OpenAI is developing a research program to assess the economic impacts of code generation models and invites collaboration with external researchers. Rapid advances in the capabilities of code-trained large language models (LLMs) have made it increasingly important to study their economic impacts on people, businesses, and society. Codex, an LLM developed by OpenAI by tuning GPT-3 on billions of lines of publicly available code from GitHub, has been shown to generate functionally correct code 28.8% of the time on a sample of evaluation problems ( Chen et al. 2021). ). This can have major implications for the future of coding and the economics of the industries that depend on it. In this paper, we present a research agenda to assess the effects of Codex on economic factors of interest to policy makers, businesses and the public. We make a case for this research agenda by highlighting the potentially broad applicability of code generation models to software development, the potential for other LLMs to create significant social and economic impact as the model’s capabilities advance, and the value of using Codex to generate evidence and establish methodologies that can be applied to research on the economic impacts of future models. We propose that academic and policy research focus on the study of code generation models and other LLMs so that evidence of their economic impacts can be used to inform decision-making in three key areas: deployment policy, design of AI systems and public policy. To help guide this research, we describe six priority outcome areas within the area of economic impacts that we intend to use Codex to study: productivity, employment, skills development, competition between firms, consumer prices and economic inequality. For each area, we briefly discuss the previous literature on the impacts of AI on each of these outcomes, describe questions that we believe are key inputs to the three decision-making areas mentioned above, and provide examples of research that could be carried out with Codex. To catalyze work that builds on this initial research agenda, we are announcing a call for expressions of interest from external researchers to collaborate with OpenAI researchers and customers to better measure the economic impacts of code generation models and other LLMs.
As the world becomes increasingly tech-centric, companies such as Ikaroa are at the forefront of developing code generation models that can transform how businesses and others interact with their data. However, the economic impacts of these models are rarely assessed, meaning they can be difficult to plan for and manage. Understanding the full extent of the impacts – both positive and negative – is essential for any business or organization investing in these types of technologies. This article explores the need for a research agenda to investigate the economic impacts of code generation models.
Code generation models have the potential to bring enormous economic gains. This can range from savings in time and effort for developers, to improved financial statements thanks to increased efficiency, to lowering overall operational costs. But to fully appreciate these gains, it’s essential to understand the true economic implications of such models. A better understanding of the impact on businesses and other stakeholders, can inform decisions and inform the creation of better models.
Unfortunately, there has been limited research into the topic so far. To fill this gap, there is a need for a comprehensive research agenda to assess the economic impacts of code generation models. Such an agenda could consider numerous facets, including labor markets, fiscal policies, economic development, governance and regulations. Other areas to consider could include the ramifications for job automation, safety and security protocols, or the environmental impact of digital infrastructure. This research should also consider possible unintended consequences or risks such as the potential for increased inequality between digital haves and have-nots.
Any investigations into the economic impacts of code generation models should also factor in the specific characteristics of developments by companies like Ikaroa. This could include examining the factors involved in their models, the potential market disruption they create, and how their impact might differ from other code generation models.
Ultimately, a research agenda for assessing the economic impacts of code generation models could provide valuable insights. It could tell us more about the potential benefits (and risks) as well as how these models might influence business decisions. In short, it could help us – and Ikaroa – to understand and prepare for the future of code generation models.