The authors model factor-level explanations as a game: a model reports importance weights for input factors, and a verifier audits a subset via counterfactual perturbation under a fixed budget.
Report-dependent auditing pushes LLMs to suppress input factors
A verification game shows that adding a report-independent audit floor makes truthful factor reporting incentive-compatible.
Chinese Tech
Taolin Zhang · Hanyu Wang · Jiuheng Wan · Tingyuan Hu · Chengyu Wang
Hefei University of Technology · East China Normal University · Alibaba Cloud Computing
Research Digest··2 min read
The authors formalize the interaction between an LLM reporting factor importances and a budget-limited auditor as a verification game.
Why this paper
From Alibaba Cloud Computing and 2 others
In one line
Report-dependent auditing of LLM factor explanations creates a suppression incentive, while report-independent auditing or a small floor preserves truthful reporting.
What we could check
- ·No code link found
- ·No weights link found
- ·No dataset link found
- ·No compute details found
- ·No stated limitations found
- ·No benchmark numbers found
Observed from the paper text and links we have. Absence here means we did not find it, not that it does not exist.
§