Pith. sign in

REVIEW 3 major objections 5 minor 32 references

Cross-model Transferability among Large Language Models on the Platonic Representations of Concepts

T0 review · 3 major / 5 minor · reviewed 2026-08-10 · deepseek-v4-flash

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

Pith's one-line read A single linear map transfers steering-vector control between different LLMs, including from a 0.5B model to a 7B model.

desk verdict Useful empirical study of linear SV transfer across LLMs; the weak-to-strong result is real, but the geometric claim outruns the behavioral evidence. read the letter →

arxiv 2501.02009 v2 pith:M3IX46UM submitted 2025-01-02 cs.CL cs.AI

Youcheng Huang , Chen Huang , Duanyu Feng , Wenqiang Lei , Jiancheng Lv This is my paper · ORCID
classification cs.CLcs.AI
keywords steeringvectorscross-modeltransferlinearrepresentationhypothesisLLMinterpretabilityweak-to-strongconceptalignmentactivationPlatonic
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

Different large language models, trained on different data, appear to represent the same concept as a direction in their hidden states, and these directions can be aligned from one model to another by a single learned linear map. The paper shows that a steering vector extracted from one model, multiplied by this map, can modulate another model's text generation toward the same concept, across eleven concepts including harmfulness, happiness, fear, fairness, and sycophancy. The same linear map works for many concepts, and steering vectors from a 0.5B model can control a 7B model, suggesting that LLM concept spaces share a roughly linear, common structure rather than being idiosyncratic to each model. If this is right, behavior-control vectors and concept probes could be built once and reused across model families and scales.

What carries the argument

The load-bearing object is the transformation matrix $T_D$, fitted by ordinary least squares on paired representations from a shared corpus, and applied to steering vectors through Eq. (3.2), $\bar{\lambda}_W^{m_t} = \bar{\lambda}_W^{m_s} T_D$. Steering vectors are concept directions in hidden-state space, defined either as the mean difference of contrastive text pairs (CAA) or as their first principal component (RepE). $T_D$ does the argument's work: if one map aligns many different concept directions, and does so for concepts not used to fit it, then the two models' concept geometries are related by a single linear change of coordinates. The paper supports that conclusion with the numerical similarity of $T$ matrices fitted on different concepts and with t-SNE visualizations that show the mapped representations behaving like rotated, scaled, or flipped versions of the source representations.

What would settle it

Take a target model whose self-modulation is known to work strongly, learn $T_D$ on a corpus disjoint from the test concept, and sweep $\beta$ and layers with a transferred vector; if the transferred vector's effect is indistinguishable from a random matrix of the same norm while the target's own vector still steers, then the OLS alignment is not carrying the concept and the claimed geometry is an artifact of activation addition.

Watch

Extended reading notes

Core claim

The paper sets out to show that the concept geometry of one LLM can be transplanted into another by a linear map: a steering vector $\bar{\lambda}_W^{m_s}$ for concept $W$ in source model $m_s$ becomes a steering vector for the same concept in target model $m_t$ as $\bar{\lambda}_W^{m_t} = \bar{\lambda}_W^{m_s} T_D$, where $T_D$ is obtained by ordinary least squares regression of the two models' representations of a shared corpus $D$. The claim is that this single matrix aligns concept directions across models well enough that the transformed vector changes target-model behavior toward the intended concept, that the same $T_D$ generalizes to concepts whose data played no part in fitting it, and that the alignment holds from a 0.5B model to a 7B model. On eleven benchmark concepts across Llama2, Qwen2, and Llama3.1, the paper reports that cross-model modulation is on par with each model's own self-modulation in most cases, and that removing or randomizing $T$ destroys the effect.

Load-bearing premise

The argument rests on the premise that adding a scaled, transformed vector to a target model's hidden states actually steers that model's behavior along the intended concept, so a behavior change after transfer counts as evidence that the two concept directions are aligned.

Editorial extensions

If this is right

  • If correct, a steering-vector library built once on one open model can be ported to other model families without retraining, because cross-model modulation matches self-modulation in most tested configurations.
  • Because one transformation matrix per model pair transfers many concepts, concept-specific tuning of the map is unnecessary; only the scaling factor $\beta$ needs per-concept adjustment.
  • Weak-to-strong transfer implies that concept diagnostics and behavior controls can be developed on a small, inexpensive model and applied to a larger model from the same family, without modifying the larger model's weights.
  • The linearity of the transfer reinforces the linear representation hypothesis: concept directions are geometric objects that survive a change of coordinates between models.

Reading between the lines

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

  • Editorial extension: because only paired text representations are needed to fit $T_D$, the same procedure should port other linear objects, such as classifier probes or semantic axes, from one model to another; testing on held-out probes would show whether the alignment is a property of geometry rather than of steering alone.
  • The paper leaves implicit that weak-to-strong transfer turns small open models into control surfaces for larger or closed ones, since a user needs the small model's weights and a few hundred paired sentences from the target, not the target's weights.
  • A safety consequence the authors do not emphasize: if harmful-direction transfer succeeds this easily, a refusal or safety intervention that works on one family may be bypassable in another family by transferring a small model's harmful vector through $T_D$; this is testable with an alignment evaluation like HarmBench.
  • The Platonic claim could be quantified by measuring how much of the target model's behavioral variance a single $T_D$ explains across layers, families, and scales; a systematic decrease in residual error with model scale would support the convergence-to-shared-reality reading.
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 / 5 minor

Summary. The paper proposes L-Cross Modulation, a method for transferring steering vectors (SVs) between LLMs by learning a linear map T via ordinary least squares between the hidden-state spaces of a source and a target model. The transformed SV is then added to the target's hidden states with a scaling factor β to modulate its behavior. The authors evaluate on eleven benchmark concepts across three LLMs and report three findings: (1) cross-model SV transfer effectively controls the target; (2) the same T generalizes across concepts; and (3) SVs from a weak model (Qwen2-0.5B) can steer stronger models (e.g., Qwen2-7B) to produce harmful content. Ablations show that a random transformation and direct cross-model addition without T are largely ineffective.

Significance. If the claims hold, this is a valuable contribution to interpretability: it provides evidence for a linear shared geometry of concept representations across LLMs, directly relevant to the Platonic Representation Hypothesis, and it offers a practical recipe for cross-model safety control. Strengths include the use of established CAA/RepE datasets and metrics, the out-of-sample generalization tests (cross-concept and weak-to-strong), and the explicit ablation of T. However, the central geometric claim is currently supported only by indirect behavioral evidence, and the statistical support is incomplete.

major comments (3)
  1. [§4.2, Eq. (3.2)] The paper's central claim, formalized in Eq. (3.2), is that the transformed source SV equals the target model's own concept direction. The evidence presented is behavioral: adding β times the transformed vector changes target outputs, and a random-T ablation fails. This underdetermines the geometric claim. Random T is not a matched control because it does not preserve the norm or orientation of the transformed vector; with layer ranges as broad as 9–27 (Table 5) and β hand-tuned per condition, concept, and source–target pair (Tables 5–6), many non-aligned directions could plausibly steer the target. The authors should directly measure the alignment between the transformed source SV and the target's own SV for the same concept, e.g., reporting cosine similarity or normalized distance, compared with baselines such as untransformed source SVs and norm-matched random T. Without such a check, the title and Section 6 conclusions about shared underlying structure go beyond what the behavioral experiments establish.
  2. [§4.1, Tables 1 and 3] All headline results are point estimates at manually selected β, with no confidence intervals. Since each evaluation uses 50 items (Appendix B), proportions such as the harmful-output rates in Table 1 have large binomial uncertainty; the aggregate failure counts (4/216 in §4.2, 17/216 in §4.3) similarly lack variance estimates. Without intervals, the reader cannot judge whether the gap between L-Cross Modulation and No Modulation is reliable, nor whether the claimed superiority over Self Modulation in 31/42 and 7/12 cases is within noise. Please report bootstrap confidence intervals or per-item scores for the main comparisons.
  3. [§4.2, Table 2] The random-T ablation is not a valid control for the role of T. A matrix with random entries produces a transformed vector with uncontrolled norm and orientation; its failure does not rule out the possibility that any sufficiently large, unaligned vector in the chosen layer range produces the observed behavior. Moreover, the w/o T variant is only tested on the same-dimensionality pair (Llama2 and Llama3.1), leaving the role of T for Qwen2 targets unablated. The authors should use matched random directions (same output norm as the learned T, or random orthogonal matrices) and include all three target models.
minor comments (5)
  1. [§3, Eq. (3.1)] The OLS problem has no intercept term. The authors should state whether representations were mean-centered; if not, adding a bias term or centering is a standard part of a linear alignment and is likely to improve the fit.
  2. [§4.3, Figure 3] t-SNE is a nonlinear dimensionality reduction that preserves local structure; using it to infer global linear relationships such as rotation, scaling, and flipping is not methodologically sound. The quantitative T-similarity measures in Table 4 are more appropriate and should be the primary evidence for the geometric claim.
  3. [§4.4, Figure 4] The weak-to-strong claim in the abstract is supported in the main text only by the harmfulness concept; the additional concepts in Appendix E are shown only as plots. Please include numeric results so that the generality of the claim can be verified.
  4. [General] The manuscript does not include a code availability statement. Since L-Cross Modulation is a simple OLS procedure, releasing the implementation would improve reproducibility.
  5. [Throughout] Typos and wording issues: 'Crorss Modulation' in Table 2, 'adoptd' in Section 2, 'services as a foundation' in the Introduction, and 'L-Corss Modulation' in Appendix D should be corrected.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: T is fitted to representations, and behavioral tests on held-out prompts are out-of-sample.

full rationale

No circular step is present in the derivation chain. The transformation matrix T is learned by ordinary least squares on sentence-level representations from a shared corpus (Eq. 3.1) and then applied to a source model's steering vector to form a predicted target steering vector (Eq. 3.2). The evaluation is behavioral and out-of-sample: CAA/RepE held-out test prompts and external classifiers (HarmBench, GPT-4o-mini) are used, so the target model's responses are not part of the fitted quantities. The target model's own steering vector is never used to fit T, and T is not fitted to the evaluation outcome; the manually tuned scalar beta (acknowledged in the Limitations section) affects modulation strength but does not make the geometric alignment claim true by construction. RQ2 (concept-unrelated T) and RQ3 (weak-to-strong) are genuine generalization tests relative to the fitted T. Self-citations (Huang et al. 2024; Feng et al. 2024) appear only in Related Work and are not load-bearing. Concerns about the random-T ablation and beta selection are evidence-quality or underdetermination issues, not circularity.

Assumptions & free parameters 3 free parameters · 4 assumptions · 0 invented entities

The central claim rests on the standard steering-vector mechanism inherited from CAA/RepE, on the validity of an OLS representation alignment, on hand-tuned beta values, and on concept-specific evaluation metrics. There are no invented entities. The transformation matrix T and the beta values are the two fitted objects; T is estimated from corpus representations, beta is tuned on the behavioral outcome. These are the main things a re-implementation would need to control.

free parameters (3)
  • beta (modulation strength) = Per (mt, ms, concept) values; e.g., Harm Llama2: self 4.0, Qwen2 source 8.0, Llama3.1 source 1.5 (Table 5)
    Manually tuned modulation strength; no automatic optimization; all headline numbers are at chosen beta, and Appendix D shows strong dependence on beta.
  • T_D transformation matrix = d_ms x d_mt matrix solved by OLS (Eq. 3.1)
    The map is fit to source-target sentence representations, not derived; the finding that it transfers is empirical. The learned matrices are not released, so the result cannot be independently checked.
  • transformer layer indices for SV extraction and injection = e.g., CAA: Llama2 layer 13, Qwen2 layer 18, Llama3.1 layer 13; RepE ranges in Table 5
    The paper states in Section 2 that the choice of layers is a hyperparameter; these choices affect which representations are aligned and added.
assumptions (4)
  • domain assumption Adding a scaled steering vector to hidden states modulates behavior
    Inherited from CAA/RepE; Section 2, "Modulating LLM via Scaled SVs". The paper's central measurement of transfer always uses this operation, so a failure or non-linearity here would invalidate the conclusion.
  • domain assumption Steering vectors are linear directions in representation space and are comparable across models
    Motivated by Park et al. 2024 and Jiang et al. 2024; Section 3 assumes a linear coordinate change T exists and that SVs are vectors in that space.
  • domain assumption The OLS fit on last-token representations of a shared corpus aligns the semantic coordinate systems
    Equation 3.1 treats sentence representations as paired observations; if these representations are not semantically aligned, T may be an arbitrary map that happens to work for the tested corpus.
  • domain assumption The evaluation metrics measure the target concepts
    Section 4.1 relies on the HarmBench classifier, GPT-4o-mini relevance scores, and binary choice probabilities; if these scores are noisy or miss the concept, the reported transfer percentages are not meaningful.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Cross-model Transferability among Large Language Models on the Platonic Representations of Concepts." pith.science (2026). https://pith.science/paper/M3IX46UM

@misc{pith2026250102009,
  author       = {Pith},
  title        = {Pith review of: Cross-model Transferability among Large Language Models on the Platonic Representations of Concepts},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/M3IX46UM}},
  note         = {Machine review of arXiv:2501.02009}
}
read the original abstract

Understanding the inner workings of Large Language Models (LLMs) is a critical research frontier. Prior research has shown that a single LLM's concept representations can be captured as steering vectors (SVs), enabling the control of LLM behavior (e.g., towards generating harmful content). Our work takes a novel approach by exploring the intricate relationships between concept representations across different LLMs, drawing an intriguing parallel to Plato's Allegory of the Cave. In particular, we introduce a linear transformation method to bridge these representations and present three key findings: 1) Concept representations across different LLMs can be effectively aligned using simple linear transformations, enabling efficient cross-model transfer and behavioral control via SVs. 2) This linear transformation generalizes across concepts, facilitating alignment and control of SVs representing different concepts across LLMs. 3) A weak-to-strong transferability exists between LLM concept representations, whereby SVs extracted from smaller LLMs can effectively control the behavior of larger LLMs.

Figures

Figures reproduced from arXiv: 2501.02009 by the authors.

Figure 1
Figure 1. In Plato’s Allegory of the Cave, prisoners [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. L-Cross Modulation uses linear transforma [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗
Figure 3
Figure 3. T-SNE visualization of representations {λδ}. The green, purple, and yellow dots correspond to the concepts of AIC., CORR., and HALLU., respectively. ms Llama2 Qwen2 Llama3.1 mt Qwen2 Llama3.1 Llama2 SSIM ↑ 0.94 0.95 0.87 ME ↓ 1.14 0.07 1.76 ∥∆∥F ↓ 573.65 27.63 572.17 Random T SSIM ↑ 0.13 0.05 0.08 ME ↓ 54.03 55.75 51.34 ∥∆∥F ↓ 3855.13 3831.53 4117.45 [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figures from the paper (6 more)
Figure 4
Figure 4. Figure 4: In Self-Modulation, varying β results in a max￾imum 54.0% harmful outputs of Qwen2 0.5B. However, the harmful SV derived from Qwen2 0.5B effectively modulate Qwen2 7B to generate 88.0% harmful outputs. How effective are the SVs derived from a weak LLM on modulating str…
Figure 5
Figure 5. Figure 5: Weak-to-Strong L-Cross Modulation where SVs are extracted from a weak model of Qwen2-0.5B. We conduct additional weak-to-strong L-Cross Modulation using the seven concepts in CAA. See [PITH_FULL_IMAGE:figures/full_fig_p017_5.png]
Figure 6
Figure 6. Figure 6: Weak-to-Strong L-Cross Modulation results [PITH_FULL_IMAGE:figures/full_fig_p017_6.png]
Figure 7
Figure 7. Figure 7: The probabilities that LLMs assign to the [PITH_FULL_IMAGE:figures/full_fig_p018_7.png]
Figure 9
Figure 9. Figure 9: The probabilities that LLMs assign to the [PITH_FULL_IMAGE:figures/full_fig_p018_9.png]
Figure 12
Figure 12. Figure 12: The evaluation metrics of concepts HARM and FAIR in the setting of Self Modulation and L-Cross Modulation (where concept-unrelated T is utilized (cf. see Section 4.3), with different modulation strengths β [PITH_FULL_IMAGE:figures/full_fig_p019_12.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

32 extracted references · 8 canonical work pages

  1. [1]

    Rishi Bommasani, Drew A Hudson, Ehsan Adeli, Russ Altman, Simran Arora, Sydney von Arx, Michael S Bernstein, Jeannette Bohg, Antoine Bosselut, Emma Brunskill, et al. 2021. On the opportunities and risks of foundation models. arXiv preprint arXiv:2108.07258

    arXiv 2021

  2. [2]

    Collin Burns, Haotian Ye, Dan Klein, and Jacob Steinhardt. 2023. https://openreview.net/forum?id=ETKGuby0hcs Discovering latent knowledge in language models without supervision . In The Eleventh International Conference on Learning Representations

    2023

  3. [3]

    Xinyun Chen, Renat Aksitov, Uri Alon, Jie Ren, Kefan Xiao, Pengcheng Yin, Sushant Prakash, Charles Sutton, Xuezhi Wang, and Denny Zhou. 2023. Universal self-consistency for large language model generation. arXiv preprint arXiv:2311.17311

    arXiv 2023

  4. [4]

    Abhimanyu Dubey, Abhinav Jauhri, Abhinav Pandey, Abhishek Kadian, Ahmad Al-Dahle, Aiesha Letman, Akhil Mathur, Alan Schelten, Amy Yang, Angela Fan, Anirudh Goyal, Anthony Hartshorn, Aobo Yang, Archi Mitra, Archie Sravankumar, Artem Korenev, Arthur Hinsvark, Arun Rao, Aston Zhang, Aurelien Rodriguez, Austen Gregerson, Ava Spataru, Baptiste Rozi \`e re, Bet...

    arXiv 2024

  5. [5]

    Duanyu Feng, Bowen Qin, Chen Huang, Youcheng Huang, Zheng Zhang, and Wenqiang Lei. 2024. Legend: Leveraging representation engineering to annotate safety margin for preference datasets. arXiv preprint arXiv:2406.08124

    arXiv 2024

  6. [6]

    Youcheng Huang, Fengbin Zhu, Jingkun Tang, Pan Zhou, Wenqiang Lei, Jiancheng Lv, and Tat-Seng Chua. 2024. Effective and efficient adversarial detection for vision-language models via a single vector. arXiv preprint arXiv:2410.22888

    arXiv 2024

  7. [7]

    Minyoung Huh, Brian Cheung, Tongzhou Wang, and Phillip Isola. 2024. https://openreview.net/forum?id=BH8TYy0r6u Position: The platonic representation hypothesis . In Forty-first International Conference on Machine Learning

    work page 2024

  8. [8]

    Yibo Jiang, Goutham Rajendran, Pradeep Kumar Ravikumar, Bryon Aragam, and Victor Veitch. 2024. https://openreview.net/forum?id=otuTw4Mghk On the origins of linear representations in large language models . In Forty-first International Conference on Machine Learning

    work page 2024

Show all 32 references
  1. [9]

    Yuping Lin, Pengfei He, Han Xu, Yue Xing, Makoto Yamada, Hui Liu, and Jiliang Tang. 2024. https://doi.org/10.18653/v1/2024.emnlp-main.401 Towards understanding jailbreak attacks in LLM s: A representation space analysis . In Proceedings of the 2024 Conference on Empirical Meth...

    2024 doi

  2. [10]

    Wenhao Liu, Xiaohua Wang, Muling Wu, Tianlong Li, Changze Lv, Zixuan Ling, Zhu JianHao, Cenyuan Zhang, Xiaoqing Zheng, and Xuanjing Huang. 2024. https://doi.org/10.18653/v1/2024.acl-long.572 Aligning large language models with human preferences through representation engineeri...

    2024 doi

  3. [11]

    Mantas Mazeika, Long Phan, Xuwang Yin, Andy Zou, Zifan Wang, Norman Mu, Elham Sakhaee, Nathaniel Li, Steven Basart, Bo Li, David Forsyth, and Dan Hendrycks. 2024. https://openreview.net/forum?id=f3TUipYU3U Harmbench: A standardized evaluation framework for automated red teamin...

    2024

  4. [12]

    Neel Nanda, Andrew Lee, and Martin Wattenberg. 2023. https://doi.org/10.18653/v1/2023.blackboxnlp-1.2 Emergent linear representations in world models of self-supervised sequence models . In Proceedings of the 6th BlackboxNLP Workshop: Analyzing and Interpreting Neural Networks...

    2023 doi

  5. [13]

    Kiho Park, Yo Joong Choe, and Victor Veitch. 2024. https://openreview.net/forum?id=UGpGkLzwpP The linear representation hypothesis and the geometry of large language models . In Forty-first International Conference on Machine Learning

    2024

  6. [14]

    Plato. c. 375 BC. Republic

  7. [15]

    Nina Rimsky, Nick Gabrieli, Julian Schulz, Meg Tong, Evan Hubinger, and Alexander Turner. 2024. https://doi.org/10.18653/v1/2024.acl-long.828 Steering llama 2 via contrastive activation addition . In Proceedings of the 62nd Annual Meeting of the Association for Computational L...

    2024 doi

  8. [16]

    Dale Schuurmans, Hanjun Dai, and Francesco Zanini. 2024. Autoregressive large language models are computationally universal. arXiv preprint arXiv:2410.03170

    2024 arXiv

  9. [17]

    Shuqian Sheng, Yi Xu, Tianhang Zhang, Zanwei Shen, Luoyi Fu, Jiaxin Ding, Lei Zhou, Xiaoying Gan, Xinbing Wang, and Chenghu Zhou. 2024. https://doi.org/10.18653/v1/2024.emnlp-main.398 R ep E val: Effective text evaluation with LLM representation . In Proceedings of the 2024 Co...

    2024 doi

  10. [18]

    Yusheng Su, Xiaozhi Wang, Yujia Qin, Chi-Min Chan, Yankai Lin, Huadong Wang, Kaiyue Wen, Zhiyuan Liu, Peng Li, Juanzi Li, Lei Hou, Maosong Sun, and Jie Zhou. 2022. https://doi.org/10.18653/v1/2022.naacl-main.290 On transferability of prompt tuning for natural language processi...

    2022 doi

  11. [19]

    Nishant Subramani, Nivedita Suresh, and Matthew Peters. 2022. https://doi.org/10.18653/v1/2022.findings-acl.48 Extracting latent steering vectors from pretrained language models . In Findings of the Association for Computational Linguistics: ACL 2022, pages 566--581, Dublin, I...

    2022 doi

  12. [20]

    Curt Tigges, Oskar John Hollinsworth, Atticus Geiger, and Neel Nanda. 2023. Linear representations of sentiment in large language models. arXiv preprint arXiv:2310.15154

    2023 arXiv

  13. [21]

    Stone, Peter Albert, Amjad Almahairi, Yasmine Babaei, Nikolay Bashlykov, Soumya Batra, Prajjwal Bhargava, Shruti Bhosale, Daniel M

    Hugo Touvron, Louis Martin, Kevin R. Stone, Peter Albert, Amjad Almahairi, Yasmine Babaei, Nikolay Bashlykov, Soumya Batra, Prajjwal Bhargava, Shruti Bhosale, Daniel M. Bikel, Lukas Blecher, Cristian Cant \'o n Ferrer, Moya Chen, Guillem Cucurull, David Esiobu, Jude Fernandes,...

    2023 arXiv

  14. [22]

    Alexander Matt Turner, Lisa Thiergart, Gavin Leech, David Udell, Juan J Vazquez, Ulisse Mini, and Monte MacDiarmid. 2023. Activation addition: Steering language models without optimization. arXiv preprint arXiv:2308.10248

    2023 arXiv

  15. [23]

    Laurens van der Maaten and Geoffrey E. Hinton. 2008. https://api.semanticscholar.org/CorpusID:5855042 Visualizing data using t-sne . Journal of Machine Learning Research, 9:2579--2605

    2008

  16. [24]

    Sheikh, and Eero P

    Zhou Wang, Alan Conrad Bovik, Hamid R. Sheikh, and Eero P. Simoncelli. 2004. https://api.semanticscholar.org/CorpusID:207761262 Image quality assessment: from error visibility to structural similarity . IEEE Transactions on Image Processing, 13:600--612

    2004

  17. [25]

    Zihao Wang, Lin Gui, Jeffrey Negrea, and Victor Veitch. 2023. https://openreview.net/forum?id=SGlrCuwdsB Concept algebra for (score-based) text-controlled generative models . In Thirty-seventh Conference on Neural Information Processing Systems

    2023

  18. [26]

    Chunqiu Steven Xia, Matteo Paltenghi, Jia Le Tian, Michael Pradel, and Lingming Zhang. 2024. Fuzz4all: Universal fuzzing with large language models. In Proceedings of the IEEE/ACM 46th International Conference on Software Engineering, pages 1--13

    2024

  19. [27]

    An Yang, Baosong Yang, Binyuan Hui, Bo Zheng, Bowen Yu, Chang Zhou, Chengpeng Li, Chengyuan Li, Dayiheng Liu, Fei Huang, Guanting Dong, Haoran Wei, Huan Lin, Jialong Tang, Jialin Wang, Jian Yang, Jianhong Tu, Jianwei Zhang, Jianxin Ma, Jin Xu, Jingren Zhou, Jinze Bai, Jinzheng...

    2024 arXiv

  20. [28]

    Yichi Zhang, Yinpeng Dong, Siyuan Zhang, Tianzan Min, Hang Su, and Jun Zhu. 2024. https://api.semanticscholar.org/CorpusID:269187585 Exploring the transferability of visual prompting for multimodal large language models . 2024 IEEE/CVF Conference on Computer Vision and Pattern...

    2024

  21. [29]

    Andy Zou, Long Phan, Sarah Chen, James Campbell, Phillip Guo, Richard Ren, Alexander Pan, Xuwang Yin, Mantas Mazeika, Ann-Kathrin Dombrowski, et al. 2023 a . Representation engineering: A top-down approach to ai transparency. arXiv preprint arXiv:2310.01405

    2023 arXiv

  22. [31]

    Zico Kolter, and Matt Fredrikson

    Andy Zou, Zifan Wang, J. Zico Kolter, and Matt Fredrikson. 2023 c . https://api.semanticscholar.org/CorpusID:260202961 Universal and transferable adversarial attacks on aligned language models . ArXiv, abs/2307.15043

    2023 arXiv

  23. [32]

    online" 'onlinestring :=

    ENTRY address archivePrefix author booktitle chapter edition editor eid eprint eprinttype howpublished institution journal key month note number organization pages publisher school series title type volume year doi pubmed url lastchecked label extra.label sort.label short.list...

  24. [33]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 gl...

Pith tools

Reviewed August 10, 2026 · model on record in the stance chip above.