Pith. sign in

REVIEW 3 major objections 4 minor 54 references

REVEAL: A Rubric-Guided Agent for Explicit Evidence Sufficiency Verificationin Long-Video Question Answering

T0 review · 3 major / 4 minor · reviewed 2026-08-14 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read Long-video QA errors trace to stopping retrieval at relevance; an explicit sufficiency check recovers the decisive evidence.

desk verdict A genuinely new inference-time mechanism with impressive benchmark numbers, but the main ablation does not isolate sufficiency verification from rubric-based reranking. read the letter →

arxiv 2608.08612 v1 pith:ICVPFNZV submitted 2026-08-09 cs.CV cs.AI

Caijun Yan , Yang Zhou , Meixing Shi , Haoran Sun , Yichen Li , Yuxiang Cai , Yankai Jiang This is my paper · ORCID
classification cs.CVcs.AI
keywords long-videoquestionansweringevidencesufficiencyrubric-guidedagentretrieval-augmentedreasoningvideomemorycontrastiverubricconstructionagenticretrievalmultimodallargelanguagemodel
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

Long-video question answering, the paper argues, is limited less by how much relevant context a retrieval loop can find than by when it decides to stop: relevance-based stopping accepts topically related segments even when the decisive temporal, causal, or fine-grained detail is absent. REVEAL replaces that implicit stopping rule with an explicit evidence-sufficiency check. A fixed library of weighted criteria, built automatically from contrastive pairs of decisive and misleading evidence, is selected per question; the verifier scores the accumulated evidence against those criteria and, if the score falls below a threshold, produces a concrete missing-evidence analysis that drives a targeted repair query. On five long-video QA benchmarks, the training-free system reports state-of-the-art accuracy, and the single largest ablation gain comes from adding the sufficiency check. The paper's point is that “enough evidence” is a property that can be verified independently of relevance, and that making it explicit is what retrieves the decisive clues.

What carries the argument

The load-bearing mechanism is the rubric-guided retrieve–verify–answer loop. A rubric library is a fixed set of weighted evidence-sufficiency criteria, each with explicit score boundaries (0 missing, 0.5 partial, 1 sufficient), induced offline from contrastive pairs of evidence sets separated by whether they flip the answer model from wrong to right. At inference, the verifier selects the subset of criteria pertinent to the question, computes a weighted aggregate score, and either accepts the evidence or emits a missing-evidence analysis; the planner turns that analysis into targeted retrieval over a hierarchical offline memory of variable-length, visually coherent segments, with event-timeline and global-narrative summaries, while an online memory tracks admitted evidence, visited segments, query history, and per-round retrieval quality. This separation of “how much evidence is enough” (the scalar score) from “what is still missing” (the analysis) is what makes the repair retrieval directed rather than redundant.

What would settle it

On a held-out set of long-video questions with manually annotated decisive-evidence intervals, compare the verifier's missing-evidence analysis with the actual content of those intervals; if questions whose answer flips from wrong to right after one targeted repair are not the ones the verifier flagged as missing that clue, the claimed sufficiency signal is not what drives the gain.

Watch

Extended reading notes

Core claim

On its own terms, REVEAL's central discovery is that an automatically built rubric of evidence-sufficiency criteria, applied inside the retrieval loop, turns the retrieval-stopping decision into a calibrated, inspectable judgment. The system's verifier selects the question-relevant criteria from a shared rubric pool, assigns each a score of 0, 0.5, or 1 for the accumulated evidence, and stops only when the weighted confidence passes a threshold; on failure it returns a missing-evidence analysis rather than a generic “keep searching” signal. The planner then uses that analysis together with query history and retrieval-quality flags to issue targeted queries, or falls back to uniform timeline sampling when directed search is exhausted. After at most three rounds, the answerer receives a compact evidence cache. The authors report that this verification step—not larger memory, not more retrieved segments—produces the decisive jump in accuracy, and that the same fixed rubric pool transfers across egocentric and third-person benchmarks without per-dataset tuning.

Load-bearing premise

The load-bearing premise is that the rubric library, built from training questions and filtered by a single model's judgments of what makes evidence sufficient, keeps its meaning on unseen videos, questions, and answer models without retuning.

Editorial extensions

If this is right

  • If sufficiency, not relevance, is the right stopping criterion, retrieval systems for long videos should report an evidence-sufficiency score alongside each answer, so users can see when a response is under-supported.
  • Because a single rubric pool transfers across both egocentric and third-person benchmarks, evidence-sufficiency criteria appear to be reusable reasoning requirements rather than dataset-specific heuristics.
  • The finding that accuracy peaks with about a quarter of the evidence volume, and declines when more relevant-but-redundant segments are added, implies that context budgets should be set by sufficiency rather than by maximum context length.
  • The approach is training-free and uses one backbone model for planning, verification, and answering, suggesting that the same control loop can be dropped into existing retrieval-augmented pipelines without fine-tuning.

Reading between the lines

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

  • A natural extension the paper does not test: because the contrastive-pair filter leans on the answer model's own correctness, swapping in a stronger or weaker teacher should shift the sharpness of every rubric boundary, so verifier calibration becomes a measurable proxy for teacher quality.
  • The verifier's missing-evidence analysis is already an explanation of what the video fails to establish; surfacing it to users as the justification for an answer, or for an admission of under-support, is an interpretability feature the paper mentions only in passing.
  • The same contrastive sufficiency logic should transfer to image QA and document QA, where topically relevant passages also omit decisive details; testing REVEAL's loop outside video would reveal whether “evidence sufficiency” is a general agentic-retrieval control or a video-specific fix.
  • The stopping threshold is treated as a tuned constant; varying it as a user-facing confidence knob would let the same system trade retrieval cost against answer reliability, something the efficiency analysis suggests is feasible.
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

3 major / 4 minor

Summary. REVEAL is a training-free agentic framework for long-video question answering. It builds an offline hierarchical memory (L1 segments, L2 event timelines, L3 global narratives) from visually coherent, variable-length segments, and combines it with a question-conditioned online memory that tracks retrieved evidence, search history, and the verifier's missing-evidence analysis. The core novelty is an automatically constructed rubric library, induced from contrastive pairs of decisive and misleading evidence, that an inference-time verifier uses to decide whether accumulated evidence is sufficient (Eq. 5) and, if not, to produce concrete diagnoses that drive targeted re-retrieval. The paper reports state-of-the-art accuracy on five long-video QA benchmarks with a single open backbone (Qwen3.5-27B), and a component ablation in Table 2 attributes the largest gains to rubric-guided sufficiency verification.

Significance. If the central claim holds, REVEAL is a valuable contribution: it demonstrates that an interpretable, automatically constructed evidence-sufficiency criterion can be inserted into the inference loop of a long-video QA agent without any gradient-based training, and it provides extensive same-backbone comparisons with existing RAG and memory-augmented methods. The paper also ships useful analyses: the evidence-volume curve in Figure 4a shows that verification improves accuracy while using far fewer segments, the calibration analysis in Figure 4b uses oracle target-time annotations, and the per-category breakdown in Figure 4c localizes the gains to evidence-sensitive reasoning types. The detailed implementation notes (supplementary Sec. B) make the system largely reproducible. However, the central mechanistic claim that explicit sufficiency verification, rather than better rubric-based relevance ranking, drives the gains is not yet isolated by the reported experiments, and the main benchmark results are partly affected by hyperparameter tuning on the same dataset.

major comments (3)
  1. [§B.3, Table 2] The implementation described in §B.3 admits evidence via 'the candidates are merged across queries and rubric-reranked to 8 evidence pieces before being admitted to the cache.' The 'Rubric Suff.' row in Table 2 therefore adds both the verifier/repair loop and rubric-based candidate reranking relative to M+OnMem, while the 'disable verification' analysis in Figure 4a also removes the whole rubric component rather than only the sufficiency judgment. The reported +7.8/+10.9 gain thus conflates two distinct mechanisms, and the paper's headline claim that explicit evidence-sufficiency verification, rather than improved relevance ranking, drives the gains is not yet supported. Please add an ablation that includes rubric reranking without the verifier/repair loop, and ideally one that keeps verification while using relevance-only admission, so that the two mechanisms can be separated.
  2. [§B.5, Table 1] Section B.5 states that the hyperparameters K, τ, and θ were tuned on Video-MME-L during development, and Table 1 reports Video-MME-L as one of the five main benchmarks. The Video-MME-L accuracy of 79.1 is therefore not fully held-out, and the same applies to the Video-MME-L row in Table 2. This weakens the headline claim of consistent state-of-the-art performance across all five benchmarks. Please either tune on a separate development set and report fixed-config results on all benchmarks, or explicitly qualify Video-MME-L as a development-tuned result in the main text.
  3. [§3.2, Eq. (3)] The rubric library is constructed from contrastive pairs filtered by the answer model A's own correctness (Eq. 3), and the same backbone Qwen3.5-27B later serves as verifier and answerer. The sufficiency criteria therefore partially encode the decision boundary of the exact model they are meant to audit; if these criteria capture model-specific blind spots or training-distribution regularities rather than generic evidence-sufficiency requirements, the verifier's verdicts may be miscalibrated at test time. The 'one rubric pool across benchmarks' analysis in §4.4 does not fully settle this, because all five evaluation benchmarks share the same construction and evaluation protocol. A concrete transfer test would be to build the rubric library with a different teacher model, or on one benchmark family only (e.g., third-person), and then evaluate the same backbone on the remaining benchmarks; such an experiment would substantially strengthen the transferability claim.
minor comments (4)
  1. [Abstract, §B.4] The phrase 'without any extra training' is stronger than what is actually done: the rubric library is produced by an offline, LLM-based refinement loop over training pairs described in §B.4. Suggest rewording to 'without gradient-based training' or 'without fine-tuning' to avoid overclaiming.
  2. [§4.1, Table 1] The paper reports that each result comes from a single evaluation run, and some differences in Table 1 are small (e.g., 79.1 vs. 77.0 for Vgent on Video-MME-L). Reporting variance across multiple runs, or at least stating this as a limitation, would make the consistent-improvement claim more robust.
  3. [Figure 4c] The per-category analysis includes a category ('Spatial Perception') with only three questions, so the claim that no category regresses rests on a very small sample for that category. Consider aggregating small categories or explicitly noting the low sample size.
  4. [§B.2, §1] Section B.2 notes that L1 segments are approximately 10 seconds long on average, which is the same scale as the fixed 10s chunks criticized in §1. The intended distinction is variable-length, event-boundary-aware grouping rather than a different segment scale; the paper should state this explicitly to avoid an apparent inconsistency.

Circularity Check

0 steps flagged · score 0.0 of 10

No constructional circularity: the rubric construction is self-referential but not tautological, and the main ablation confound is an attribution issue rather than a circular derivation.

full rationale

REVEAL's claimed derivation chain is not circular in the constructional sense. The rubric library is induced offline from contrastive pairs filtered by Equation (3), i.e., pairs in which the answer model A is correct with E+, incorrect with E-, and incorrect with no evidence; the resulting criteria are therefore a distillation of A's own decision boundary, and the same A later serves as verifier and answerer. This is self-referential, but not tautological: the rubrics are fixed on the training split, applied unchanged to unseen test questions, and the reported outcome is accuracy against external gold answers, not agreement with A's self-judgment. No load-bearing self-citation, imported uniqueness theorem, or ansatz smuggled in via citation is present; the framework's components are described with their own prompts and equations. The strongest concern in the paper is an attribution confound rather than circularity: Section B.3 states that in each retrieval round 'the candidates are merged across queries and rubric-reranked to 8 evidence pieces before being admitted to the cache,' so the 'Rubric Suff.' ablation in Table 2 adds both rubric-based reranking and sufficiency verification together, and the reported +7.8/+10.9 gain cannot be isolated to verification alone. That is a validity risk for the mechanistic claim, but it does not make any prediction equal to its inputs by construction, so it does not raise the circularity score.

Assumptions & free parameters 5 free parameters · 5 assumptions · 2 invented entities

The central claim rests on several fitted hyperparameters (theta, K, tau, rubric weights) and on domain assumptions about the transferability of the rubric library, the reliability of the backbone as teacher, and the representativeness of the selected benchmarks. No new physical entities are introduced; the rubric pool and memory hierarchy are software artifacts with inspectable behavior.

free parameters (5)
  • theta (visual similarity threshold) = 0.8
    Controls frame grouping into event segments; tuned on Video-MME-L development set (Section B.5).
  • K (max retrieval rounds) = 3
    Maximum number of retrieve-verify-answer iterations; tuned on Video-MME-L dev (Section B.5).
  • tau (sufficiency threshold) = 0.7
    Threshold on weighted rubric score that triggers answering; tuned on Video-MME-L dev (Section B.5).
  • Rubric weights w_k = Not reported
    Fitted during rubric refinement against held-out contrastive pairs (Section B.4); no final values given.
  • Retrieval and dedup thresholds = Dedup 0.9, query rejection Jaccard 0.9 / BGE cosine 0.70, top 8 L1 / 4 L2 / 2 L3
    Fixed design choices in the executor and online memory (Section B.3); not tuned but affect retrieval behavior.
assumptions (5)
  • domain assumption Visual similarity grouping at 1 fps with threshold 0.8 yields natural event units that preserve coherent events.
    Section 3.1 and B.2: if the threshold splits coherent events or merges distinct ones, the memory fragments or conflates evidence, undermining retrieval.
  • domain assumption Contrastive pair filter (Eq. 3) ensures E+ is genuinely sufficient and E- is genuinely insufficient.
    Section 3.2: relies on the answer model's judgments being reliable for the training set; if the backbone is systematically wrong on some types, the induced rubrics encode those errors.
  • domain assumption A single rubric pool with weighted threshold tau=0.7 transfers across all five benchmarks without per-benchmark tuning.
    Section 4.4 and B.5: the main results depend on this transferability; the paper provides no per-benchmark adaptation.
  • domain assumption The answer model Qwen3.5-27B is a reliable teacher for labeling decisive vs misleading evidence.
    Section 3.2: all contrastive pairs are filtered by this model's correctness, so its biases directly shape the rubric library.
  • domain assumption The selected benchmarks and their subsets are representative of long-video QA.
    Appendix A: includes a subset of LVBench and the validation split of LongVideoBench; results may not fully reflect the entire benchmarks.
invented entities (2)
  • Unified rubric pool independent evidence
    purpose: A library of weighted criteria induced offline from contrastive evidence pairs; it scores evidence sufficiency and controls the retrieve-verify-answer loop.
    The rubric pool is an inspectable artifact whose transferability is tested on held-out benchmarks, though it is not a physical entity with independent falsifiability.
  • Offline-online memory (L1/L2/L3 hierarchy and online state O_t) independent evidence
    purpose: A structured video representation and per-question reasoning state that supports multi-scale retrieval and tracks evidence gaps and search history.
    The memory is a software construct whose contribution is validated in ablations; it is not a physical entity but has measurable behavior.

how reviews work

0 comments
Cite this review

Pith. "Pith review of REVEAL: A Rubric-Guided Agent for Explicit Evidence Sufficiency Verificationin Long-Video Question Answering." pith.science (2026). https://pith.science/paper/ICVPFNZV

@misc{pith2026260808612,
  author       = {Pith},
  title        = {Pith review of: REVEAL: A Rubric-Guided Agent for Explicit Evidence Sufficiency Verificationin Long-Video Question Answering},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/ICVPFNZV}},
  note         = {Machine review of arXiv:2608.08612}
}
read the original abstract

Recently, retrieval-augmented and memory-augmented methods have emerged as two promising paradigms for long-video question answering. However, existing methods typically rely on rigid, fixed-length temporal chunking (e.g., 10s) and static offline memory banks, which not only fragment coherent continuous events but also fail to adapt during real-time reasoning. Moreover, whether using multi-scale summaries or multimodal knowledge graphs, current approaches prioritize retrieval relevance while overlooking evidence sufficiency, often stopping to answer once only semantically relevant clues are retrieved, even when key temporal, causal, or fine-grained action evidence is still missing. To tackle these challenges, we propose REVEAL, a rubric-guided agent framework. As a foundation, we introduce an adaptive visual-similarity-based preprocessing pipeline that groups visually coherent adjacent frames into natural event units to construct an offline-online video memory---capturing global video context offline while dynamically maintaining question-conditioned memory online. Built upon this structured memory, REVEAL uses an automatically constructed rubric library to explicitly verify whether retrieved evidence satisfies sufficiency criteria, pinpoints missing clues upon verification failure, and directs targeted re-retrieval for complementary information. Without any extra training, REVEAL consistently outperforms both closed-source and open-source state-of-the-art methods across extensive experiments. These results show that explicitly verifying evidence sufficiency, rather than stopping at semantic relevance, retrieves the decisive clues that prior methods miss and yields more reliable long-video reasoning.

Figures

Figures reproduced from arXiv: 2608.08612 by the authors.

Figure 1
Figure 1. Comparison of retrieval-augmented, memory-augmented, and REVEAL paradigms for long-video QA. REVEAL [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Overview of REVEAL. A hierarchical offline memory supports multi-scale retrieval; the online memory tracks [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Contrastive rubric construction pipeline. A teacher [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: Analysis of the sufficiency verifier: (a) accuracy vs. number of retrieved segments; (b) verifier calibration on Ego-R1; [PITH_FULL_IMAGE:figures/full_fig_p007_4.png]
Figure 5
Figure 5. Figure 5: A representative retrieve–verify–repair trace [PITH_FULL_IMAGE:figures/full_fig_p014_5.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

54 extracted references · 38 canonical work pages

  1. [1]

    Fu, Chaoyou and Dai, Yuhan and Luo, Yongdong and Li, Lei and Ren, Shuhuai and Zhang, Renrui and Wang, Zihan and Zhou, Chenyu and Shen, Yunhang and Zhang, Mengdan and others , journal=. Video-

    work page

  2. [2]

    Zhou, Junjie and Shu, Yan and Zhao, Bo and Wu, Boya and Liang, Zhengyang and Xiao, Shitao and Qin, Minghao and Yang, Xi and Xiong, Yongping and Zhang, Bo and others , journal=

    work page

  3. [3]

    Wu, Haoning and Li, Dongxu and Chen, Bei and Li, Junnan , journal=

    work page

  4. [4]

    Wang, Weihan and He, Zehai and Hong, Wenyi and Cheng, Yean and Zhang, Xiaohan and Qi, Ji and Gu, Xiaotao and Huang, Shiyu and Xu, Bin and Dong, Yuxiao and others , journal=

    work page

  5. [5]

    Wang, Xiaohan and Zhang, Yuhui and Zohar, Orr and Yeung-Levy, Serena , journal=

    work page

  6. [6]

    Fan, Yue and Ma, Xiaojian and Wu, Rujie and Du, Yuntao and Li, Jiaqi and Gao, Zhi and Li, Qing , journal=

    work page

  7. [7]

    Yang, Zeyuan and Chen, Delin and Yu, Xueyang and Shen, Maohao and Gan, Chuang , journal=

    work page

  8. [8]

    Chen, Boyu and Yue, Zhengrong and Chen, Siran and Wang, Zikang and Liu, Yang and Li, Peng and Wang, Yali , journal=

    work page

Show all 54 references
  1. [9]

    Retrieval-Augmented Generation for Knowledge-Intensive

    Lewis, Patrick and Perez, Ethan and Piktus, Aleksandra and Petroni, Fabio and Karpukhin, Vladimir and Goyal, Naman and K. Retrieval-Augmented Generation for Knowledge-Intensive. Advances in Neural Information Processing Systems , volume=

  2. [10]

    Yao, Shunyu and Zhao, Jeffrey and Yu, Dian and Du, Nan and Shafran, Izhak and Narasimhan, Karthik and Cao, Yuan , booktitle=

  3. [11]

    Advances in Neural Information Processing Systems , year=

    Reflexion: Language Agents with Verbal Reinforcement Learning , author=. Advances in Neural Information Processing Systems , year=

  4. [12]

    Liu, Yang and Iter, Dan and Xu, Yichong and Wang, Shuohang and Xu, Ruochen and Zhu, Chenguang , booktitle=

  5. [13]

    and others , booktitle=

    Zheng, Lianmin and Chiang, Wei-Lin and Sheng, Ying and Zhuang, Siyuan and Wu, Zhanghao and Zhuang, Yonghao and Lin, Zi and Li, Zhuohan and Li, Dacheng and Xing, Eric P. and others , booktitle=. Judging

  6. [14]

    arXiv preprint arXiv:2310.08491 , year=

    Prometheus: Inducing Fine-grained Evaluation Capability in Language Models , author=. arXiv preprint arXiv:2310.08491 , year=

    arXiv

  7. [15]

    Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

    Sigmoid Loss for Language Image Pre-Training , author=. Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

  8. [16]

    Chen, Jianlv and Xiao, Shitao and Zhang, Peitian and Luo, Kun and Lian, Defu and Liu, Zheng , journal=

  9. [17]

    International Conference on Machine Learning , pages=

    Learning Transferable Visual Models from Natural Language Supervision , author=. International Conference on Machine Learning , pages=

  10. [18]

    arXiv preprint arXiv:2512.05774 , year=

    Active Video Perception: Iterative Evidence Seeking for Agentic Long Video Understanding , author=. arXiv preprint arXiv:2512.05774 , year=

    arXiv

  11. [19]

    Kurpath, Mohammed Irfan and Kaithakkodan, Jaseel Muhammad and Zhou, Jinxing and Mullappilly, Sahal Shaji and Almansoori, Mohammad and Ahsan, Noor and Kalmakhanbet, Beknur and Shikhar, Sambal and Lalla, Rishabh and Lahoud, Jean and others , journal=

  12. [20]

    Waheed, Abdul and Wu, Zhen and Alharthi, Dareen and Kim, Seungone and Raj, Bhiksha , journal=

  13. [21]

    and Wei, Jason and Hicks, Rebecca Soskin and Bowman, Preston and Qui

    Arora, Rahul K. and Wei, Jason and Hicks, Rebecca Soskin and Bowman, Preston and Qui. arXiv preprint arXiv:2505.08775 , year=

    arXiv

  14. [22]

    Chen, Guo and Liu, Yicheng and Huang, Yifei and He, Yuping and Pei, Baoqi and Xu, Jilan and Wang, Yali and Lu, Tong and Wang, Limin , journal=

  15. [23]

    Liu, Ye and Lin, Kevin Qinghong and Chen, Chang Wen and Shou, Mike Zheng , journal=

  16. [24]

    arXiv preprint , year=

    The Rules of the Game: A Survey of Rubrics for Large Language Models , author=. arXiv preprint , year=

  17. [25]

    Liu, Tianci and Xu, Ran and Yu, Tony and Hong, Ilgee and Yang, Carl and Zhao, Tuo and Wang, Haoyu , journal=

  18. [26]

    Li, Sunzhu and Zhao, Jiale and Wei, Miteto and Ren, Huimin and Zhou, Yang and Yang, Jingwen and Liu, Shunyu and Zhang, Kaike and Chen, Wei , journal=

  19. [27]

    Fan, Zhiyuan and Wang, Weinong and Wu, Xing and Zhang, Debing , journal=

  20. [28]

    Gao, Shanghua and Su, Yuchang and Sui, Pengwei and Ginder, Curtis and Zitnik, Marinka , journal=

  21. [29]

    Qi, Zhengyang and Dickens, Charles and Pham, Derek and Dsouza, Amanda and Parchami, Armin and Sala, Frederic and Varma, Paroma , journal=

  22. [30]

    and Batra, Sanjit Singh and Tillman, Robert E

    Yang, Zhichao and Janghorbani, Sepehr and Zhang, Dongxu and Han, Jun and Qian, Qian and Ressler, Andrew and Lyng, Gregory D. and Batra, Sanjit Singh and Tillman, Robert E. , journal=

  23. [31]

    arXiv preprint arXiv:1707.06347 , year=

    Proximal Policy Optimization Algorithms , author=. arXiv preprint arXiv:1707.06347 , year=

    arXiv

  24. [32]

    Yeo, Jeong Hun and Chung, Sangyun and Park, Sungjune and Kim, Dae Hoe and Moon, Jinyoung and Ro, Yong Man , journal=

  25. [33]

    Tian, Shulin and Wang, Ruiqi and Guo, Hongming and Wu, Penghao and Dong, Yuhao and Wang, Xiuying and Yang, Jingkang and Zhang, Hao and Zhu, Hongyuan and Liu, Ziwei , journal=

  26. [34]

    Maaz, Muhammad and Rasheed, Hanoona and Khan, Salman and Khan, Fahad Shahbaz , journal=. Video-

  27. [35]

    Lin, Bin and Ye, Yang and Zhu, Bin and Cui, Jiaxi and Ning, Munan and Jin, Peng and Yuan, Li , journal=. Video-

  28. [36]

    Wang, Peng and Bai, Shuai and Tan, Sinan and Wang, Shijie and Fan, Zhihao and Bai, Jinze and Chen, Keqin and Liu, Xuejing and Wang, Jialin and Ge, Wenbin and others , journal=

  29. [37]

    Shen, Xiaoqian and Xiong, Yunyang and Zhao, Changsheng and Wu, Lemeng and Chen, Jun and Zhu, Chenchen and Liu, Zechun and Xiao, Fanyi and Varadarajan, Balakrishnan and Bordes, Florian and others , journal=

  30. [38]

    Song, Enxin and Chai, Wenhao and Wang, Guanhong and Zhang, Yucheng and Zhou, Haoyang and Wu, Feiyang and Chi, Xun and Guo, Xun and Ye, Tian and Zhang, Yanting and others , booktitle=

  31. [39]

    He, Bo and Li, Hengduo and Jang, Young Kyun and Jia, Menglin and Cao, Xuefei and Shah, Ashish and Shrivastava, Abhinav and Lim, Ser-Nam , booktitle=

  32. [40]

    Luo, Yongdong and Zheng, Xiawu and Li, Guilin and Yin, Shukang and Lin, Haojia and Fu, Chaoyou and Huang, Jinfa and Ji, Jiayi and Chao, Fei and Luo, Jiebo and Ji, Rongrong , journal=. Video-

  33. [41]

    Yeo, Woongyeong and Kim, Kangsan and Yoon, Jaehong and Hwang, Sung Ju , journal=

  34. [42]

    Breaking the Exploration Bottleneck: Rubric-Scaffolded Reinforcement Learning for General

    Zhou, Yang and Li, Sunzhu and Liu, Shunyu and Fang, Wenkai and Zhang, Kongcheng and Zhao, Jiale and Yang, Jingwen and Zhou, Yihe and Lv, Jianwei and Zheng, Tongya and Lu, Hengtong and Chen, Wei and Xie, Yan and Song, Mingli , journal=. Breaking the Exploration Bottleneck: Rubr...

  35. [43]

    International Conference on Learning Representations (ICLR) , year=

    Chasing the Tail: Effective Rubric-based Reward Modeling for Large Language Model Post-Training , author=. International Conference on Learning Representations (ICLR) , year=

  36. [44]

    2025 , doi=

    Wang, Ziyang and Yoon, Jaehong and Yu, Shoubin and Islam, Md Mohaiminul and Bertasius, Gedas and Bansal, Mohit , booktitle=. 2025 , doi=

    2025

  37. [45]

    Bai, Shuai and Cai, Yuxuan and Chen, Ruizhe and others , journal=

  38. [46]

    arXiv preprint arXiv:2604.15804 , year=

    arXiv

  39. [47]

    2025 , howpublished=

    2025

  40. [48]

    Guti. From. arXiv preprint arXiv:2502.14802 , year=

    arXiv

  41. [49]

    Shen and others , year=

  42. [50]

    Long and others , year=

  43. [51]

    Zuo, Jialong and Deng, Yongtai and Kong, Lingdong and Yang, Jingkang and Jin, Rui and Zhang, Yiwei and Sang, Nong and Pan, Liang and Liu, Ziwei and Gao, Changxin , journal=

  44. [52]

    Asai, Akari and Wu, Zeqiu and Wang, Yizhong and Sil, Avirup and Hajishirzi, Hannaneh , booktitle=. Self-

  45. [53]

    Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing , pages=

    Active Retrieval Augmented Generation , author=. Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing , pages=

    2023

  46. [54]

    arXiv preprint arXiv:2401.15884 , year=

    Corrective Retrieval Augmented Generation , author=. arXiv preprint arXiv:2401.15884 , year=

    arXiv

Pith tools

Reviewed August 14, 2026 · model on record in the stance chip above.