Independent researcher measuring frontier LLM/VLM failure modes: context-driven valuation bias, author misattribution, miscalibrated confidence, and confabulation. Former first-author neuroscience researcher (PNAS 2007, Salk Institute); 20+ years of experimental design applied to AI behavior.
Current studies
- ai-halo-valuation-bias — six-model VLM study on how visual/text context inflates or deflates price estimates (N=4,604)
- llm-author-misattribution — six-model study on author-age misattribution, miscalibrated confidence, and post-hoc confession without correction (N=3,225)