Cosine's Genie: Refreshing the New Standard of AI in Software Engineering

2024-08-13

Cosine, the AI startup from the UK, has successfully raised $2.5 million in seed funding and proudly announced the creation of "the world's top AI software engineer" - Genie. This achievement is not just empty talk! Genie has performed exceptionally well in the industry-recognized SWE-Bench, a benchmark test for evaluating AI model software engineering skills, with an outstanding score of 30.08%, surpassing the previous high score of 19.27% held by Factory Code Droid. What's even more remarkable is that Genie's performance not only surpasses many well-known AI models, such as Devin with only 13.8% and OpenAI GPT-4 with a score of 12.47% in the same test, but also demonstrates its unparalleled strength. In the development of Genie, Cosine has taken a unique approach by focusing on simulating the reasoning ability of human software engineers. They have extensively trained Genie using an exclusive dataset that meticulously records the problem-solving process of real-world software engineers, ensuring that Genie can "learn and apply" effectively. Alistair Pullen, the CEO of Cosine, excitedly stated, "We have made a crucial breakthrough in digitizing the human reasoning process, enabling our AI model to handle tasks far beyond the capabilities of current software development teams." Genie not only has the ability to independently solve code errors, build functionalities, and refactor code, but also seamlessly collaborates with human developers to enhance work efficiency. Through integration with GitHub, Genie can directly import issues and generate detailed work instructions, optimizing the entire development process. Its precise document recognition and in-place editing capabilities empower development teams, significantly improving productivity. The seed round financing was led by SOMA and Uphonest Capital, with investments from Lakestar and Focal, among other investment firms. Ellen Ma, a partner at Uphonest Capital, has full confidence in Cosine, stating, "Cosine has successfully taught AI to reason, bringing true AI partnership to enterprises." Since its establishment in 2022, Cosine has rapidly grown with the support of Y Combinator and has established operational centers in San Francisco and London. Looking ahead, Cosine plans to further expand Genie's capabilities, covering more programming languages and frameworks, while also exploring miniaturized models to handle simple tasks and building more powerful models to tackle complex challenges. With the rapid development of AI in the field of software engineering, Cosine's Genie has undoubtedly become the new benchmark in the industry. However, facing continuous investment and technological innovation, competition in this field will also become increasingly fierce, and more breakthroughs and progress are eagerly anticipated.