OpenAI Cofounder Launches New Open Source LLM ‘Nanochat’
It’s been over a year since Andrej Karpathy, co-founder of OpenAI, left the company. In the time since his departure, Karpathy has become a vocal advocate for ‘vibe coding’—a term he coined to describe the practice of using AI tools to automate coding tasks. However, his recent project, Nanochat, has challenged that narrative. Despite his public support for AI-assisted coding, Karpathy admits that he wrote the entire model by hand, rejecting his earlier approach of relying on AI tools like Claude and Codex.
Nanochat, according to Karpathy, is a ‘minimal, from scratch, full-stack training/inference pipeline’ that is designed to let anyone build a large language model with a ChatGPT-style chatbot interface in a matter of hours and for as little as $100. Karpathy said the project contains about 8,000 lines of ‘quite clean code,’ which he wrote by hand—not necessarily by choice, but because he found AI tools couldn’t do what he needed. He explained that while he attempted to use AI tools for assistance, they ‘just didn’t work well enough at all and net unhelpful.’
The release of Nanochat raises questions about the practical limits of AI-assisted coding. While Karpathy’s advocacy for vibe coding has influenced the broader tech community, his latest project demonstrates that, at least in some cases, human coding remains essential. The project’s open-source nature may encourage further innovation in the field, allowing developers to experiment with building their own LLMs without the need for extensive resources.
Karpathy’s admission that he manually coded Nanochat also highlights the ongoing debate within the tech industry about the role of AI in software development. While some argue that AI tools can streamline and enhance the coding process, others, like Karpathy, remain skeptical. His project serves as a reminder that, despite the rapid advancements in AI, human expertise and manual coding are still crucial components of modern software development.
The release of Nanochat has sparked discussions among developers and researchers about the future of AI-assisted programming. As more tools emerge to help automate coding tasks, projects like Nanochat provide a counterpoint, emphasizing the value of human input in the development of complex systems. Karpathy’s project may inspire a new wave of open-source initiatives aimed at making large language model development more accessible to a wider audience of developers.