Jacob E. Thomas argues that persona benchmarks can reward a historically plausible voice without testing whether its reasoning is grounded in what a person could have known at a particular time. He uses model portrayals of Alexander Hamilton to illustrate his proposed Miranda hypothesis about culturally dominant representations influencing generated personas.
Jacob E. Thomas proposes grounding the system in primary documents, anchoring it to a specific moment and involving domain experts in evaluation. He contrasts inspectable context-based configurations with fine-tuning, presenting his architectural claims as arguments rather than established results.
The proposed Prism experiment compares four periods of Abraham Lincoln’s life under three prompting conditions. A historian-designed rubric scores anachronism, documentary consistency and contextual plausibility. Jacob E. Thomas says the protocol is preregistered and has not been run at scale; the predicted outcomes remain to be tested.
Watch on YouTube
