Degradation of openness? A comparative study on the evolution of open source in software and large language models
Year: 2025
Type: dissertation
Abstract: This thesis explores how the enablers and challenges of open source phenomenon in the context of large language models (LLMs) differ to those that influenced the evolution of open-source software (OSS). While open source is well-established in OSS, its application within the rapidly evolving field of generative AI and LLMs remains underexplored. The concept of open source was examined through re-sources, processes, phenomena, and technical requirements. Using inductive approach, empirical data was gathered with semi-structured interviews with 16 experts ranging from developers, researchers, open-source com-munity members, and legal professionals. The findings reaffirm key enablers of open source, such as collaboration and re-source accessibility, but highlight that resource intensity, the opaque nature of LLMs, legal uncertainties, and commercial pressures constrain openness. Consequently, the concept of open source requires adaptation to fit the unique context of LLMs, leading to the introduction of Open Source LLMs (OLLMs), a concept evolving in response to legal, technological, and market dynamics. For managers, the study emphasizes the need of sustainable business models in LLM development, identifying four dominant approaches: selling access, selling closed models, creating ecosystems, or offering services. Openness can serve as a strategic tool, balancing proprietary interests with the benefits of releasing, guided by a framework that aligns openness levels with user needs. The study’s theoretical contribution includes the OLLM concept and a framework for understanding the evolution of both OSS and LLMs. It also highlights that, given the unique challenges, openness in LLMs may reinforce existing power structures rather than create a more inclusive and participatory environment.
Source: Aaltodoc (Aalto University)
Author: Juntura, Paula
Language: English
Subfield: Computer Science Applications
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W6999656620