Convergence Point Theory: Why LLM uncertainty is determined by the topic, not the model
Existing research on LLM response uncertainty has been looking in different directions. Hallucination, knowledge conflict, RLHF limitations, prompt sensitivity, calibration failure — these have all been studied separately, and I kept wondering why no o…