Pith. sign in

REVIEW 10 cited by

Eliminating Position Bias of Language Models: A Mechanistic Approach

Not yet reviewed by Pith; the record is open.

This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.

SPECIMEN: schema-true, not a live event

T0 review · schema-true

One-sentence machine reading of the paper's core claim.

pith:XXXXXXXX · record.json · timestamp

arxiv 2407.01100 v3 pith:T7UFRGIR submitted 2024-07-01 cs.CL cs.LG

Eliminating Position Bias of Language Models: A Mechanistic Approach

classification cs.CL cs.LG
keywords biaspositionattentionmodelsperformancedocumentsreasoningapproach
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved
0 comments
Share X Bluesky LinkedIn Reddit HN
read the original abstract

Position bias has proven to be a prevalent issue of modern language models (LMs), where the models prioritize content based on its position within the given context. This bias often leads to unexpected model failures and hurts performance, robustness, and reliability across various applications. Our mechanistic analysis attributes the position bias to two components employed in nearly all state-of-the-art LMs: causal attention and relative positional encodings. Based on the analyses, we propose to eliminate position bias (e.g., different retrieved documents' orders in QA affect performance) with a training-free zero-shot approach. Our method changes the causal attention to bidirectional attention between documents and utilizes model attention values to decide the relative orders of documents instead of using the order provided in input prompts, therefore enabling Position-INvariant inferencE (PINE) at the document level. By eliminating position bias, models achieve better performance and reliability in downstream tasks, including LM-as-a-judge, retrieval-augmented QA, molecule generation, and math reasoning. Notably, PINE is especially useful when adapting LMs for evaluating reasoning pairs: it consistently provides 8 to 10 percentage points performance gains, making Llama-3-70B-Instruct perform even better than GPT-4-0125-preview and GPT-4o-2024-08-06 on the RewardBench reasoning set.

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Forward citations

Cited by 10 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. Is Position Bias in Dense Retrievers Built In-or Learned from Data?

    cs.IR 2026-05 unverdicted novelty 7.0

    Training data position distribution controls positional bias in dense retrievers, with balanced training reducing sensitivity by 57-87%.

  2. Towards Order Fairness: Mitigating LLMs Order Sensitivity through Dual Group Advantage Optimization

    cs.LG 2026-05 unverdicted novelty 7.0

    DGAO uses reinforcement learning to optimize LLMs for both accuracy and order stability by balancing intra-group accuracy advantages and inter-group stability advantages.

  3. Inside the Unfair Judge: A Mechanistic Interpretability Account of LLM-as-Judge Bias

    cs.LG 2026-07 conditional novelty 6.5

    LLM-as-judge scoring biases concentrate in low-dimensional, type-specific activation subspaces that support bidirectional causal steering and cross-domain failure prediction.

  4. MARS-RA: Rank Aggregation for Credit Assignment via Multimodal Comparisons in Embodied Multi-Agent Cooperation

    cs.AI 2026-07 conditional novelty 6.0

    Credit assignment via LMM pairwise comparisons plus Bradley–Terry rank aggregation and potential-based shaping improves cooperative MARL under sparse rewards and dynamic agent counts.

  5. Logit-Attention Divergence: Mitigating Position Bias in Multi-Image Retrieval via Attention-Guided Calibration

    cs.CV 2026-05 unverdicted novelty 6.0

    A training-free attention-guided debiasing framework mitigates position bias in MLLM multi-image retrieval by exploiting the observed mismatch between biased logits and aligned attention maps, yielding over 40% accura...

  6. Mitigating Coordinate Prediction Bias from Positional Encoding Failures

    cs.CV 2025-10 unverdicted novelty 6.0

    VPSG corrects predictable directional coordinate biases in MLLMs by shuffling visual positional encodings to isolate unconditioned tendencies and steering digit decoding with a lightweight finite-state machine, yieldi...

  7. LISA: Linear-Indexed Sparse Attention for Efficient Long-Context Reasoning

    cs.AI 2026-05 reject novelty 5.0

    LISA replaces standard attention with linear attention plus indexer-selected sparse attention, claiming O(nM) inference cost and accuracy gains, but the complexity analysis ignores the indexer's full-context scoring.

  8. LLM-Oriented Information Retrieval: A Denoising-First Perspective

    cs.IR 2026-05 unverdicted novelty 5.0

    Denoising to maximize usable evidence density and verifiability is becoming the primary bottleneck in LLM-oriented information retrieval, conceptualized via a four-stage framework and addressed through a pipeline taxo...

  9. The PICCO Framework for Large Language Model Prompting: A Taxonomy and Reference Architecture for Prompt Structure

    cs.CL 2026-04 accept novelty 5.0

    PICCO is a five-element reference architecture (Persona, Instructions, Context, Constraints, Output) for structuring LLM prompts, derived from synthesizing prior frameworks along with a taxonomy distinguishing prompt ...

  10. LLM-Oriented Information Retrieval: A Denoising-First Perspective

    cs.IR 2026-05 unverdicted novelty 4.0

    Argues for a denoising-first paradigm in LLM-oriented information retrieval, framing challenges via a four-stage progression and providing a taxonomy of signal-to-noise optimization techniques across the pipeline.