Scholarlink

<p>Large language models (LLMs) increasingly function as infrastructure mediating knowledge access, making paradigm choice between open-source and closed-source development relevant to questions of responsible deployment. This article examines whether open-source development offers normative advantages for ethically deploying LLMs. We first establish that technical considerations alone are insufficient to determine paradigm choice: leading open-weight models achieve approximate performance parity with closed-source alternatives, while fully open-source models demonstrate a narrowing but real performance gap, with neither paradigm demonstrating clear overall superiority. Given this approximate parity, we argue paradigm choice warrants evaluation on normative grounds. Adopting Sen and Nussbaum’s capabilities approach, we operationalize four independently grounded dimensions: transparency (epistemic capability), reproducibility (adaptive capability), community contributions (participatory capability), and auditability (contestation capability). We argue that open-source development expands substantive freedoms along each dimension as a matter of normative validity, independent of whether enabling conditions are currently met in any given context. The enabling conditions examined in Sect.&#xa0;<InternalRef RefID="Sec20">6</InternalRef>—computational infrastructure, technical capacity building, sustainable funding, and governance mechanisms—bear on the implementation of this argument, not its justification. We acknowledge contexts where alternative paradigms may prove appropriate and identify investments required to actualize open-source’s capability-expanding potential.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

The open-source advantage in large language models (LLMs)

  • 期刊论文

  • 发表日期：

    2026年7月9日

查看全文
  • Jiya Manchanda,
  • Laura Boettcher,
  • Matheus Westphalen,
  • Jasser Jasser

摘要

Large language models (LLMs) increasingly function as infrastructure mediating knowledge access, making paradigm choice between open-source and closed-source development relevant to questions of responsible deployment. This article examines whether open-source development offers normative advantages for ethically deploying LLMs. We first establish that technical considerations alone are insufficient to determine paradigm choice: leading open-weight models achieve approximate performance parity with closed-source alternatives, while fully open-source models demonstrate a narrowing but real performance gap, with neither paradigm demonstrating clear overall superiority. Given this approximate parity, we argue paradigm choice warrants evaluation on normative grounds. Adopting Sen and Nussbaum’s capabilities approach, we operationalize four independently grounded dimensions: transparency (epistemic capability), reproducibility (adaptive capability), community contributions (participatory capability), and auditability (contestation capability). We argue that open-source development expands substantive freedoms along each dimension as a matter of normative validity, independent of whether enabling conditions are currently met in any given context. The enabling conditions examined in Sect. 6—computational infrastructure, technical capacity building, sustainable funding, and governance mechanisms—bear on the implementation of this argument, not its justification. We acknowledge contexts where alternative paradigms may prove appropriate and identify investments required to actualize open-source’s capability-expanding potential.