Pith. sign in

REVIEW 2 major objections 6 minor 41 references

LODESTAR: Trustworthy Entropy Is Navigated, Not Merely Measured -- Reinforced Polarizer Keeps a Frozen LLM from Being Confidently Misled by the Wrong Evidence

T0 review · 2 major / 6 minor · reviewed 2026-08-16 · deepseek-v4-flash

Pith's one-line read A single learned string inserted into the prompt lets a frozen LLM avoid confidently wrong answers in retrieval-augmented QA.

desk verdict A genuinely new and unusually honest method for steering a frozen LLM's selection uncertainty; the headline gain is real, but the estimator choice on the eval pools makes the OOD claims softer than stated. read the letter →

arxiv 2608.11922 v1 pith:5RMACBHE submitted 2026-08-12 cs.CL cs.IRcs.LG

classification cs.CLcs.IRcs.LG
keywords retrieval-augmentedgenerationentropy-basedselectionconfidentlywrongpolarizerfrozenLLMreinforcementlearningquestionansweringpromptintervention
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

Retrieval-augmented question answering suffers from a quietly dangerous failure: a misleading retrieved passage can make a frozen answering model more confident, not less, so the standard “keep the lowest-entropy answer” rule selects exactly the wrong evidence. The paper argues this is not a flaw of entropy itself but of treating the model’s uncertainty as a fixed measurement. It introduces LODESTAR, which learns one short natural-language string offline by reinforcement learning and inserts it between the passage and the question; this “polarizer” steers the responder’s first-token entropy upward on misleading passages and leaves supporting passages near the entropy floor. On 5,008 questions across five QA benchmarks, the same minimum-entropy rule with one string inserted lifts mean answer $F_1$ from 0.5148 to 0.5339, exact match to 0.4136, and lowers the rate of selecting a judged-misleading passage from 30.3% to 26.0%. The paper’s central claim is that entropy should be navigated—changed as an input—rather than merely measured.

What carries the argument

The central object is the polarizer $\psi^\star$, a short fixed natural-language string trained once offline by reinforcement learning and inserted between the retrieved passage and the question, never into the model’s weights. It works by “directed entropy”: the reward is the within-question separation $\Delta_{Mis-Sup}\bar H_L(q;\psi)-\Delta_{Mis-Sup}\bar H_L(q)$ between the responder’s mean first-token entropy on judged-misleading and supporting passages, so the string is optimized to raise entropy on misleading candidates while leaving supporting ones near the floor. At inference, selection is still $\arg\min_a \bar H_L(q,p,\psi^\star)$, with no extra model, no sampling, and no gold answer; the only artifact carried from training is the string itself.

What would settle it

Take the published polarizer and apply it, with the same frozen respondent and candidate pools, to a new benchmark from an unseen domain; if the judged-misleading selection rate does not fall below the pool’s base rate and answer $F_1$ does not rise, the transfer assumption is false.

Watch

Extended reading notes

Core claim

LODESTAR establishes that within-question entropy can be repaired by a text intervention. When a misleading passage is read alone, the frozen respondent’s answer-token entropy is lower than on a supporting passage—1.22 nats versus 1.33 pooled across the five pools—so argmin-entropy selection is drawn to the wrong answer. The paper trains a polarizer $\psi^\star$ with a policy-gradient objective whose reward is the baseline-corrected within-question separation of the responder’s first-token entropy between judged-misleading and supporting passages. Inserted as $[p; \psi^\star; q]$, the learned string raises entropy on the misleading side by roughly an order of magnitude in the worked example, moving the argmin to the supporting passage. Across 5,008 questions and fourteen published selectors re-run under the same frozen respondent and candidate pools, this one fixed string gives the highest mean answer $F_1$ (0.5339), exact match (0.4136), and independent-judge accuracy (0.6435), with the string itself, by ablation, contributing $+0.0191 \pm 0.0054$ $F_1$ across three training seeds; the gain holds both on Natural Questions and on the four out-of-domain benchmarks.

Load-bearing premise

The entire transfer claim rests on the assumption that a short English string trained only on Natural Questions can shift a frozen model’s first-token confidence on misleading passages in the same way on other datasets; if the steering effect is specific to NQ-like passages, the out-of-domain gains disappear.

Editorial extensions

If this is right

  • Because inference is one forward pass per candidate and reads no gold answer, the polarizer can be dropped into existing retrieval-augmented pipelines at negligible cost.
  • The ablation shows the string, not the training procedure, accounts for the gain: removing it returns $F_1$ and misleading-rate to the plain entropy baseline, positive on 5/5 datasets under all three seeds.
  • The polarizer is respondent-specific: on two additional frozen respondents, every diagonal pairing (train and evaluate on the same model) improves, while several off-diagonal transfers hurt, so the string should be trained for the model that will answer.
  • The learned string outperforms both hand-written cautionary instructions and prompt-search optimizers under the same reward, indicating the RL training finds a better optimum than direct search.
  • Entropy selection alone reads a misleading passage more often than random draw (30.3% vs. the pool’s 28.9%); LODESTAR is the configuration that consistently stays below that floor.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • If the mechanism is robust, prompt-level steering of first-token entropy could be a general “epistemic navigation” tool for frozen models—e.g., to make a model less credulous about retrieved evidence in other tasks like fact-checking or medical QA—without fine-tuning.
  • The polarizer’s converged text is an interpretable instruction to doubt that the passage addresses the exact question; independent training seeds converge on similar wording, suggesting the reward has a stable linguistic optimum that could be used to audit what reinforcement learning actually optimizes.
  • The reward’s training labels come from two LLM judges with Cohen’s kappa 0.675; improving label agreement, or replacing gold-exact-match definitions of support, would likely sharpen the entropy separation and could yield larger gains than the current +3.7% $F_1$.
  • A natural stress test would be to train the polarizer on one respondent and evaluate on an unseen model family; the paper’s cross-respondent matrix already shows negative off-diagonal transfers, so a pretrained “universal polarizer” is not implied.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

2 major / 6 minor

Summary. The paper proposes LODESTAR, an offline-trained natural-language 'polarizer' string inserted between a retrieved passage and the question in the prompt of a frozen LLM respondent. The claim is that a single fixed string, trained by GRPO to maximize the within-question separation of the respondent's first-token entropy between judged-misleading and gold-matching passages, is sufficient to make lowest-entropy answer selection less susceptible to confidently wrong answers. Across 5,008 questions from five QA benchmarks (NQ-Open plus four OOD sets), with a frozen Llama-3.1-8B respondent and shared bge-m3 top-10 pools, the three-seed mean of LODESTAR reaches 0.5339 mean answer F1 versus 0.5148 for plain first-token entropy selection, 0.4769 for the retriever's top-1, and is reported as paired-significant against fourteen re-purposed published configurations; the polarizer ablation reduces the rate of selecting judged-misleading passages from 30.3% to 26.0%. The manuscript also reports cross-respondent transfer (positive diagonals, four negative off-diagonal facets) and a detailed reproduction of CTRL-RAG in an appendix.

Significance. If the results hold, LODESTAR would be a practically valuable, cheap intervention — no extra model, sampling, or gold answers at inference — and a clean demonstration that a frozen respondent's uncertainty signal can be steered by prompt-level text rather than merely measured. The paper's strengths are its unusually complete empirical discipline: all three training seeds are reported individually, with a same-seed execution-level rerun (Appendix E); per-question paired tests (bootstrap, t, Wilcoxon) are tabulated in full (Tables 7, 9, 10); every baseline deviation from official scripts is disclosed in Appendix J, including the semantic-entropy first-wins tie-break inflation and the SPS task mismatch; and the CTRL-RAG case study derives a mechanism (Eq. 7) for the observed format degeneration. The controlled ablation of Section 6 (removing only the polarizer string) is well designed, and the hand-written vs searched vs GRPO-learned ordering (0.5129 / 0.5131–0.5160 / 0.5339) is a strong comparison.

major comments (2)
  1. [Section 4.2 and Appendix C (Table 6)] The choice of the L=1 first-token entropy estimator over the aligned all-token mean was made by comparing selection F1 on the very same five evaluation pools that later produce every headline number in Tables 2 and 3 (0.5148 vs 0.5001 on those pools). The four 'out-of-domain' datasets participated in this choice, so no tested pool is untouched by estimator selection, and the headline claims of 'highest mean F1 of any inference-ready selector' and of out-of-domain transfer (0.5274 to 0.5476) are more fragile than the paper's wording suggests: the untrained entropy signal LODESTAR builds on was selected on the test pools, while the fourteen baselines were not given test-pool-based tuning of their own scoring rules. The Table 3 ablation survives this critique because it keeps H1 fixed, and Figure 2's positive Llama diagonal under the all-token selector suggests the polarizer effect is not H1-specific, but the main tables never connect these dots. I ask the authors to either (a) re-select the estimator on NQ-Open training questions only, or (b) report the key tables under both H1 and the aligned all-token estimator (retraining the polarizer for the latter if needed), or (c) state the limitation in the main text and qualify the 'highest F1' and OOD claims accordingly.
  2. [Sections 3 and 6, Eq. (4), Table 3, Appendix G] The two headline outcomes are not equally independent of the training labels. The reward in Eq. (4) contrasts passages labeled MISLEADING by a dual-judge consensus (Cohen's kappa = 0.675) against passages labeled SUPPORTING by exact match with gold, and Table 3's judged-misleading rate is then computed with the same Section 3 judge rule extended across all pools. While the paper is transparent about this and about the kappa value, the title-level claim that the polarizer keeps the respondent from being confidently misled by the wrong evidence rests partly on a metric that shares its label source with the training reward, and the independent audit in Appendix G is computed only on the 600 consensus rows, leaving disagreement rows unaudited. The F1 and EM results are label-independent and are the strongest evidence for the paper; I ask the authors to additionally report the misleading-rate ablation using the independent 70B audit judge over the full audit set (including disagreement rows), or to state in the main text that the misleading-rate metric evaluates the training objective in-family rather than serving as an independent audit.
minor comments (6)
  1. [Section 1 and abstract] The phrasing 'one fixed string, enough by itself to outperform every published method the paper tests' should carry the frozen-respondent, inference-ready qualifier consistently, since the paper's own Table 12 reports CTRL-RAG's best checkpoint at 0.5920 mean F1 against LODESTAR's 0.5339 on identical pools, with the disambiguation residing only in an appendix.
  2. [Table 2 (also Tables 9 and 10)] The Reference column entry for the LODESTAR row contains an illegible glyph sequence that appears to be a corrupted superscript marker; it should be removed or replaced with a proper reference or star notation.
  3. [Section 4.1 and Table 1] All main results use the bge-m3 top-10 pool, whose misleading share (28.9% macro) is the lowest band in Table 1; since the paper motivates the problem by showing that rerankers raise the misleading rate to as much as 35.0%, evaluating LODESTAR on at least one reranked pool would test the mechanism where the problem is most severe.
  4. [Section 3 and Appendix G] The third-party audit (96% agreement) is computed on the 600 consensus rows, and this scope is stated only in the appendix; the main-text sentence in Section 3 should say explicitly that the audit covers only rows on which the two judges already agree, leaving the disagreement cases unaudited.
  5. [Section 6 and Figure 2] Section 6 could cite the Llama diagonal of Figure 2 as evidence that the polarizer's benefit is not specific to the H1 estimator, since those facets use a mean-normalized all-token selector; this would directly address the estimator-dependence concern raised by Appendix C.
  6. [General] The manuscript describes LODESTAR as a new method but provides no link to released code or artifacts; given the reliance on exact prompt templates, judge protocols, and cached probes, a statement of where the training and evaluation code will be made available is needed for reproducibility.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the polarizer is trained on an entropy-separation reward built from disclosed gold and judge labels, while the headline F1 numbers are external measurements on held-out pools and are not defined by the training objective.

full rationale

The claimed derivation is not circular. LODESTAR's objective (Eq. 4) rewards the within-question separation of the respondent's entropy between Mis(q) and Sup(q), Δ_Mis−Sup H̄L(q;ψ) − Δ_Mis−Sup H̄L(q); selection (Eq. 3) is argmin over H1(q,p,ψ*). The paper's headline F1, exact-match, and judge scores are measured against gold answers and are not quantities that Eq. 4 optimizes, so the result is not equivalent to its training signal by construction. The labels are gold- and judge-derived, but the paper explicitly discloses this: "gold-freedom is a property of inference, not of training-data construction"; using gold-derived labels for training is standard supervision, not circularity. No load-bearing claim rests on a self-citation: the reference list contains no overlapping-author citation, and no uniqueness theorem is imported. The Appendix C choice of H1 over the aligned all-token mean is made on the same five evaluation pools used for all headline numbers, which is a real test-set reuse and estimator-selection caveat for the out-of-domain and "highest mean F1" claims; however, it is not a circular reduction because the polarizer itself is trained only on the NQ-Open training pool and the ablation (Section 6) holds H1 fixed for both configurations, so the +0.0191 polarizer effect and the misleading-read-rate reduction remain meaningful comparisons. The correct criticism of Appendix C is test-set reuse, which belongs under correctness risk rather than circularity; the central derivation is otherwise self-contained.

Assumptions & free parameters 4 free parameters · 4 assumptions · 0 invented entities

The paper introduces no new physical or mathematical entities. The polarizer is a learned string artifact, and 'directed entropy' is a defined quantity, not a postulated entity. The free parameters listed are modeling choices that shape the training signal and the final numbers.

free parameters (4)
  • First-token entropy estimator (L=1) = L=1
    Chosen after comparing H1 with the aligned all-token mean on the same five evaluation pools (Appendix C), a choice made on the test benchmarks rather than on a separate holdout.
  • Entropy-shift clip = ±2
    Per-passage entropy shifts are clipped to ±2 in the reward (Appendix B); a hand-set constant that shapes the training signal.
  • SUPPORTING label criterion = exact match after normalization
    Passages are labeled SUPPORTING only when the respondent's answer exactly matches a gold answer (Section 3), which couples the training labels to answer correctness and to the normalization scheme.
  • Judge agreement rule = both judges must agree MISLEADING
    A passage is MISLEADING only when two LLM judges both agree (Section 3); changing this rule changes the reward and therefore the learned polarizer.
assumptions (4)
  • domain assumption Two LLM judges' labels are reliable proxies for passage misleadingness.
    The reward (Eq. 4) and ablation metrics rely on Mis/Sup labels produced by gpt-oss-120b and Qwen2.5-72B (Section 3); inter-judge agreement is only Cohen's kappa 0.675.
  • domain assumption First-token entropy is a sufficient selection signal.
    The method and baseline use H1 throughout (Section 4.2); the paper cites Clark et al. (2025) cautioning that H1 may not approximate word entropy, and validates it only on the five evaluation pools (Appendix C).
  • domain assumption A fixed natural-language string can durably steer a frozen LLM's uncertainty.
    The polarizer is one short string that must raise entropy on misleading passages and hold supporting passages low across domains; only empirical evidence (Section 6) supports this.
  • domain assumption RL training on NQ-Open transfers out-of-domain.
    The polarizer is trained only on 1,939 NQ-Open training questions (Section 4.2) and evaluated on four OOD datasets; Figure 2 tests this but with negative off-diagonal transfers.

how reviews work

0 comments
Cite this review

Pith. "Pith review of LODESTAR: Trustworthy Entropy Is Navigated, Not Merely Measured -- Reinforced Polarizer Keeps a Frozen LLM from Being Confidently Misled by the Wrong Evidence." pith.science (2026). https://pith.science/paper/5RMACBHE

@misc{pith2026260811922,
  author       = {Pith},
  title        = {Pith review of: LODESTAR: Trustworthy Entropy Is Navigated, Not Merely Measured -- Reinforced Polarizer Keeps a Frozen LLM from Being Confidently Misled by the Wrong Evidence},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/5RMACBHE}},
  note         = {Machine review of arXiv:2608.11922}
}
abstract

Predictive-distribution entropy makes a strong selection rule in retrieval-augmented question answering: across five QA benchmarks, keeping the candidate answer that a frozen respondent LLM produces with the lowest answer-token entropy lifts mean answer $F_1$ from 0.4769 to 0.5148 over the retriever's top-ranked passage, with no gold answers. Yet this lowest-entropy rule, which prior entropy-based selectors adopt, fails in a specific and consequential way: a misleading passage makes the respondent confidently wrong, driving its entropy down precisely where the signal looks most trustworthy. We show that the failure comes from the passage the respondent reads -- and the context that passage is read in is an input we can intervene on. We introduce LODESTAR, to our knowledge the first method to score a text intervention by the uncertainty it induces in a third-party frozen respondent, compared across one question's candidates. LODESTAR uses reinforcement learning to train, once and offline, a polarizer -- a short fixed natural-language string inserted into the respondent's prompt and never into its weights; its training labels are built offline from gold answers and two LLM judges, and inference reads neither. Evaluating every competing selector under the same frozen respondent and the same candidate pools on 5,008 questions, LODESTAR attains the highest mean $F_1$ of any inference-ready selector (0.5148 to 0.5339), the highest exact match (0.4136), and the highest GPT-4o judge score of the frozen-respondent configurations judged (0.6435); its three-seed mean wins all 70 method-by-dataset $F_1$ cells against fourteen published configurations while remaining paired-significant against every one. The gain holds both in-domain and out-of-domain, and ablating the polarizer shows it is what makes the respondent read a misleading passage less often (26.0% against 30.3%).

Figures

Figures reproduced from arXiv: 2608.11922 by the authors.

Figure 1
Figure 1. Why minimum-entropy selection is misled, and what LODESTAR changes. (a) The misleading passage makes the frozen respondent become confidently wrong: the respondent’s en￾tropy is lowest on that passage (0.0013, the minimum over all ten, against the supporting pas￾sage’s 0.0021), and the lowest-entropy rule keeps the answer that passage induces, which is wrong. (b) LODESTAR inserts one learned polarizer ψ ⋆ between ea… view at source ↗
Figure 2
Figure 2. Cross-respondent transfer of the polarizer under a mean-normalized-entropy (H¯L) selector. Rows are the respondent the polarizer ψ was trained against, columns the frozen respondent that selects and answers; each of the eighteen facets prints answer F1 with ψ inserted (top) and ∆ against the same-pass configuration without it (bottom). Every diagonal is positive on both panels; all four negative facets are off-diago… view at source ↗
Figure 3
Figure 3. CTRL-RAG reproduction, the RL stage measured two ways. Both panels are the SFT-anchored seed-42 run whose checkpoints [PITH_FULL_IMAGE:figures/full_fig_p023_3.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

41 extracted references · 23 canonical work pages

  1. [1]

    LLM-as-judge protocol.Because bothF 1 and EM are lexical, we additionally audit answer cor- rectness with an LLM judge, so that no conclusion rests on string overlap alone

    The polarizer ablation, the cleanest test of the contribution, remains significant under EM on both tests. LLM-as-judge protocol.Because bothF 1 and EM are lexical, we additionally audit answer cor- rectness with an LLM judge, so that no conclusion rests on string overlap alone. To keep the audit independent of our own design choices, the judging prompt i...

  2. [5]

    Revanth Gangi Reddy, JaeHyeok Doo, Yifei Xu, Md Arafat Sultan, Deevya Swain, Avirup Sil, and Heng Ji

    doi: 10.1038/s41586-024 -07421-0. Revanth Gangi Reddy, JaeHyeok Doo, Yifei Xu, Md Arafat Sultan, Deevya Swain, Avirup Sil, and Heng Ji. FIRST: Faster improved listwise reranking with single token decoding. InProceedings of the 2024 Conference on Empirical Methods in Natural Language Processing (EMNLP), pp. 8642–8652,

  3. [6]

    Aaron Grattafiori, Abhimanyu Dubey, Abhinav Jauhri, Abhinav Pandey, Abhishek Kadian, Ahmad Al-Dahle, Aiesha Letman, Akhil Mathur, Alan Schelten, Alex Vaughan, et al

    doi: 10.18653/v1/2024.emnlp-main.491. Aaron Grattafiori, Abhimanyu Dubey, Abhinav Jauhri, Abhinav Pandey, Abhishek Kadian, Ahmad Al-Dahle, Aiesha Letman, Akhil Mathur, Alan Schelten, Alex Vaughan, et al. The Llama 3 herd of models.arXiv preprint arXiv:2407.21783,

  4. [8]

    arXiv:2508.05909

    doi: 10.1609/aaai.v40i37.40371. arXiv:2508.05909. Gautier Izacard, Mathilde Caron, Lucas Hosseini, Sebastian Riedel, Piotr Bojanowski, Armand Joulin, and Edouard Grave. Unsupervised dense information retrieval with contrastive learning. Transactions on Machine Learning Research,

  5. [9]

    Yi Jiang, Sendong Zhao, Jianbo Li, Haochun Wang, and Bing Qin

    arXiv:2112.09118. Yi Jiang, Sendong Zhao, Jianbo Li, Haochun Wang, and Bing Qin. GainRAG: Preference align- ment in retrieval-augmented generation through gain signal synthesis. InAnnual Meeting of the Association for Computational Linguistics (ACL),

  6. [10]

    GainRAG: Preference Alignment in Retrieval-Augmented Generation through Gain Signal Synthesis

    arXiv:2505.18710. Mandar Joshi, Eunsol Choi, Daniel S. Weld, and Luke Zettlemoyer. TriviaQA: A large scale dis- tantly supervised challenge dataset for reading comprehension. InProceedings of the 55th Annual Meeting of the Association for Computational Linguistics (ACL), pp. 1601–1611,

  7. [11]

    Huan Ma, Jiadong Pan, Jing Liu, Yan Chen, Joey Tianyi Zhou, Guangyu Wang, Qinghua Hu, Hua Wu, Changqing Zhang, and Haifeng Wang

    arXiv:2503.20783. Huan Ma, Jiadong Pan, Jing Liu, Yan Chen, Joey Tianyi Zhou, Guangyu Wang, Qinghua Hu, Hua Wu, Changqing Zhang, and Haifeng Wang. Semantic energy: Detecting LLM hallucination be- yond entropy.arXiv preprint arXiv:2508.14496,

  8. [12]

    Charles Moslonka, Hicham Randrianarivo, Arthur Garnier, and Emmanuel Malherbe

    ACL Anthology 2025.acl-long.319. Charles Moslonka, Hicham Randrianarivo, Arthur Garnier, and Emmanuel Malherbe. Learned hal- lucination detection in black-box LLMs using token-level entropy production rate. InEuropean Conference on Information Retrieval (ECIR), volume 16483 ofLecture Notes in Computer Sci- ence, pp. 115–130. Springer,

Show all 41 references
  1. [13]

    doi: 10.1007/978-3-032-21289-4

  2. [14]

    arXiv:2509.04492. OpenAI. GPT-4o system card.arXiv preprint arXiv:2410.21276,

  3. [15]

    gpt-oss-120b & gpt-oss-20b Model Card.arXiv preprint arXiv:2508.10925,

    OpenAI. gpt-oss-120b & gpt-oss-20b Model Card.arXiv preprint arXiv:2508.10925,

  4. [16]

    Qwen Team

    ACL Anthology 2025.naacl-long.236. Qwen Team. Qwen3.5: Towards Native Multimodal Agents, 2026.https://qwen.ai/blog ?id=qwen3.5. Pranav Rajpurkar, Jian Zhang, Konstantin Lopyrev, and Percy Liang. SQuAD: 100,000+ questions for machine comprehension of text. InProceedings of the ...

  5. [19]

    Zhipeng Song, Yizhi Zhou, Xiangyu Kong, Jiulong Jiao, Xinrui Bao, Xu You, Xueqing Shi, Yuhang Zhou, and Heng Qi

    arXiv:2310.16789. Zhipeng Song, Yizhi Zhou, Xiangyu Kong, Jiulong Jiao, Xinrui Bao, Xu You, Xueqing Shi, Yuhang Zhou, and Heng Qi. Less is more for RAG: Information gain pruning for generator-aligned reranking and evidence selection.arXiv preprint arXiv:2601.17532,

  6. [20]

    11 Zhehao Tan, Yihan Jiao, Dan Yang, Junjie Wang, Duolin Sun, Jie Feng, Xidong Wang, Lei Liu, Yue Shen, Jian Wang, and Jinjie Gu

    ACL Anthology 2025.findings-acl.852. 11 Zhehao Tan, Yihan Jiao, Dan Yang, Junjie Wang, Duolin Sun, Jie Feng, Xidong Wang, Lei Liu, Yue Shen, Jian Wang, and Jinjie Gu. CTRL-RAG: Contrastive likelihood reward based reinforcement learning for context-faithful RAG models.arXiv pre...

  7. [21]

    Aditya Taparia, Ransalu Senanayake, Kowshik Thopalli, and Vivek Narayanaswamy

    arXiv:2412.01572. Aditya Taparia, Ransalu Senanayake, Kowshik Thopalli, and Vivek Narayanaswamy. The anatomy of uncertainty in LLMs.arXiv preprint arXiv:2603.24967,

  8. [22]

    Large language models can accu- rately predict searcher preferences

    Paul Thomas, Seth Spielman, Nick Craswell, and Bhaskar Mitra. Large language models can accu- rately predict searcher preferences. InProceedings of the 47th International ACM SIGIR Confer- ence on Research and Development in Information Retrieval (SIGIR), pp. 1930–1940,

  9. [23]

    arXiv:2309.10621

    doi: 10.1145/3626772.3657707. arXiv:2309.10621. Pat Verga, Sebastian Hofst ¨atter, Sophia Althammer, Yixuan Su, Aleksandra Piktus, Arkady Arkhangorodsky, Minjie Xu, Naomi White, and Patrick Lewis. Replacing judges with juries: Evaluating LLM generations with a panel of diverse...

  10. [24]

    CRITIC-R1: Learning structured critics for retrieval-augmented generation.arXiv preprint arXiv:2605.29886,

    Wenhan Xiao, Ziwei Zhang, Chuanyue Yu, Xingcheng Fu, Qingyun Sun, Runhua Xu, and Jianxin Li. CRITIC-R1: Learning structured critics for retrieval-augmented generation.arXiv preprint arXiv:2605.29886,

  11. [25]

    Qwen2.5 technical report.arXiv preprint arXiv:2412.15115,

    An Yang, Baosong Yang, Beichen Zhang, Binyuan Hui, Bo Zheng, Bowen Yu, Chengyuan Li, Dayiheng Liu, Fei Huang, Haoran Wei, et al. Qwen2.5 technical report.arXiv preprint arXiv:2412.15115,

  12. [26]

    Qwen3 technical report.arXiv preprint arXiv:2505.09388,

    An Yang, Anfeng Li, Baosong Yang, Beichen Zhang, Binyuan Hui, Bo Zheng, Bowen Yu, Chengyuan Gao, Chengen Huang, Chenxu Lv, et al. Qwen3 technical report.arXiv preprint arXiv:2505.09388,

  13. [27]

    arXiv:2406.19215

    doi: 10.18653/v1/2025.acl-long.1312. arXiv:2406.19215. Qiying Yu, Zheng Zhang, Ruofei Zhu, Yufeng Yuan, Xiaochen Zuo, Yu Yue, Weinan Dai, Tiantian Fan, Gaohong Liu, Juncai Liu, Lingjun Liu, Xin Liu, Haibin Lin, Zhiqi Lin, Bole Ma, Guangming Sheng, Yuxuan Tong, Chi Zhang, Mofan...

  14. [28]

    Jingyang Zhang, Jingwei Sun, Eric Yeats, Yang Ouyang, Martin Kuo, Jianyi Zhang, Hao Frank Yang, and Hai Li

    arXiv:2503.14476. Jingyang Zhang, Jingwei Sun, Eric Yeats, Yang Ouyang, Martin Kuo, Jianyi Zhang, Hao Frank Yang, and Hai Li. Min-k%++: Improved baseline for detecting pre-training data from large language models. InInternational Conference on Learning Representations (ICLR), ...

  15. [30]

    Following Liu et al

    replaces a learned value baseline with the statistics of the sampled group: the advantage of polarizerψi is its reward standardized over theGrewards of that group, ˆAi = S(ψi)−mean S(ψj) G j=1 std S(ψj) G j=1 ,(6) broadcast to every token ofψ i. Following Liu et al. (2025) and...

  16. [31]

    Winning all70cells

    is read from the very prompt that produces the scored answer. We isolate that factor by re-scoring IGP’s per-passage all-token mean entropy¯HL(q,p)under the shared answering prompt, with the same frozen respondent on the same five pools (Table 6). The two macro means are close...

  17. [33]

    (2024) in its multi-reference branch

    Table 8:The LLM-judge prompt, adoptedverbatimfrom the official release of Farquhar et al. (2024) in its multi-reference branch. Braced tokens are the per-example fields substituted at query time. Prompt template Instruction We are assessing the quality of answers to the follow...

  18. [34]

    0.3450 0.4450 0.6070 0.3135 0.2060 0.3833 4.3 × 10−9 4.0 × 10−7 Min-K% Prob (Shi et al.,

  19. [35]

    0.3210 0.4440 0.5890 0.3075 0.2150 0.3753 6.1 × 10 −18 1.1 × 10−13 EigenScore (Chen et al., 2024a) 0.3120 0.4430 0.5930 0.3026 0.2050 0.3711 1.6 × 10 −17 2.4 × 10−13 IGP (Song et al.,

  20. [36]

    0.3280 0.4010 0.6040 0.3194 0.2010 0.3707 9.0 × 10 −16 4.5 × 10−12 MBA-RAG (Tang et al.,

  21. [37]

    0.3010 0.3800 0.5480 0.2669 0.1930 0.3378 2.6 × 10 −42 4.3 × 10−31 SEARCH-BASED GUIDANCE MIPROv2 (Opsahl-Ong et al.,

  22. [38]

    0.3990 0.0800 0.2990 0.0923 0.1120 0.1965 1.2 × 10 −201 1.3 × 10−128 REFERENCE LINES random 0.1928 0.1731 0.4795 0.1752 0.1550 0.2351— —rank1 0.3160 0.4120 0.5750 0.3056 0.1890 0.3595 1.5 × 10 −21 2.6 × 10−16 oracle 0.5710 0.6930 0.7700 0.5099 0.3960 0.5880 2.4 × 10 −229 3.5 ×...

  23. [39]

    The passage may address a similar entity or context butnotthe specific one asked for, and it might contain outdated, fictional, or misattributed information

    0.5190 0.6110 0.6850 0.6161 0.5570 0.5976 2.6 × 10 −10 5.0 × 10−8 REFERENCE LINES random 0.3560 0.3195 0.5797 0.3008 0.4057 0.3923— —rank1 0.5130 0.6220 0.6840 0.5625 0.4930 0.5749 4.4 × 10 −24 3.9 × 10−18 oracle 0.8240 0.9220 0.8710 0.7778 0.7600 0.8310 1.4 × 10 −255 2.1 × 10...

  24. [40]

    Figure 3 plots what the run was maximising while that surface form decayed

    and at0.6420on the seed-42run this table’s reference row reports. Figure 3 plots what the run was maximising while that surface form decayed. The reward rises over exactly the steps in which the sampled responses pin to the length cap, and CTRL-RAG’s objective scores no proper...

  25. [41]

    runs its official implementation, reflective evolution with Pareto candidate selection and gpt-oss-120b as the reflection LM, at two settings. At its official sample-efficiency setting, all six runs we launched—three seeds at each of two reflection-minibatch sizes—returned the...

  26. [2013]

    INSIDE: LLMs’ internal states retain the power of hallucination detection

    Chao Chen, Kai Liu, Ze Chen, Yi Gu, Yue Wu, Mingyuan Tao, Zhihang Fu, and Jieping Ye. INSIDE: LLMs’ internal states retain the power of hallucination detection. InInternational Conference on Learning Representations (ICLR), 2024a. arXiv:2402.03744. Jianlyu Chen, Shitao Xiao, P...

  27. [2016]

    Simple entity-centric ques- tions challenge dense retrievers

    Christopher Sciavolino, Zexuan Zhong, Jinhyuk Lee, and Danqi Chen. Simple entity-centric ques- tions challenge dense retrievers. InProceedings of the 2021 Conference on Empirical Methods in Natural Language Processing (EMNLP),

  28. [2021]

    Zhihong Shao, Peiyi Wang, Qihao Zhu, Runxin Xu, Junxiao Song, Xiao Bi, Haowei Zhang, Mingchuan Zhang, Y

    arXiv:2109.08535. Zhihong Shao, Peiyi Wang, Qihao Zhu, Runxin Xu, Junxiao Song, Xiao Bi, Haowei Zhang, Mingchuan Zhang, Y . K. Li, Y . Wu, and Daya Guo. DeepSeekMath: Pushing the limits of math- ematical reasoning in open language models.arXiv preprint arXiv:2402.03300,

  29. [2022]

    Zhanghao Hu, Qinglin Zhu, Siya Qi, Yulan He, Hanqi Yan, and Lin Gui

    arXiv:2106.09685. Zhanghao Hu, Qinglin Zhu, Siya Qi, Yulan He, Hanqi Yan, and Lin Gui. Beyond perplexity: Let the reader select retrieval summaries via spectrum projection score. InAAAI Conference on Artificial Intelligence, pp. 31104–31112,

  30. [2023]

    12 APPENDIX CONTENTS A Notation.Every symbol used in the paper, collected

    arXiv:2306.05685. 12 APPENDIX CONTENTS A Notation.Every symbol used in the paper, collected. B GRPO objective and hyper-parameters.The objective of Equation 5 in full, and every hyper-parameter. C Aligned-probe ablation.What IGP’s gain term scores once its probe reads the answ...

  31. [2024]

    Jonathan Berant, Andrew Chou, Roy Frostig, and Percy Liang

    arXiv:2310.11511. Jonathan Berant, Andrew Chou, Roy Frostig, and Percy Liang. Semantic parsing on Freebase from question-answer pairs. InProceedings of the 2013 Conference on Empirical Methods in Natural Language Processing (EMNLP), pp. 1533–1544,

  32. [2025]

    arXiv:2507.22209

    doi: 10.18653/v1/2025.ijcnlp-short.4. arXiv:2507.22209. Sebastian Farquhar, Jannik Kossen, Lorenz Kuhn, and Yarin Gal. Detecting hallucinations in large language models using semantic entropy.Nature, 630:625–630,

  33. [2026]

    Akari Asai, Zeqiu Wu, Yizhong Wang, Avirup Sil, and Hannaneh Hajishirzi

    arXiv:2507.19457. Akari Asai, Zeqiu Wu, Yizhong Wang, Avirup Sil, and Hannaneh Hajishirzi. Self-RAG: Learning to retrieve, generate, and critique through self-reflection. InInternational Conference on Learning Representations (ICLR),

Pith tools

Reviewed August 16, 2026 · model on record in the stance chip above.