Apply for community grant: Academic project (gpu and storage)
Dear Hugging Face Support Team,
I hope this email finds you well.
We are a team of researchers from the Nuffield Centre for Experimental Social Science (CESS) at the University of Oxford, and we would like to request a community grant, including GPU access and persistent storage, to support an innovative project: A large-scale leaderboard comparing the performance of various LLMs (both open- and closed-source) across a wide range of social science experiments.
Project Overview
Our goal is to develop a leaderboard that systematically evaluates LLMs on their ability to replicate human behavior in experimental social science settings. These experiments include:
- Randomized Controlled Trials (RCTs)
- Behavioral Experiments (e.g., public goods games, trust games)
- Field & Natural Experiments
- Survey & Social Network Experiments
- Psychological Experiments (e.g., conformity, decision-making)
For each experiment type, LLMs will be evaluated using rigorous, domain-specific metrics widely recognized by the experimental social science community. Additionally, we will assess how closely their responses align with actual human behavior (i.e., ground truth).
Value Proposition:
This project will provide three key benefits:
- Standardized Benchmarking – An apple-to-apple comparison of LLMs across diverse social science tasks.
- Promoting Open-Source Models – Encouraging the adoption of transparent, reproducible AI in experimental research.
- Guiding Researchers – Providing valuable insights for social scientists on how LLMs can (or cannot) be used as proxies for human decision-making.
Our findings will be published in an academic paper to maximize outreach and impact within both the AI and social science communities.
Grant Request:
To successfully build and maintain this leaderboard, we kindly request:
- Access to Hugging Face's community GPU resources to efficiently run large-scale model evaluations.
- Persistent storage for dataset management and experimental results.
We believe this project aligns with Hugging Face’s mission to democratize AI research and advance its real-world applicability in the social sciences.
We greatly appreciate your time and consideration and look forward to your positive response. Please let us know if you require any further details.
Best regards,
Dr. Raymond Low
On behalf of the Nuffield Centre for Experimental Social Science (CESS), University of Oxford
I hope this message finds you well. I wanted to follow up on my previous request for a Hugging Face community grant to support an academic project aimed at developing a large-scale leaderboard that evaluates the performance of various LLMs (both open- and closed-source) across a diverse set of social science experiments.
I’d love to hear any updates or feedback on the request and whether any additional information is needed to move forward. Looking forward to your thoughts!
Best regards,
Dr Raymond Low
Nuffield Centre for Experimental Social Science (CESS), University of Oxford
Hi @iamraymondlow , thanks for the ping, but unfortunately, your use case is out of scope of the Space grant. My understanding is that all existing leaderboards have secured their own backend GPU resources independently. cc @clefourrier for more info.
Hi!
We sometimes provide small (lowest GPU) & temporary (<1month) grants to help PhD students setup their leaderboard as POC, and find sponsors for compute - your use case falls out of scope, as we don't provide the main source of compute for large scale leaderboards.
It would make sense to see if your university (Oxford) can fund the compute or run the models (that's what happened for the Hallucinations leaderboard for ex, running on the cluster of the Uni of Edinburgh iirc), or find an external sponsor. Another option would be to contact model providers to see if you can get access to endpoints for your research.
Good luck!