claim

active

claim:typicality-bias-is-present-across-all-annotator-demographic-groups-but-varies-in-magnitude-potentially-reflecting-differences-in-linguistic-norms-captured-by-pretraining-data

Typicality bias is present across all annotator demographic groups but varies in magnitude, potentially reflecting differences in linguistic norms captured by pretraining data

Finding from PRISM dataset analysis showing typicality bias varies by ethnicity and region, with implications for fairness

Source paper

extracted_from

Verbalized Sampling: How to Mitigate Mode Collapse and Unlock LLM Diversity

(2025) · Jiayi Zhang · Simon C.H. Yu · Derek Chong · Anthony Sicilia +3

Neighborhood — ranked by edge-count

Papers (1)

paper

Verbalized Sampling: How to Mitigate Mode Collapse and Unlock LLM Diversity
mentions

Related by similarity (8)

cosine ≥ 0.65 · no typed edge

Entities in the same semantic neighborhood but without a typed relation to this one — candidates for new edges or unrecognized duplicates.

Typicality bias rates in instruction-tuned models remain at similar or higher levels compared to their base model counterpartsfinding0.858
Shows typicality bias is preserved through instruction tuning and RLHF, not introduced by alignment
Typicality Biasconcept0.834
The human tendency to prefer more typical, familiar, fluent, and predictable text in annotation tasks, identified as a fundamental data-level cause of mode collapse
Typicality Bias Ratemethod0.831
Measurement of how often human annotators prefer the response with higher base model log-probability
Typicality bias rate exceeds 50% chance baseline by 4-12 percentage points across all base models on all four preference datasetsfinding0.831
Systematic evidence that base models implicitly prefer human-preferred responses, indicating preference biases emerge during pretraining
Typicality bias with weight α>0 mathematically sharpens the aligned policy via a power transform of the reference distributionclaim0.788
Theoretical result showing that any positive typicality bias weight γ-sharpens the reference distribution, amplifying modes
Typicality bias in preference data is a fundamental and pervasive data-level cause of mode collapse in aligned LLMsclaim0.785
The central thesis of the paper, distinguishing it from algorithmic explanations of mode collapse
Bias in language modelsconcept0.771
Features related to gender, racial, ethnic biases, slurs, and hate speech.
We hypothesize that native self-report, fine-tuned introspection models, and trained activation-to-language systems will show different performance on bias-resistant localization and strength benchmarkshypothesis0.758
Comparative prediction motivating future work contrasting different approaches to LLM self-knowledge