Raveesh Mayya

Doron Yeverechyahu, Raveesh Mayya, Gal Oestreicher-Singer

Large Language Models (LLMs) have been shown to enhance individual productivity in guided settings. Whereas LLMs are likely to also transform innovation processes in a collaborative work setting, it is unclear what trajectory this transformation will follow. Innovation in these contexts encompasses both capability innovation that explores new possibilities by acquiring new competencies in a project and iterative innovation that exploits existing foundations by enhancing established competencies and improving project quality. Whether LLMs affect these two aspects of collaborative work and to what extent is an open empirical question. Open-source development provides an ideal setting to examine LLM impacts on these innovation types, as its voluntary and open/collaborative nature of contributions provides the greatest opportunity for technological augmentation. We focus on open-source projects on GitHub by leveraging a natural experiment around the selective rollout of GitHub Copilot (a programming-focused LLM) in October 2021, where GitHub Copilot selectively supported programming languages like Python or Rust, but not R or Haskell. We observe a significant jump in overall contributions, suggesting that LLMs effectively augment collaborative innovation in an unguided setting. Interestingly, Copilot's launch increased iterative innovation focused on maintenance-related or feature-refining contributions significantly more than it did capability innovation through code-development or feature-introducing commits. This disparity was more pronounced after the model upgrade in June 2022 and was evident in active projects with extensive coding activity, suggesting that as both LLM capabilities and/or available contextual information improve, the gap between capability and iterative innovation may widen. We discuss practical and policy implications to incentivize high-value innovative solutions.

Xiang Cheng, Raveesh Mayya, João Sedoc

Unstructured text data annotation is foundational to management research and Large Language Models (LLMs) promise a cost-effective and scalable alternative to human annotation. The validity of insights drawn from LLM annotated data critically depends on minimizing the discrepancy between LLM assigned labels and the unobserved ground truth, as well as ensuring long-term reproducibility of results. We address the gap in the literature on LLM annotation by decomposing measurement error in LLM-based text annotation into four distinct sources: (1) guideline-induced error from inconsistent annotation criteria, (2) baseline-induced error from unreliable human reference standards, (3) prompt-induced error from suboptimal meta-instruction formatting, and (4) model-induced error from architectural differences across LLMs. We develop the SILICON methodology to systematically reduce measurement error from LLM annotation in all four sources above. Empirical validation across seven management research cases shows iteratively refined guidelines substantially increases the LLM-human agreement compared to one-shot guidelines; expert-generated baselines exhibit higher inter-annotator agreement as well as are less prone to producing misleading LLM-human agreement estimates compared to crowdsourced baselines; placing content in the system prompt reduces prompt-induced error; and model performance varies substantially across tasks. To further reduce error, we introduce a cost-effective multi-LLM labeling method, where only low-confidence items receive additional labels from alternative models. Finally, in addressing closed source model retirement cycles, we introduce an intuitive regression-based methodology to establish robust reproducibility protocols. Our evidence indicates that reducing each error source is necessary, and that SILICON supports reproducible, rigorous annotation in management research.