concept
active
concept:primacy-and-recency-biasesPrimacy and recency biases
Human and LM memory biases paralleled by the U-shape.
Neighborhood — ranked by edge-count
Papers (1)
paper
Concepts (1)
concept
- Lost-in-the-middle effectassociated_withLM weakness on middle context positions, echoed here.
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.
- 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
- Assumptions or preferences (e.g., parsimony) that determine how a learning system generalizes beyond training data
- Deep networks are biased toward finding simple fits to data, and this bias increases with model size, driving convergence
- The tendency of deep networks to implicitly favor simpler solutions that fit the data, driving convergence
- Finding from PRISM dataset analysis showing typicality bias varies by ethnicity and region, with implications for fairness
- Measurement of how often human annotators prefer the response with higher base model log-probability
- Systematic evidence that base models implicitly prefer human-preferred responses, indicating preference biases emerge during pretraining
- Overall conclusion.