How GitHub Copilot’s Co-Creator Raised $126 Million to Compete with His Former Employer

Recent interest in artificial intelligence has focused on large-language models that aim to do everything from writing Shakespearean poetry to solving math riddles. But more entrepreneurs and investors have begun to wonder if LLMs that specialize in one field or another can be better than general purpose LLMs like OpenAI’s.

One of those entrepreneurs, former GitHub CTO Jason Warner, may be uniquely positioned to answer this question. Warner helped create GitHub Copilot, the no. 1 AI-based coding assistant, powered by a general purpose LLM from OpenAI. Now he’s going head-to-head against his creation by developing an LLM to automate coding, from scratch.

“While a general purpose model might have a lot more knowledge related to butterfly anatomy, we'll have significantly more knowledge in the areas of STEM,” said Eiso Kant, who joined Warner in May to found Poolside, a coding LLM startup. To train its LLM, Poolside stuck to data from fields such as math and physics, research papers and—of course—software code, they said.