claim
active
claim:noise-in-uncertainty-dynamics-is-largely-an-artifact-of-sampling-rather-than-llm-sensitivity-to-each-token-or-stepNoise in uncertainty dynamics is largely an artifact of sampling rather than LLM sensitivity to each token or step.
Abstract claim.
Source paper
extracted_from(2026) · Eric Bigelow · Amir Zur · Satchel Grant · Tal Haklay +6
Neighborhood — ranked by edge-count
Papers (1)
paper
Findings (3)
finding
- Noise matches null.
- Main result 1.
- Supports multinomial model.
Concepts (1)
concept
- Multinomial sampling noise modelsupportsCounts c_t ~ Multinomial(S,o_t).
Questions (1)
question
- Motivates model.
Related by similarity (8)
cosine ≥ 0.65 · no typed edgeEntities in the same semantic neighborhood but without a typed relation to this one — candidates for new edges or unrecognized duplicates.
- Connection between active inference neuronal dynamics and predictive processing theory.
- Do we assume that the model knows something we don't, or are these interference weights contributing noise?question0.775Motivating question.
- What is the right statistical model for explaining uncertainty dynamics in text generation?question0.775Central scientific question.
- NIS+ learns macro-dynamics matching ground-truth SIR dynamics from noisy micro-level data.finding0.773Experimental result from Yang et al. (2023) reported in the survey.
- Central interpretive claim of the paper supported by multiple convergent analyses
- Fig. 7d,e.
- Opening framing.
- Justifies using internal indicators rather than behavioral tests for AI consciousness