners understand not only how a language model is trained but also how it can become part of a functioning software application.

Programming is central to the LLM101n learning concept. The public syllabus identifies Python, C, and CUDA among the technologies and programming languages relevant to the course. Python is widely used for machine learning development, while C and CUDA are useful for understanding lower-level computation and GPU acceleration. The course is intended to help learners connect high-level machine-learning concepts with the underlying operations that make neural-network training possible.

Eureka Labs also proposes a collaborative approach to learning. Its original announcement explains that course materials would be available online, while digital and physical cohorts would allow groups of students to progress through the curriculum together. Cohorts can provide opportunities for discussion, collaboration, peer learning, and shared project work. This approach is intended to combine the flexibility of online education with some of the benefits of studying alongside other learners.

The AI Teaching Assistant is the central technology concept behind the company’s wider educational vision. It is intended to support students while they work through course materials, making it easier to ask questions at the point where confusion occurs. Instead of requiring every learner to follow exactly the same pace, an AI assistant could provide additional explanations and guidance based on the learner’s needs. The company also imagines using AI assistants that can communicate across different languages, helping make educational support more widely accessible. However, the announcement describes this as part of the company’s broader vision rather than a verified list of currently supported languages or a fully released multilingual product.

The proposed platform could eventually support courses in subjects beyond artificial intelligence. Eureka Labs describes a future in which subject-matter experts design high-quality curricula and AI assistants help students work through them. This approach could be applied to technical subjects such as physics, mathematics, computer science, and engineering, as well as other areas where students benefit from individual explanations and guided learning. The company’s first announced course focuses on building an AI model, but its wider ambition is to create an educational environment that can support many different subjects.

Eureka Labs’ approach also has implications for educators. Teachers could concentrate on developing strong course materials, organising the learning progression, and identifying important educational outcomes, while AI assistants help provide additional guidance to students. This could make it easier to offer more personalised support without requiring a teacher to respond individually to every routine question. The intended model does not remove the teacher from education; instead, it gives the teacher another way to support students and extend the reach of carefully designed teaching materials.

It is important to distinguish the company’s announced plans from its publicly available resources. Eureka Labs introduced LLM101n as its first planned course in July 2024. The public GitHub repository contains a syllabus and an explanation of the intended project, but its notice states that the course did not yet exist when the repository was archived in August 2024. The repository therefore provides useful information about the curriculum and learning objectives, but it should not be treated as proof that a complete, supported course is available for enrolment.

There is also an important development in the company’s leadership. In May 2026, Andrej Karpathy announced that he had joined Anthropic to work on pre-training research. He also stated that he remained interested in education and planned to return to that work in the future. This means prospective learners should distinguish Eureka Labs’ original educational vision and published course outline from a currently operating learning platform with confirmed enrolment, support, and course schedules.

Eureka Labs is therefore best understood as an AI-native education initiative focused on combining expert-designed curricula with AI teaching assistance. Its most clearly defined project is LLM101n, which aims to teach learners how to build a language model through practical programming and machine-learning exercises. Its wider concept could be relevant to educators, students, developers, and institutions interested in more personalised forms of AI-supported learning. The company’s published materials provide a clear picture of its educational direction, although the availability of a completed course and a fully operational AI teaching platform should be confirmed before learners rely on it as an active course provider.

Pricing

Eureka Labs’ original educational model describes online course materials alongside digital and in-person learning cohorts. The company did not establish a detailed public tuition schedule in its launch announcement.

• Online course materials: Intended to be available online as part of the LLM101n learning experience.
• Digital cohorts: Planned group-based learning format for students progressing through the course together.
• Physical cohorts: Planned in-person learning format that would allow students to study and collaborate.
• AI Teaching Assistant: Intended to support learners as they work through the educational materials.
• LLM101n: The announced first course focuses on building a language model from the fundamentals through training and deployment.

The online educational materials were presented as a central part of the company’s learning model, while cohort-based participation was planned as an additional learning format. Current enrolment arrangements and any cohort fees should be confirmed through official Eureka Labs announcements.

Review

Eureka Labs presents an interesting approach to online education by combining expert-designed course materials with AI Teaching Assistants. Its central idea addresses a genuine challenge in learning: students often need individual explanations and guidance, but access to experienced teachers is limited. An AI assistant could help make educational support more available by answering questions and guiding learners through difficult concepts while the teacher remains responsible for the curriculum.

The planned LLM101n course is particularly appealing to learners interested in understanding how language models are built. Its proposed syllabus covers a broad range of topics, including neural networks, transformers, tokenisation, model training, GPU computation, fine-tuning, inference, and deployment. The practical focus on building a small language model could help students develop a deeper understanding of AI than they might gain from simply using existing chatbot applications.

Another strength is the proposed combination of online learning and group cohorts. Digital materials offer flexibility, while digital and physical cohorts can provide opportunities for collaboration and discussion. This combination could make the learning experience more engaging than studying entirely through recorded lessons or isolated tutorials.

The educational concept also has potential beyond the first course. If the teacher-and-AI model is developed successfully, it could support many subjects and make specialised educational guidance more accessible. The idea of using AI to help teachers scale their expertise is relevant to schools, universities, independent educators, and online learning providers.

However, Eureka Labs has an important limitation: its original announcement describes a planned educational platform, not a complete product with a verified course catalogue and established enrolment process. The LLM101n repository was archived with a notice explaining that the course did not yet exist at that time. This makes it difficult to evaluate the finished learning experience, the quality of its AI Teaching Assistant, or the consistency of its course support.

The proposed syllabus is also technically demanding. Building a language model from scratch involves programming, mathematics, neural-network concepts, and an understanding of computing hardware. Although the course was intended to have relatively modest computer-science prerequisites, students would still need to invest considerable time in learning and debugging code. Beginners may require additional preparation before attempting the more advanced chapters.

Another consideration is the current status of the project. Andrej Karpathy joined Anthropic in May 2026 and stated that he planned to return to his education work in the future. This makes it particularly important to verify current development and course availability before treating Eureka Labs as an active education service.

Eureka Labs has a compelling educational vision and a technically ambitious first course. Its strongest appeal is to students and developers who want to understand how AI models are trained and deployed, rather than simply learn how to use existing AI tools. Its long-term potential depends on the completion of its planned courses, the development of its AI Teaching Assistant, and the availability of a supported learning experience.