Relational Preference Encoding in Looped Transformer Internal States

• URL: http://arxiv.org/abs/2604.09870v1
• Date: Fri, 10 Apr 2026 20:00:49 GMT
• Title: Relational Preference Encoding in Looped Transformer Internal States
• Abstract summary: We investigate how looped transformers encode human preference in their internal iteration states.<n>Using Ouro-2.6B-Thinking, a 2.6B- parameter looped transformer with iterative refinement, we extract hidden states from each loop iteration.<n>We train lightweight evaluator heads to predict human preference on the Anthropic HH-RLHF dataset.<n>Our pairwise evaluator achieves 95.2% test accuracy on 8,552 unseen examples, surpassing a full-batch L-BFGS probe (84.5%) while the base model remains completely frozen.
• Score: 0.0
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: We investigate how looped transformers encode human preference in their internal iteration states. Using Ouro-2.6B-Thinking, a 2.6B-parameter looped transformer with iterative refinement, we extract hidden states from each loop iteration and train lightweight evaluator heads (~5M parameters) to predict human preference on the Anthropic HH-RLHF dataset. Our pairwise evaluator achieves 95.2% test accuracy on 8,552 unseen examples, surpassing a full-batch L-BFGS probe (84.5%) while the base model remains completely frozen. Our central finding is that loop states encode preference predominantly relationally: a linear probe on pairwise differences achieves 84.5%, the best nonlinear independent evaluator reaches only 65% test accuracy, and linear independent classification scores 21.75%, below chance and with inverted polarity. Interpreted precisely, the evaluator functions as a model-internal consistency probe, measuring how stably Ouro's own learned value system organizes its representations rather than how well it predicts noisy human annotations. We also document a systematic architecture search that established a genuine 70% ceiling for independent scoring, and show how the 50% argument-swap protocol required to prevent degenerate pairwise solutions deflated pairwise training metrics by about 31 points at peak, creating the false appearance that pairwise and pointwise evaluators shared the same ceiling. Finally, we show that a cosine learning-rate dead zone at epoch 2 accidentally acted as early stopping, preserving the generalization peak before overfitting degraded test accuracy from 95.2% to 62.4% by epoch 5. Cross-epoch flip-test analysis shows that antisymmetry correlation remains stable while strict sign-flip rate mainly tracks scorer bias. We propose the flip test as a mandatory diagnostic for pairwise preference evaluators.

Related papers

• Parameter-Efficient Quantum NLP for Paraphrase Detection: Performance, Robustness, and Entanglement [0.05999777817331314]
  We evaluate a 10-qubit hybrid quantum-classical variational circuit for paraphrase detection.<n>On QQP the circuit achieves 75.53% +/- 0.75% accuracy.<n>On MRPC the optimal 2-layer variant reaches 92% of BERT-base accuracy at 54,000 lower parameter cost.<n>These results constitute the first systematic parameter-matched multi-benchmark empirical validation of hybrid variational circuits for NLP.
  arXiv  Detail & Related papers  (2026-09-13T14:06:49Z)
• GradCuit: Credit-Assigned Gradient Flow Enables Robust and Interpretable Test-Time Latent Reasoning [61.67411833252235]
  We introduce GradCuit, which inserts optimizable latent states at a selected Transformer layer between the hidden representations of the prompt and the generated continuation.<n>Across five instruction-tuned backbones, three reasoning benchmarks, and two answer formats, GradCuit achieves an average accuracy of 64.5%.<n>For interpretability, token-level gradient attribution reveals that latent influence concentrates on reasoning-connector tokens, while layer analysis identifies early-to-middle Transformer layers as the most effective optimization space.
  arXiv  Detail & Related papers  (2026-08-03T17:55:24Z)
• Cross-Fitted Residual Utility for Primary-Preserving Cognitive Decision Correction in Automatic Modulation Classification [1.547549252134621]
  We study a post-inference problem through cross-fitted residual utility and a primary-preserving cognitive decision policy.<n>A structured KAN-Fourier classifier supplies the default probability, while neural and non-neural candidates provide observable evidence.<n>On RMLA, RMLB, and HISAR, the complete system improves overall accuracy from 63.632% to 66.332%, 65.161% to 66.168%, and 77.769% to 79.867%.
  arXiv  Detail & Related papers  (2026-08-03T11:07:37Z)
• Operational Proto-Introspection in Looped Language Models: Process-Quality Taps, Executable Branching, and the Readout-Control Boundary [0.0]
  Can a language model read the quality of its ongoing computation?<n>Can an external intervention turn that readout into better outcomes?<n>We test both questions in a frozen 2.6B looped transformer, Ouro-RLTT.
  arXiv  Detail & Related papers  (2026-07-20T22:40:36Z)
• Efficient Visual Pointing for Embodied AI:Agent-Driven Data Synthesis, Cross-Block Attention, and Iterative Correction [55.11480729304395]
  We describe our PointArena 2026 solution, which achieves 77.2% overall accuracy and ranks second on the benchmark.<n>The ap proach targets three failure modes. First, agent-driven syn thesis builds large semantic and anchor-relative candidate pools.<n>Second, a determinis tic steerable-data pipeline creates a verified 10,000-sample main set, plus reserve samples, using masks, templates, and path verification.
  arXiv  Detail & Related papers  (2026-06-29T06:39:03Z)
• Automated Proving of Shannon-Type Entropy Inequalities via Fine-Tuned Language Models and Guided Tree Search [50.16356451328644]
  Proving Shannon-type entropy inequalities is a fundamental task in information theory.<n>We investigate whether small-scale large language models, fine-tuned on atomic proof steps, can automate this process.<n>We show that GPT-5.5 solves 1.7% samples under zero-shot prompting while Psitip solves 33.3% samples.
  arXiv  Detail & Related papers  (2026-06-04T05:43:12Z)
• Let the Results Speak: A Replication-First Paradigm for LLM Behavioral Benchmarking [22.825786049667602]
  We propose a replication-first paradigm to anchor validity to a single human-rater consensus.<n>We certify the instrument via four properties -- reliability across K runs, cross-instrument replication across architecturally distinct judges, historical-footprint calibration via judges from earlier training cohorts, and pre-registered prediction.<n>We test it on emotional accompaniment by letting the self-evolve data-driven across: the dimensions are not pre-stipulated and the procedure stabilizes to a 9-dimension set.
  arXiv  Detail & Related papers  (2026-05-27T03:41:11Z)
• CriterAlign: Criterion-Centric Rationale Alignment for Code Preference Judging [95.02210956333374]
  We propose a criterion-centric framework that adapts rubric-based judging to pairwise preference evaluation.<n>On BigCodeReward, CriterAlign improves a Qwen2.5-VL-32B monolithic judge from 60.4% to 66.3% accuracy.
  arXiv  Detail & Related papers  (2026-05-19T10:59:19Z)
• Putting HUMANS first: Efficient LAM Evaluation with Human Preference Alignment [53.72927532626824]
  We show that subsets of just 50 examples (0.3% of data) can achieve over 0.93 Pearson correlation with full benchmark scores.<n>To better predict preferences, we trained regression models on these selected subsets, achieving 0.98 correlation.<n>This demonstrates that in regression modeling, well-curated subsets outpredict the full benchmark, showing quality over quantity.
  arXiv  Detail & Related papers  (2026-04-20T00:57:31Z)
• Aligning to Illusions: Choice Blindness in Human and AI Feedback [2.912535004465788]
  In a human choice blindness study, 91% of surreptitiously swapped preferences go undetected, extending choice blindness to third-person evaluative comparison of unfamiliar text.<n>We find detection relies on shallow text matching rather than genuine self-monitoring.<n>In a dose-response experiment across two architectures, one-sixth to one-third of labels must be corrupted before the reward signal halves.<n>A Best-of-N evaluation confirms this translates to downstream policy degradation.
  arXiv  Detail & Related papers  (2026-03-09T14:10:36Z)
• Variational Inference for Bayesian MIDAS Regression [0.0]
  We develop a Coordinate Ascent Variational Inference (CAVI) algorithm for regression with linear weight parameteri zations.<n>CAVI produces posterior means nearly identical to a block Gibbs sampler benchmark while achieving speedups of 107x to 1,772x.<n>Weight function parameters maintain excellent calibration (coverage above 92%) across all con gurations.
  arXiv  Detail & Related papers  (2026-02-23T08:51:26Z)
• CoRefine: Confidence-Guided Self-Refinement for Adaptive Test-Time Compute [10.548368675645403]
  CoRefine is a confidence-guided self-refinement method that achieves competitive accuracy using a fraction of the tokens.<n>The controller consumes full-trace confidence to decide whether to halt, re-examine, or try a different approach.<n>We extend this to CoRefine-Tree, a hybrid sequential-parallel variant that adaptively balances exploration and exploitation.
  arXiv  Detail & Related papers  (2026-02-09T17:44:41Z)
• iResolveX: Multi-Layered Indirect Call Resolution via Static Reasoning and Learning-Augmented Refinement [9.184762707302099]
  Indirect call resolution is a key challenge in reverse engineering and control-flow graph recovery.<n>We present iResolveX, a framework that combines conservative static analysis with learning-based refinement.<n>iResolveX supports both conservative, recall-preserving and F1-optimized configurations and outperforms state-of-the-art systems.
  arXiv  Detail & Related papers  (2026-01-25T15:42:34Z)
• Ensemble Threshold Calibration for Stable Sensitivity Control [0.0]
  We present an end-to-end framework that achieves exact recall with sub-percent variance over tens of millions of geometry pairs.<n>Our approach consistently hits a recall target within a small error, decreases redundant verifications relative to other calibrations, and runs end-to-end on a single TPU v3 core.
  arXiv  Detail & Related papers  (2025-10-02T15:22:28Z)
• CLUE: Non-parametric Verification from Experience via Hidden-State Clustering [64.50919789875233]
  We show that correctness of a solution is encoded as a geometrically separable signature within the trajectory of hidden activations.<n>ClUE consistently outperforms LLM-as-a-judge baselines and matches or exceeds modern confidence-based methods in reranking candidates.
  arXiv  Detail & Related papers  (2025-10-02T02:14:33Z)
• Reference-Free Rating of LLM Responses via Latent Information [53.463883683503106]
  We study the common practice of asking a judge model to assign Likert-scale scores to free-text responses.<n>We then propose and evaluate Latent Judges, which derive scalar ratings from internal model signals.<n>Across a broad suite of pairwise and single-rating benchmarks, latent methods match or surpass standard prompting.
  arXiv  Detail & Related papers  (2025-09-29T12:15:52Z)
• Time Is a Feature: Exploiting Temporal Dynamics in Diffusion Language Models [57.474294329887236]
  Diffusion large language models (dLLMs) generate text through iterative denoising.<n>Current decoding strategies discard rich intermediate predictions in favor of the final output.<n>We introduce two complementary methods that exploit temporal consistency.
  arXiv  Detail & Related papers  (2025-08-12T17:59:57Z)
• Mitigating LLM Hallucinations via Conformal Abstention [70.83870602967625]
  We develop a principled procedure for determining when a large language model should abstain from responding in a general domain. We leverage conformal prediction techniques to develop an abstention procedure that benefits from rigorous theoretical guarantees on the hallucination rate (error rate) Experimentally, our resulting conformal abstention method reliably bounds the hallucination rate on various closed-book, open-domain generative question answering datasets.
  arXiv  Detail & Related papers  (2024-04-04T11:32:03Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.