Pith.

Computation and Language · arXiv:2611.15268 v1 · submitted 11 Nov 2026 · 56 references

Knowledge in the Harness, Reliability in the Weights: Teacher-Free Co-Evolution of an LLM Agent

Yujia Zhang, Xinqi Ren, Shixiong Zhou, Jia Meng

The paper says

“Small open models suit private customer-service deployments but trail frontier models in two ways: they miss what the written policy requires in a given situation, and they act inconsistently on what they know.”

more of the abstract

We study whether such a model can close both gaps from its own deployment logs and their reference outcomes, without a stronger teacher model. Our starting point is a Gemma 4 12B agent that reads its domain policy from on-demand skills rather than from its system prompt. In each round, the deployed system serves a fresh batch of tasks; the base model diagnoses failed conversations and writes rules into the skills, and a rule survives only if it passes code checks, counterexample self-review, a paired retest, and a noise-calibrated validation gate. Reinforcement learning (GRPO) then trains a LoRA adapter on complete multi-turn, tool-using conversations. On the retail and airline domains of τ2-bench, without training or selecting on official tasks, the agent after two rounds improves on this starting point by 12.5 pass1 points in retail and 19.5 in airline, and by 18.4 and 22.0 pass4 points. It also exceeds the benchmark's reference agent, which runs the same model with the whole policy in its system prompt, by 11.8 and 8.0 pass1 points and by 22.8 and 16.0 pass4 points, closing most of the gap to three frontier models under the same protocol. A 2 × 2 ablation shows that the rules rescue tasks the agent never solved and carry policy knowledge that RL with the rules in context does not internalize, whereas the trained weights make solved tasks reliable.

from the abstract

Pith says

Machine review · deepseek-v4-flash · 13 Nov 2026 · what was read was not recorded · not checked by a person

It can be wrong. Challenge this review or reply to it.

Holds up. A well-engineered demonstration that a 12B model can co-evolve its own harness and weights under programmatic checks, but the evidence is limited to two domains, two rounds, and one base model. This line is wrong

3 minor

How often objections from reviews like this one are right: not yet measured

Are you an author? What you can do.

Not checked: figures, code, what the 56 references say.

3 minor objections

The paper says, at §3.1

Pith says: minor objection 1 of 3

The definition of pass@k is given, but in Table 2 and the text it appears as "pass1" and "pass4". Consider using "pass@1" and "pass@4" consistently to match the notation in the metrics section and avoid ambiguity with column labels.

The paper says, at Table 2 caption

Pith says: minor objection 2 of 3

The caption states "The reference agent places the whole policy in its system prompt; all other rows use the skill-based harness." It would help to remind the reader that the reference agent is a strong baseline (same model, full policy in prompt) and is run under the same protocol, so the gains are not from a simple system-prompt difference.

show 1 more

The paper says, at §5.4, Table 3

Pith says: minor objection 3 of 3

The confidence intervals for the bin-wise means in Table 3 are not shown. Adding bootstrap intervals (as done in Table 11) would let readers assess the uncertainty behind statements such as "rules alone reach 44% success" and "weights alone reach 25% success" for the never-solved bin.

What would settle it

Recompute the correlation between V and official gains including the two rejected harness candidates (from the retry attempts in Table 14 footnote). If the correlation remains above 0.9, the proxy assumption is further validated; if it drops below 0.8, the gate decisions may rely on V-specific patterns that do not transfer identically.

from the skeptic's reading

How the review went

4 readings by deepseek-v4-flash · 13 Nov 2026

  1. Reader holds up confidence high
  2. Skeptic partly agrees 30 s · 26,784 tokens read, 3,743 written
  3. Editor wrote the verdict above 8 s · 27,547 tokens read, 888 written
  4. Referee accept: 3 minor 20 s · 27,006 tokens read, 2,393 written

Not run on this paper: the simulated authors' defence.

Pith did not check this review's objections against the paper.

The review in full

The review in brief, machine-written

This paper shows something real: a 12B Gemma 4 agent improves on τ²-bench by writing its own policy rules and training its own LoRA adapter, with no stronger teacher. The gains are meaningful—12.5 pass@1 in retail, 19.5 in airline—and the ablation cleanly separates what rules do (rescue never-solved tasks) from what RL does (stabilize usually-solved ones). The engineering is careful: programmatic checks, self-review, noise-calibrated gates, and a generated validation set that correlates at r=0.98 with official-task gains.

The soft spots are the ones the authors already state. Two domains on one benchmark, two rounds, one model. The validation set is faithful but not identical to the real tasks; the loop relies on a non-standard user simulator. The conclusions are solid within that scope, but how well the method generalizes to other domains, models, or longer evolution remains open.

Who this paper is for: anyone building private-deployment customer-service agents who wants a practical recipe for closed-loop improvement. It is also useful for researchers studying self-improving LLM agents as a concrete, well-documented case study.

Verdict: this paper deserves a fair peer review. The limitations are honest and the claims are commensurate with the evidence. It is not a paradigm shift, but it is a competent piece of work that moves the needle on an applied problem.

The strongest objection

After a thorough review, I find that the central claim—that a 12B model can co-evolve harness and weights from its own logs under programmatic checks and noise-calibrated gates—is well supported by the evidence. The weakest assumption identified by the reader (the validation set V being a faithful proxy for official tasks) is defended with a high correlation (r=0.98) between gains on V and official tasks, and the authors acknowledge the use of a non-standard simulator and the custom setting. The limitations are openly stated, and the ablation experiments, retest simulations, and gate calibration provide internal validity. No single concern undermines the main conclusion; the paper is methodologically sound and honestly evaluated. I do not have a load-bearing concern that would change the verdict.

The paper presents a well-engineered system with clear experimental design, honest reporting, and strong empirical results. The reader's weakest assumption—that the generated validation set V is a faithful proxy for official tasks—is the most consequential assumption, but the paper provides substantial evidence for it: high correlation (r=0.98), matched knowledge-point coverage, and within-domain agreement within 2.1 points. The limitations, including the single base model, two domains, two rounds, and non-standard simulator, are explicitly acknowledged in the Limitations section. The internal comparisons are valid, and the ablation study convincingly shows the complementary roles of harness rules and trained weights. I agree with the reader's ACCEPT verdict; no additional concern rises to the level of changing it. The suggested concrete test—including rejected candidates—would strengthen the proxy evidence further but is not required to uphold the current claims.

The referee's report

If the claimed results hold, this work shows that a small open model can self-improve in both instruction-following and behavioral consistency without any frontier model in the loop. This is directly relevant to privacy-sensitive customer-service deployments where sending data to a larger API is not an option. The methodology is carefully designed: training batches, validation sets, and official evaluation tasks are disjoint; all harness edits go through multiple automated checks; and the gate decisions are calibrated to measured dialogue noise. The authors openly acknowledge the limitations (single base model, two domains, two rounds, custom simulator), and the central claim is supported by held-out official results, a convincing ablation, and a high validation-to-official correlation.

Pith's reading of the paper (a machine's summary, not the authors')

This paper argues that a small open model can improve its performance on multi-turn customer-service tasks through a self-contained co-evolution loop: it diagnoses failed conversations, writes and tests new rules (the harness), then trains its weights with reinforcement learning on its own rollouts. The key insight is that the two layers play complementary roles—rules supply knowledge the model lacks, while training makes the model apply that knowledge consistently. On the τ²-bench retail and airline domains, a Gemma 4 12B agent starting from a skill-based system that trails frontier models by a wide margin closes most of the gap after two rounds, with gains that transfer to held-out official tasks.

What carries the argument. The central mechanism is the teacher-free co-evolution loop where a Gemma 4 12B model serves as agent, editor, and trainee. The harness is organized as on-demand skills (split from the domain policy), and the editor diagnoses errors from evidence packs built from failed conversations. Edits are constrained by root-cause prescription, self-review, paired retesting, and a validation gate on a generated validation set. GRPO then trains a LoRA adapter on the agent's own rollouts under the new harness. The separation of policy into skills allows local edits and observable knowledge use, and the noise-calibrated gates compensate for dialogue variability.

Core claim. The central discovery is that a 12B model can act as both the agent and the editor, and that its own unreliable judgments can be compensated by programmatic checks and noise-calibrated gates. The evolved rules (harness) and the trained weights (via GRPO) serve different functions: rules rescue tasks the agent never solved and encode policy knowledge that reinforcement learning does not internalize when the rules are present in context, while the trained weights make already-solved tasks more reliable. The 2×2 ablation shows that each layer contributes independently and that their combination yields the largest gains, closing the gap to frontier models (GPT-5.6-sol, Claude Sonnet 5, GLM-5.2) on the benchmark.

The premise it rests on. The generated validation set V is a faithful proxy for the official τ²-bench tasks, so that gains measured on V transfer to the official task distribution.

If the paper is right (Pith's inferences, not the authors')

Small models can reach near-frontier performance on structured customer-service tasks without any teacher data, reducing privacy and infrastructure costs.

The complementary roles of harness rules and trained weights suggest that closed-loop systems should actively evolve both layers rather than relying on one alone.

The self-generated validation set serves as a reliable proxy for official task distributions, enabling version selection without accessing the benchmark's test set.

The co-evolution loop can be extended to other domains with verifiable outcomes, such as booking, scheduling, or compliance dialogues.

The fact that GRPO under the evolved rules fails to internalize rule knowledge confirms that outcome-based RL alone is insufficient for acquiring new policy facts; explicit rule editing is necessary.

The noise-calibrated gating framework could be adapted to other settings where a weak model must evaluate its own candidate improvements, providing a principled way to trade off false positives and false negatives.

Because the loops repair structural defects caused by the skill split—such as the baggage-allowance dependency in airline—this method may be especially valuable when a policy is naturally modular but creates cross-document dependencies.

The finding that rules rescue unsolved tasks while weights stabilize solved ones suggests a general decomposition for self-improving systems: first expand coverage via explicit knowledge, then consolidate via training.

Using the same base model as editor may set an upper bound on the complexity of rules it can write; larger or more capable editors could plausibly accelerate the loop or handle more intricate policy defects.

Check this review

Pith stored no record of this review's passes and took no fingerprint of it. Reviews made by code from 13 Nov 2026 on carry both.

How often reviews like this one are right

Reviews like this one: written by deepseek-v4-flash, where what was read was not recorded.

How often objections from reviews like this one are right: not yet measured

Every value above, in the review's public record (JSON)

Reviews by people

This review was written by a machine. No person has reviewed this paper on Pith yet. A scientist listed on Pith can review it for a fee, and their review is published here whatever it concludes.

Request a human review

How Pith checks papers · The full audit of this paper

The paper

Abstract

Small open models suit private customer-service deployments but trail frontier models in two ways: they miss what the written policy requires in a given situation, and they act inconsistently on what they know. We study whether such a model can close both gaps from its own deployment logs and their reference outcomes, without a stronger teacher model. Our starting point is a Gemma 4 12B agent that reads its domain policy from on-demand skills rather than from its system prompt. In each round, the deployed system serves a fresh batch of tasks; the base model diagnoses failed conversations and writes rules into the skills, and a rule survives only if it passes code checks, counterexample self-review, a paired retest, and a noise-calibrated validation gate. Reinforcement learning (GRPO) then trains a LoRA adapter on complete multi-turn, tool-using conversations. On the retail and airline domains of τ2-bench, without training or selecting on official tasks, the agent after two rounds improves on this starting point by 12.5 pass1 points in retail and 19.5 in airline, and by 18.4 and 22.0 pass4 points. It also exceeds the benchmark's reference agent, which runs the same model with the whole policy in its system prompt, by 11.8 and 8.0 pass1 points and by 22.8 and 16.0 pass4 points, closing most of the gap to three frontier models under the same protocol. A 2 × 2 ablation shows that the rules rescue tasks the agent never solved and carry policy knowledge that RL with the rules in context does not internalize, whereas the trained weights make solved tasks reliable.

The paper's own sentences, arranged, kept beside the review and separate from it. Each label is drawn from a closed vocabulary; the number beside it is the probability the classifier gave that label. No sentence in this record was written by a model.

abstract compiled · 2026-11-13

The authors' main claim, as the classifier picked it out 0.77

“A $2\times2$ ablation shows that the rules rescue tasks the agent never solved and carry policy knowledge that RL with the rules in context does not internalize, whereas the trained weights make solved tasks reliable.”

How the main claim is stated: asserted 1.00

What the abstract itself states, as probabilities: claims first 0.05 · states limitation 0.35 · human participants 0.06 · controlled comparison 0.97 · randomized assignment 0.02 · states quantitative result 0.99 · compares with prior results 0.97 · mentions code or data availability 0.04

The abstract, sentence by sentence

This table labels the abstract only (arXiv listing abstract v1). The number is the probability the classifier gave the label.

SentenceRoleThe authors' own workStates the main conclusionStrength
Small open models suit private customer-service deployments but trail frontier models in two ways: they miss what the written policy requires in a given situation, and they act inconsistently on what they know. background 0.92 0.12 0.03 asserted 1.00
We study whether such a model can close both gaps from its own deployment logs and their reference outcomes, without a stronger teacher model. contribution 0.78 0.30 0.05 hedged possibility 0.46
Our starting point is a Gemma 4 12B agent that reads its domain policy from on-demand skills rather than from its system prompt. method 0.96 0.76 0.03 asserted 0.99
In each round, the deployed system serves a fresh batch of tasks; the base model diagnoses failed conversations and writes rules into the skills, and a rule survives only if it passes code checks, counterexample self-review, a paired retest, and a noise-calibrated validation gate. method 1.00 0.89 0.04 asserted 0.99
Reinforcement learning (GRPO) then trains a LoRA adapter on complete multi-turn, tool-using conversations. method 1.00 0.88 0.03 asserted 0.99
On the retail and airline domains of $\tau^2$-bench, without training or selecting on official tasks, the agent after two rounds improves on this starting point by 12.5 pass$^1$ points in retail and 19.5 in airline, and by 18.4 and 22.0 pass$^4$ points. result 1.00 0.96 0.20 asserted 1.00
It also exceeds the benchmark's reference agent, which runs the same model with the whole policy in its system prompt, by 11.8 and 8.0 pass$^1$ points and by 22.8 and 16.0 pass$^4$ points, closing most of the gap to three frontier models under the same protocol. result 1.00 0.94 0.33 asserted 1.00
A $2\times2$ ablation shows that the rules rescue tasks the agent never solved and carry policy knowledge that RL with the rules in context does not internalize, whereas the trained weights make solved tasks reliable. interpretation 0.74 0.94 0.59 asserted 1.00

What kind of paper the classifier reports this is: method or tool 0.91 · evidence from observation or experiment 0.98

What was examined, and what was not

Sources read

  • arXiv listing abstract v1

Not examined

  • full text
  • figures and tables
  • references and citing sentences
  • relations to other papers

Levels run

  • P abstract

Not run

  • S sections
  • T body sentences
  • C citations
  • R cross-paper relations

How the labels were made

Compiled by classifier-1.13.0 via TypeSafe API, question bank abstract-p1-20260923, 2026-11-13: 1 request, 6,663 input tokens.

Every quoted sentence on this record is the authors'. Every label is from Pith's closed vocabulary, with the probability shown beside it. This record as JSON

Figures (5)
Figure 1 of the paper
Figure 1. One round of co-evolution. The deployed system serves a fresh batch
Figure 2 of the paper
Figure 2. Official scores over two rounds. Dashed lines:
Figure 3 of the paper
Figure 3. The 2 × 2 ablation on the official tasks: the evolved rules alone (H2-M0), the trained weights alone (H0-M2), and both (H2-M2). (a) Gains over H0-M0. (b) Success on the tasks that H0-M0 solved in 0, 1–3, or 4 of its four trials; the three systems are independent runs. layer contrasts. Knowledge stays in the harness. Under H0, the airline allowance table appears only in the book￾ing skill. M2 was trained with the evolved rules, including a copy of the allowance formula, in its context, yet H0-M2 fails all 28 of its baggage￾task conversations that skip the booking skill, as H0-M0…
Figure 4 of the paper
Figure 4. Gain in pass4 against gain in pass1 on the official tasks for sixteen layer contrasts: the eight ac￾cepted updates and eight ablation contrasts, over both domains. All eight weight contrasts lie above the di￾agonal, and all four rule contrasts on the base weights lie on or below it; three of the four rule contrasts on trained weights lie above it, as rules written in round 2 from the failures of M1 act on tasks the trained weights nearly solve. Loads Success System book loaded not loaded Reference agent – 18/40 (policy in prompt) H0-M0 30% 6/12 0/28 H0-M2 30% 7/12 0/28 H1-M0 40%…
Figure 5 of the paper
Figure 5. pass1 gain of each accepted update on V and on the official tasks. The dotted line marks equal gains. Airline n Ref. H0 M0 H2 M0 H0 M2 H2 M2 Find a reservation 27 64.8 52.8 60.2 60.2 70.4 Fallback condition 19 55.3 35.5 56.6 43.4 64.5 Multiple requests 17 39.7 17.6 38.2 25.0 50.0 Refused cancel 15 76.7 60.0 78.3 70.0 81.7 Persistent user 14 80.4 66.1 83.9 75.0 89.3 Change cabin 13 42.3 25.0 42.3 28.8 59.6 Direct flights 13 59.6 50.0 61.5 63.5 71.2 Fare difference 12 41.7 22.9 43.8 25.0 64.6 False user claim 11 72.7 52.3 75.0 61.4 79.5 Change flights 11 70.5 56.8 63.6 68.2 77.3…

Figures are the authors', reproduced from the paper.

References (56)
  1. [1] Barres, Victor and Dong, Honghua and Ray, Soham and Si, Xujie and Narasimhan, Karthik , booktitle = on Pith
  2. [2] Yao, Shunyu and Shinn, Noah and Razavi, Pedram and Narasimhan, Karthik , booktitle = on Pith
  3. [3] Shi, Quan and Zytek, Alexandra and Razavi, Pedram and Narasimhan, Karthik and Barres, Victor , booktitle = on Pith
  4. [4] Chen, Tingyang and Lu, Shuo and Zhao, Kang and Meng, Weicheng and Teng, Hanlin and Li, Tianhao and Li, Chao and Liu, Xule and Liang, Jian and Zhang, Zhizhong and Xie, Yuan and Qu, Heng and Shao, Kun and Luan, Jian , journal = on Pith
  5. [5] Xia, Peng and Chen, Jianwen and Wang, Hanyang and Liu, Jiaqi and Zeng, Kaide and Wang, Yu and Han, Siwei and Zhou, Yiyang and Zhao, Xujiang and Chen, Haifeng and Zheng, Zeyu and Xie, Cihang and Yao, Huaxiu , booktitle = on Pith
  6. [6] Harness-Aware Self-Evolving: Co-Evolving Model Weights, Harness, and Task Solutions , author =. arXiv preprint arXiv:2607.03935 , year = on PithPith's reviewarXiv
  7. [7] Xu, Chejian and Zhang, Shujian and Zhou, Wenxuan and Lambert, John and Chen, Beck and Li, Bo and Hard, Andrew and Mathews, Rajiv , journal = on Pith
  8. [8] Hebbar, Prannay and Manawat, Yogendra and Verboomen, Samuel and Ivanova, Alesia and Palanimalai, Selvam and Bhatia, Kunal and Baskaran, Vignesh , journal = on Pith
  9. [9] Chen, Mingju and Lv, Can and Zhang, Guibin and Chang, Heng and Zhou, Shiji , booktitle = on Pith
  10. [10] Chen, Zhengyu and Xiao, Teng and Zhu, Huaisheng and Yuan, Yige and Zhang, Luan and Wang, Jingang , journal = on Pith
  11. [11] Chen, Jixuan and Zhang, Jiaxin and Ye, Qinyuan and Pruksachatkun, Yada and Zhang, Haoxiang and Zhuo, Jingming and Zhang, Yifan and Dai, Yutong and Tan, Juntao and Peng, Xiangyu and Savarese, Silvio and Chen, Zeyuan and Qin, Lianhui and Wu, Chien-Sheng , journal = on Pith
  12. [12] Harness Evolution Hits a Ceiling: When Weight Training Should Begin , author =. arXiv preprint arXiv:2610.11655 , year = on PithPith's reviewarXiv
  13. [13] Co-Evolving Harnesses and Models: On-Policy Correction Helps Weaker Models Catch Up Where Imitation Fails , author =. arXiv preprint arXiv:2609.09134 , year = on PithPith's reviewarXiv
  14. [14] Fine-Tuning and Prompt Optimization: Two Great Steps that Work Better Together , author =. Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing , publisher = on Pith
  15. [15] Composing Policy Gradients and Prompt Optimization for Language Model Programs , author =. Proceedings of the ACM Conference on AI and Agentic Systems , publisher = on Pith
  16. [16] Wu, Jinyang and Yang, Shuo and Lu, Zhengxi and Zhang, Fan and Shen, Yuhao and Feng, Lang and Luo, Haoran and Lian, Zheng and Zhang, Shuai and Wen, Zhengqi and Tao, Jianhua , journal = on Pith
  17. [17] Lu, Zhengxi and Yao, Zhiyuan and Wu, Jinyang and Han, Chengcheng and Song, Xierui and Cai, Xunliang and Lu, Weiming and Xiao, Jun and Zhuang, Yueting and Shen, Yongliang , booktitle = on Pith
  18. [18] Rethinking the Evaluation of Harness Evolution for Agents , author =. arXiv preprint arXiv:2607.12227 , year = on PithPith's reviewarXiv
  19. [19] Lin, Weihong and Zhang, Xiangzheng and Sun, Lin , booktitle =. When Do Prompt-Side Agent Playbooks Transfer? on Pith
  20. [20] Ding, Wenxuan and Lu, Qin and Yu, Changlong and Li, Sha and Jin, Shuowei and Liu, Xin and Shi, Zhenyu and Durrett, Greg , journal = on Pith
show 36 more
  1. [21] Lin, Minhua and Wu, Juncheng and Wang, Zijun and Shi, Zhan and Sang, Yisi and He, Bing and Liu, Zewen and Wei, Tianxin and Wu, Zongyu and Zhang, Zhiwei and Wang, Dakuo and Zhang, Xiang and Dumoulin, Benoit and Xie, Cihang and Zhou, Yuyin and Wang, Suhang and Lu, Hanqing , book... on Pith
  2. [22] Du, Yuetian and Wang, Yucheng and Xu, He and Xu, Jiexu and Tan, Shanwen and Zhao, Bing and Yang, Boyu and Xu, Zhijie and Kong, Ming and Wei, Hu and Liu, Jie and Zhu, Qiang , journal = on Pith
  3. [23] Verify Smarter, Evolve Further: Efficient Harness Evolution through Behavior-Aware Verification , author =. arXiv preprint arXiv:2608.27311 , year = on PithPith's reviewarXiv
  4. [24] Gao, Jiaxuan and Chen, Jiaao and He, Chuyi and Xu, Shusheng and Jin, Di and Wu, Yi , journal =. From Self-Evolving Synthetic Data to Verifiable-Reward on Pith
  5. [25] The Path to Recursive Self-Improving Agents: Foundation, Framework, and Future Directions , author = on Pith
  6. [26] Yang, Yifan and Gong, Ziyang and Huang, Weiquan and Yang, Qihao and Zhou, Ziwei and Huang, Zisu and Li, Yan and Gao, Xuemei and Dai, Qi and Liu, Bei and Qiu, Kai and Yang, Yuqing and Chen, Dongdong and Yang, Xue and Luo, Chong , booktitle = on Pith
  7. [27] Li, Gaoyuan and Fan, Meihao and Liu, Yizhe and Zhang, Shaolei and Fan, Ju and Wang, Siyi and Hou, Jiaheng and Weng, Xudong and Tian, Honghan and Li, Zang , journal = on Pith
  8. [28] Li, Xiangyi and Liu, Yimin and Chen, Wenbo and Zheng, Shenghan and He, Yifeng and others , booktitle = on Pith
  9. [29] Zhao, Andrew and Huang, Daniel and Xu, Quentin and Lin, Matthieu and Liu, Yong-Jin and Huang, Gao , booktitle = on Pith
  10. [30] Agent Workflow Memory , author =. Proceedings of the 42nd International Conference on Machine Learning , year = on Pith
  11. [31] Prabhakar, Akshara and Liu, Zuxin and Zhu, Ming and Zhang, Jianguo and Awalgaonkar, Tulika and Wang, Shiyu and Liu, Zhiwei and Chen, Haolin and Hoang, Thai and Niebles, Juan Carlos and Heinecke, Shelby and Yao, Weiran and Wang, Huan and Savarese, Silvio and Xiong, Caiming , bo... on Pith
  12. [32] Zhao, Weikang and Wang, Xili and Ma, Chengdi and Kong, Lingbin and Yang, Zhaohua and Tuo, Mingxiang and Shi, Xiaowei and Zhai, Yitao and Cai, Xunliang , journal = on Pith
  13. [33] Reflexion: Language Agents with Verbal Reinforcement Learning , author =. Advances in Neural Information Processing Systems , volume = on Pith
  14. [34] Large Language Models as Optimizers , author =. The Twelfth International Conference on Learning Representations , year = on Pith
  15. [35] Agrawal, Lakshya A. and Tan, Shangyin and Soylu, Dilara and Ziems, Noah and Khare, Rishi and Opsahl-Ong, Krista and Singhvi, Arnav and Shandilya, Herumb and Ryan, Michael J. and Jiang, Meng and Potts, Christopher and Sen, Koushik and Dimakis, Alexandros G. and Stoica, Ion and ... on Pith
  16. [36] Agentic Context Engineering: Evolving Contexts for Self-Improving Language Models , author =. The Fourteenth International Conference on Learning Representations , year = on Pith
  17. [37] Yue, Yang and Chen, Zhiqi and Lu, Rui and Zhao, Andrew and Wang, Zhaokai and Yue, Yang and Song, Shiji and Huang, Gao , booktitle =. Does Reinforcement Learning Really Incentivize Reasoning Capacity in on Pith
  18. [38] Wu, Fang and Xuan, Weihao and Xing, Danny and Lu, Ximing and Liu, Mingjie and Dong, Yi and Harchaoui, Zaid and Choi, Yejin , journal =. The Invisible Leash: Why on Pith
  19. [39] Wang, Zihan and Wang, Kangrui and Wang, Qineng and Zhang, Pingyue and Li, Linjie and Yang, Zhengyuan and Jin, Xing and Yu, Kefan and Nguyen, Minh Nhat and Liu, Licheng and Gottlieb, Eli and Lu, Yiping and Cho, Kyunghyun and Wu, Jiajun and Fei-Fei, Li and Wang, Lijuan and Choi,... on Pith
  20. [40] Shao, Zhihong and Wang, Peiyi and Zhu, Qihao and Xu, Runxin and Song, Junxiao and Bi, Xiao and Zhang, Haowei and Zhang, Mingchuan and Li, Y. K. and Wu, Y. and Guo, Daya , journal = on Pith
  21. [41] Proximal Policy Optimization Algorithms , author =. arXiv preprint arXiv:1707.06347 , year = on PithPith's reviewarXiv
  22. [42] Ye, Deheng and Liu, Zhao and Sun, Mingfei and Shi, Bei and Zhao, Peilin and Wu, Hao and Yu, Hongsheng and Yang, Shaojie and Wu, Xipeng and Guo, Qingwei and Chen, Qiaobo and Yin, Yinyuting and Zhang, Hao and Shi, Tengfei and Wang, Liang and Fu, Qiang and Yang, Wei and Huang, La... on Pith
  23. [43] Li, Ziniu and Chen, Congliang and Yang, Tianyun and Ding, Tian and Sun, Ruoyu and Zhang, Ge and Huang, Wenhao and Luo, Zhi-Quan , booktitle =. Knapsack on Pith
  24. [44] One-Pass to Reason: Token Duplication and Block-Sparse Mask for Efficient Fine-Tuning on Multi-Turn Reasoning , author =. Findings of the Association for Computational Linguistics: IJCNLP-AACL 2025 , publisher = on Pith
  25. [45] Wang, Jinghui and Wang, Shaojie and Tang, Can and Cui, Yinghan and Chen, Xuxing and Zhang, Xiaojiang and Wang, Chao and Zhang, Haotian , booktitle =. Tree Training: Efficient on Pith
  26. [46] Li, Jiazheng and Lin, Hongzhou and Lu, Hong and Wen, Kaiyue and Yang, Zaiwen and Gao, Jiaxuan and Wu, Yi and Zhang, Jingzhao , booktitle = on Pith
  27. [47] Hu, Edward J. and Shen, Yelong and Wallis, Phillip and Allen-Zhu, Zeyuan and Li, Yuanzhi and Wang, Shean and Wang, Lu and Chen, Weizhu , booktitle = on Pith
  28. [48] Kwon, Woosuk and Li, Zhuohan and Zhuang, Siyuan and Sheng, Ying and Zheng, Lianmin and Yu, Cody Hao and Gonzalez, Joseph E. and Zhang, Hao and Stoica, Ion , booktitle =. Efficient Memory Management for Large Language Model Serving with on Pith
  29. [49] Adding Error Bars to Evals: A Statistical Approach to Language Model Evaluations , author =. arXiv preprint arXiv:2411.00640 , year = on PithPith's reviewarXiv
  30. [50] Some Guidelines and Guarantees for Common Random Numbers , author =. Management Science , volume = on Pith
  31. [51] Zhang, Barry and Lazuka, Keith and Murag, Mahesh , howpublished =. Equipping Agents for the Real World with. 2025 , note = on Pith
  32. [52] arXiv preprint arXiv:2607.02770 , year = on PithPith's reviewarXiv
  33. [53] Decoupled Weight Decay Regularization , author =. The Seventh International Conference on Learning Representations , year = on Pith
  34. [54] Du, Yufeng and Tian, Minyang and Ronanki, Srikanth and Rongali, Subendhu and Bodapati, Sravan Babu and Galstyan, Aram and Wells, Azton and Schwartz, Roy and Huerta, Eliu A. and Peng, Hao , booktitle =. Context Length Alone Hurts on Pith
  35. [55] Jaroslawicz, Daniel and Whiting, Brendan and Shah, Parth and Maamari, Karime , journal =. How Many Instructions Can on Pith
  36. [56] Balaji, Sumanth and Mishra, Piyush and Sachdeva, Aashraya and Agrawal, Suraj , booktitle =. Beyond on Pith

Challenges and discussion

Disagree with the review, or know something it missed? Leave a comment, or challenge this review formally: a challenge becomes a permanent entry in this paper's record, and its decision is published here.

No comments yet.

Sign in to comment