[2409.08379] The Impact of Large Language Models on Open-source Innovation: Evidence from GitHub Copilot

PDF display of the paper entitled The Effect of Long Language Models on Open innovation Source: Guide from GitHub Copilot, by Doron Yifrichiaho and two other authors

PDF view

a summary:LLMS models have been shown to enhance individual productivity in the guided settings. While LLMS is also likely to convert innovation in the preparation of cooperative work, it is not clear the path that this transformation will follow. Innovation in these contexts includes all the innovation that explores new capabilities by obtaining new competencies in the project and the repetitive innovation that exploits the current institutions by enhancing known competencies and improving the quality of the project. Whether LLMS affects these two sides of cooperative work and to what extent is an open experimental question. The development of the open source provides an ideal environment for studying the effects of LLM on these types of innovation, as the nature of voluntary and open/cooperative contributions provides the greatest opportunity for technological increase. We focus on open source projects on GitHub by taking advantage of a natural experience on the selective effect of Github Copilot (LLM that focuses on programming) in October 2021, where Github Copilot supported programming languages ​​such as Python or Rust, but not R or Haskell. We note a big leap in the total contributions, indicating that LLMS effectively increases cooperative innovation in an ungroke environment. Interestingly, the launch of Copilot has increased the repetitive innovation that focuses on maintenance contributions associated with maintenance or getting rid of features much more than innovation by developing the code or adhering to the production of features. This contrast was more clear after the upgrade of the model in June 2022 and was evident in active projects with a wide coding activity, indicating that with the improvement of both the capabilities of LLM and/or context information available, the gap between the ability and repetitive innovation may expand. We discuss practical and political effects to stimulate innovative high -value solutions.