news

한어Русский языкEnglishFrançaisIndonesianSanskrit日本語DeutschPortuguêsΕλληνικάespañolItalianoSuomalainenLatina

Machine Heart Report

Synced Editorial Department

Karpathy: Tesla and OpenAI were both my side jobs before, but now it’s my main job.

Just now, Andrej Karpathy announced his next destination on the X platform: to start an artificial intelligence + education company called Eureka Labs.

I am very happy to share with you that I am starting an AI + Education company called Eureka Labs. The announcement is as follows:

We are Eureka Labs, and we are building a new kind of school that is AI-native. How can we achieve the ideal experience of learning new knowledge? In physics, for example, we can imagine studying high-quality course material with Feynman, who guides you every step of the way. Unfortunately, passionate, educated, infinitely patient subject matter experts who speak all the world’s languages ​​are also too scarce to tutor 8 billion of us in person on demand. However, with recent advances in generative AI, this learning experience feels attainable.

Teachers are still responsible for designing course materials, but those materials are supported, leveraged, and extended by AI teaching assistants that are optimized to help guide students through those materials. This symbiotic model of "teacher + AI" can run entire courses on a common platform. If we succeed, anyone can easily learn anything, thereby expanding the reach (large numbers of people learning something) and depth of education (any one person can learn a large number of subjects, which is beyond the scope of what is currently possible without assistance). Our first product will be the world's best AI course, LLM101n.

This is an undergraduate course that teaches students how to train their own AI, much like a scaled-down AI TA course. Course materials will be available online, but we also plan to offer both digital and physical classes. Today, we are focusing on LLM101n, but we look forward to a future where AI is a key technology for advancing human potential.

Eureka Labs is the culmination of my passion for AI and education that has spanned over two decades. My interest in education has always been with me, from Rubik’s Cube tutorials on YouTube to CS231n at Stanford to the recent “Zero-to-Hero AI” series. My work in AI has ranged from academic research at Stanford to actual products at Tesla to AGI research at OpenAI. All of my work in both fields to date has been part-time, on the sidelines of my “full-time” job. So I’m excited to dive in and create something great professionally and full-time. While it’s still early, I wanted to announce the launch of the company so that I can create it publicly instead of keeping it a dark secret.

Andrej Karpathy has many identities: a disciple of Fei-Fei Li, a founding member and research scientist of OpenAI, former senior director of AI at Tesla, and a well-known blogger in YouTube's "Learning Zone".

In February this year, Karpathy announced his departure from OpenAI, saying "there was no dramatic conflict, I just wanted to try my own personal project." Since then, his next move has become the focus of public attention.

This reminds us of another well-known AI scholar, Andrew Ng. As the founder of the Google Brain project and former chief scientist of Baidu, Andrew Ng co-founded the online education platform Coursera with Daphne Koller, also a Stanford professor. Perhaps, all the bigwigs engaged in theoretical research have an educational dream in their hearts.

New company focuses on AI+education

Eureka Labs’ first product is an undergraduate-level course, LLM101n, designed to help students train their own AI models.

In June, Andrej Karpathy simply uploaded the LLM101n course catalog on GitHub, and it quickly went viral.

Specifically, LLM101n starts with the basics of language modeling and machine learning, and then goes to multimodality, RLHF, and model deployment, aiming to build a Storyteller AI large model (LLM) to create, improve, and illustrate short stories using AI, covering everything from the basics to a runnable web application similar to ChatGPT, and using Python, C, and CUDA to build projects from scratch, with minimal computer science prerequisites. This course will give students a relatively in-depth understanding of AI, LLM, and deep learning.

Currently, three weeks after uploading, the LLM101n project has received 17k Stars.

Under Karpathy’s post announcing the new company, Google chief scientist Jeff Dean sent his sincere blessings.

In the comments section, Karpathy also explained the origin of the name Eureka Labs: Eureka (derived from ancient Greek) is a wonderful feeling after understanding something, as if something clicked in the brain. Their goal is to inspire those moments in people's hearts. The reason why Labs was added to the name is that Eureka as a single word has been used by others (the service discovery framework developed by Netflix), and he has always wanted a lab...

Currently, the company has opened an official Twitter account and GitHub repository.

GitHub address: https://github.com/EurekaLabsAI

Active figures in the field of AI

Personal homepage: https://karpathy.ai/

From 2005 to 2009, Andrej Karpathy studied for his undergraduate degree at the University of Toronto in Canada, majoring in computer science and physics and minoring in mathematics. Here, he first came into contact with deep learning and listened to Hinton's course.

From 2009 to 2011, Karpathy studied for his master's degree at the University of British Columbia in Canada. His advisor was Michiel van de Panne, a professor in the Department of Computer Science, and his research focused on machine learning for agile robots in physical simulations.

From 2011 to 2016, Dr. Karpathy studied at Stanford University under the tutelage of the famous AI scholar Fei-Fei Li, focusing on the study of convolutional/recurrent neural networks and their applications in computer vision, natural language processing and cross-domains. During this period, he designed and served as the main lecturer of Stanford's first deep learning course "CS231n: Convolutional Neural Networks and Visual Recognition".

At the same time, Karpathy also had three internships. In 2011, he interned at the early stage of Google Brain, working on large-scale unsupervised learning in the field of video. In 2013, he interned at Google Research again, working on large-scale supervised learning of YouTube videos. In 2015, he interned at DeepMind, working on the deep reinforcement learning team.

Currently, Karpathy's papers have been cited 66,426 times on Google Scholar. Among them, his second most cited paper, "Large-scale Video Classification with Convolutional Neural Networks", was included in the CVPR 2014 Oral.

Karpathy is a founding member and research scientist of OpenAI. Just over a year after the founding of OpenAI, in 2017, Karpathy accepted Musk's invitation to join Tesla, replacing Chris Lattner, then head of Tesla Autopilot and creator of Apple's Swift language and LLVM compiler.

In five years, Karpathy single-handedly facilitated the development of Autopilot. As Tesla gradually expanded from the initial autonomous driving to the broader field of artificial intelligence, Karpathy was also promoted to Tesla's senior director of AI, reporting directly to Musk.

Passion for education

In fact, during his previous career changes, Karpathy had already expressed his interest in "engaging in education."

Karpathy left Tesla in July 2022. In a subsequent interview, he said it was a difficult decision. Although Tesla has not yet fully achieved automated driving, the R&D team can already develop on its own. This departure also gave him an opportunity to re-examine his love for artificial intelligence, open source, and education.

Eight months after leaving, in February 2023, Karpathy announced that he had rejoined OpenAI. During this period, in addition to completing OpenAI's research work, Karpathy also did some personal projects, including recording educational videos and developing an open source project about Llama 2 - llama2.c.

In February of this year, after an internal struggle at OpenAI, Karpathy resigned from OpenAI again and has been working on some open source courses.

Now, he has officially devoted himself to the education career he loves. We can expect Karpathy to bring more excellent course resources to the field of AI.

For more of Karpathy's course projects, see Further Reading.