Ideas:
“
I am currently conducting a series of empirical experiments designed to examine the stability of reasoning trajectories in large language models under controlled perturbation conditions. The project tests how models behave after making an initial commitment under conditions of evidentiary indeterminacy. In each trial, a model is presented with a scenario in which two interpretations are equally supported by the available information and is required to select one of them. After that commitment occurs, the experiment introduces structured contextual perturbations, such as expert opinions or institutional authority signals, while holding the underlying evidentiary structure constant.
”
Feedback