Even the Inventor of 'Vibe Coding' Says Vibe Coding Can't Cut It
In a surprising about-face, OpenAI cofounder Andrej Karpathy has revealed that he wrote an entire open-source model by hand, contradicting his previous advocacy for "vibe coding," a practice that involves using AI tools to farm out coding projects.
Karpathy released his project, nanochat, earlier this week. According to him, it's a minimal, from-scratch full-stack training-inference pipeline designed to let anyone build a large language model with a ChatGPT-style chatbot interface in just hours and for as little as $100. The project contains approximately 8,000 lines of clean code, which Karpathy wrote by hand.
"I tried to use Claude Codex agents a few times but they just didn't work well enough at all and were unhelpful," Karpathy said in his post announcing the release of nanochat. "It's basically entirely hand-written (with tab autocomplete)."
This admission marks a significant departure from Karpathy's previous stance on vibe coding, which he had described as suitable for throwaway weekend projects. In fact, it was Karpathy who coined and popularized the term vibe coding in the first place.
Background and Context
Karpathy left OpenAI over a year ago, but his influence on the tech industry remains significant. Vibe coding, a concept he championed, involves using AI tools to automate coding tasks, allowing developers to focus on higher-level thinking.
However, Karpathy's recent experience with nanochat suggests that even he has come to realize the limitations of relying solely on AI for coding projects. "It's not just about throwing code at an AI and expecting it to work," Karpathy said in a separate interview. "There are nuances and complexities that require human judgment and expertise."
Implications and Real-World Applications
Karpathy's decision to write nanochat by hand has significant implications for the tech industry, particularly for developers working on large-scale projects. While AI tools can certainly streamline certain tasks, they may not always be able to replicate the level of complexity and nuance required in high-stakes coding.
As one developer noted, "Karpathy's admission is a wake-up call for the entire industry. We need to recognize that AI is not a replacement for human expertise, but rather a tool to augment our capabilities."
Next Developments
The release of nanochat marks an important milestone in the development of large language models and chatbots. As more developers begin to explore Karpathy's project, it will be interesting to see how they adapt his approach to their own work.
In the meantime, Karpathy's decision to write nanochat by hand serves as a reminder that even the most advanced AI tools have limitations. "It's time for us to rethink our reliance on vibe coding and focus on building more robust, human-centered solutions," one industry expert said.
*Reporting by Gizmodo.*
