Introducing the 2025 MSc Thesis Prize Winners
Emerging academic work on LLM evaluation validity and AI moderation legitimacy may inform future APS thinking on AI assurance frameworks.
Key points
- Oxford Internet Institute's 2025 MSc thesis prizes recognise four students across AI, social media, and internet governance research.
- Two AI-relevant theses cover LLM benchmark validity and public legitimacy perceptions of AI content moderation approaches.
- Limited direct relevance to APS practitioners; useful as a signal of emerging academic thinking on AI evaluation and governance.
Implications for Australian agencies
- Monitor AI assurance and evaluation teams may want to monitor Kearns's forthcoming DPhil work on LLM benchmark validity, as it could inform approaches to AI capability assessment in government contexts.
Implications are AI-generated. Starting points, not advice — see methodology for how they're framed.
View original source
Copied.
"Introducing the 2025 MSc Thesis Prize Winners"
Source: Oxford Internet Institute – News
Published: 18 February 2026
URL: https://www.oii.ox.ac.uk/introducing-the-2025-msc-thesis-prize-winners/
The Oxford Internet Institute has announced its 2025 MSc Thesis Prize winners across Social Data Science and Social Science of the Internet programs. Two prizes are directly AI-relevant: Ryan Kearns's thesis on construct validity in LLM evaluations critiques the reliability of benchmark scores as measures of AI capability, and Maximilian Kroner Dale's thesis uses survey experiments to assess public legitimacy perceptions of participatory versus closed-door AI content moderation. The other prizes address social media and adolescent mental health policy evidence, and crisis-time narrative control. These represent early-stage academic contributions rather than policy-ready outputs.
Implications for Australian agencies:
- [Monitor] AI assurance and evaluation teams may want to monitor Kearns's forthcoming DPhil work on LLM benchmark validity, as it could inform approaches to AI capability assessment in government contexts.
Retrieved from SIMS, 18 July 2026.