On Tuesday, former OpenAI researcher Andrej Karpathy introduced the formation of a new AI studying platform referred to as Eureka Labs. The enterprise goals to create an “AI native” academic expertise, with its first providing targeted on educating college students how to build their very own massive language mannequin (LLM).
“It’s nonetheless early days however I wished to announce the company in order that I can build publicly as an alternative of protecting a secret that is not,” Karpathy wrote on X.
While the concept of utilizing AI in training is not significantly new, Karpathy’s method hopes to pair expert-designed course supplies with an AI-powered educating assistant based mostly on an LLM, aiming to present customized steering at scale. This mixture seeks to make high-quality training extra accessible to a worldwide viewers.
The platform’s inaugural course, LLM101n, targets an undergraduate-level viewers. It will stroll college students by means of the method of coaching an AI system referred to as a “Storyteller AI Large Language Model” that will “create, refine and illustrate little tales.” Eureka Labs will provide this course on-line at first, then by means of in-person teams sooner or later.
Karpathy no stranger to AI training
Karpathy’s roots in AI go deep. In 2015, he obtained a PhD from Stanford University beneath pc scientist Dr. Fei-Fei Li. He was one of many founding members of OpenAI as a analysis scientist, then moved to Tesla to change into its senior director of AI between 2017 and 2022. In 2023, Karpathy rejoined OpenAI for a 12 months, departing in February.
This new enterprise builds on Karpathy’s observe file on AI training. Over the previous 12 months, Karpathy has posted a number of extremely regarded tutorials overlaying AI ideas on YouTube, together with an tutorial video about how to build an LLM from scratch, which at present has 4.5 million views. The movies have showcased his means to break down complicated subjects for a broader viewers.
“@EurekaLabsAI is the end result of my ardour in each AI and training over ~2 many years,” Karpathy wrote on X. “My curiosity in training took me from YouTube tutorials on Rubik’s cubes to beginning CS231n at Stanford, to my more moderen Zero-to-Hero AI sequence. While my work in AI took me from tutorial analysis at Stanford to real-world merchandise at Tesla and AGI analysis at OpenAI. All of my work combining the 2 thus far has solely been part-time, as facet quests to my ‘actual job,’ so I’m fairly excited to dive in and build one thing nice, professionally and full-time.”
Eureka Labs’ imaginative and prescient extends past its preliminary AI course, hinting at a broader curriculum that might span numerous topics. “Today, we’re heads down constructing LLM101n,” the announcement reads, “however we glance ahead to a future the place AI is a key expertise for growing human potential. What would you like to be taught?“
