- Rating
- 3(+2)
- Confidence
- 4
- Informativeness
- 1
- Presentation
- 1
- Soundness
- 1
Read review
Paper summary
The paper propose new mechanistic interpretability method called V-Lens which consists of V-Read and V-Write. V-Read learns the affine transformation from intermediate hidden state to the last hidden state while V-Write calculate the gradient of the defined score to measure how each patch contribute to answer change. The authors tested this new interpretability methods on their own benchmark named SIMONSAYS which covers authority, evaluation awareness, and sandbagging. The qualitative results show that V-Write can accurately locate the target object of the query in the image while localizing the object better than KernelSHAP. More experiments include AUROC comparing to previous lens, reading where the answer arises within LLM layers, visual password experiments. The paper argue that V-Lens can distinguish what model perceives and what model actually generates.
Critical strengths
- The authors provided the detailed formulation of V-Lens.
- The paper curated their own benchmark SIMONSAYS to test V-Lens.
- Extensive experiments are conducted including various baselines of V-Read and V-Write.
Critical weaknesses
-
I am significantly concerned about the format and overall quality of this paper. Although the authors explicitly mentioned their usage of AI on figures and writings, I still see this paper are far below the high standard of ICLR conference. (1) all figures are presented in low-resolution (some of them are clearly generated by AI without sufficient care) (2) \vspace are excessively used around the paper (L375-376, L304-305, L235-237, L209-211, L150-152, L111-113, L29-L31 etc). Without vspace command, paper will be clearly over 9 pages and this can be the reason for desk-rejection. (3) All texts are mainly describing the experiment results with all detailed numbers without conveying the key messages (or inference) for each results (typical AI-generated text). All of this significantly reduce the readability of the paper.
-
SIMONSAYS benchmark is curated with the help of Claude (figure 6) but the authors does not describe in the main paper how they validate the quality of the dataset which is critical considering the character of the experiment they do.
-
Limited novelty of the proposed V-Lens. V-Read formulation is exactly same as tuned lens as the author admitted in L142-145. Moreover, localization ability of logit lens are already widely studied [1-3] which is neither compared to V-Write or discussed extensively. Why V-Write should be better than all previous logit-lens-based localization? It should be mentioned in Sec3 and Sec 7.
-
One of the main finding of the paper is the "Cues the models notice but do not act on. (L376-429)" but this is not something new and overlapped with previous works [4].
-
All qualitative results are comparing V-Lens only to naive logit lens not to recent lens (tuned lens or J-lens). It is already known that logit-lens results are not reliable on early-mid layers so the results are not surprising to me.
[1] Neo, C., Ong, L., Torr, P., Geva, M., Krueger, D., & Barez, F. (2025, May). Towards interpreting visual information processing in vision-language models. In International Conference on Learning Representations (Vol. 2025, pp. 57172-57189).
[2] Jiang, N., Kachinthaya, A., Petryk, S., & Gandelsman, Y. (2025, May). Interpreting and editing vision-language representations to mitigate hallucinations. In International Conference on Learning Representations (Vol. 2025, pp. 63582-63605).
[3] Krojer, B., Nayak, S., Mañas, O., Adlakha, V., Elliott, D., Reddy, S., & Mosbach, M. (2026). Latentlens: Revealing highly interpretable visual tokens in llms. arXiv preprint arXiv:2602.00462.
[4] Nooralahzadeh, F., Rohanian, O., Zhang, Y., Fürst, J., & Stockinger, K. (2026). Arbitration Failure, Not Perceptual Blindness: How Vision-Language Models Resolve Visual-Linguistic Conflicts. arXiv preprint arXiv:2604.09364.
Questions for authors
- Why V-Read is better than Tuned lens in Table 2 although their formulation are exactly the same?
- What is the intuition behind the formulation of V-Write (based on the gradient of the score)? Why not using simple tuned-lens localization?
5 comments
Response to reviewer ngeH (2/3)
W3. Novelty and comparison with lens-based localisation (continued)
Table R1 (Appendix Table 14). Lens localisers vs. V-Write. Answer removed is the fall in the answer score after attention-masking the top 20% of image tokens. Agreement is Spearman correlation with measured patching effects over 16 regions. Own − other compares removal using the target-question map with that using a map for another object in the same photograph (251 photographs).
| Localiser | hit@1 | Mask AUROC | Answer removed | Agreement | Own − other |
|---|---|---|---|---|---|
| V-Write | 0.50 | 0.72 | 3.84 | 0.47 | 3.32 |
| Logit-lens object, max over blocks | 0.38 | 0.69 | 1.66 | 0.03 | 1.75 |
| Tuned-lens object | 0.23 | 0.53 | 1.01 | −0.04 | 0.76 |
| V-Read object | 0.21 | 0.57 | 0.62 | −0.05 | 0.63 |
| LatentLens | 0.11 | 0.52 | 0.68 | −0.05 | 0.35 |
| Random | 0.13 | 0.50 | 0.04 | 0.00 | – |
Table R2 (Appendix Table 15). Object absent, model answers "no" (40 COCO photographs). The metric is the answer-score decrease after masking the top 20% of tokens.
| Localiser | Answer removed | V-Write − localiser [95% CI] |
|---|---|---|
| V-Write | 1.23 | – |
| LatentLens | 0.27 | 0.97 [0.81, 1.14] |
| Tuned-lens object | 0.20 | 1.04 [0.86, 1.21] |
| Logit-lens object, max over blocks | −0.05 | 1.28 [1.06, 1.51] |
Table R3 (Appendix Table 16). Comparison with token-level KernelSHAP (50 COCO photographs; 10,000 coalitions). KernelSHAP is an estimated reference; Fall@20% is the score decrease after masking the top 20% of tokens.
| Method | Spearman with reference [95% CI] | hit@1 | Mask AUROC | Fall@20% |
|---|---|---|---|---|
| Token-level KernelSHAP (reference) | 1.00 | 0.54 | 0.70 | 3.76 |
| V-Write | 0.71 [0.67, 0.74] | 0.54 | 0.75 | 3.87 |
| KernelSHAP, 16 groups | 0.27 [0.24, 0.29] | 0.26 | 0.82 | 2.30 |
| IG from zero | 0.18 | 0.42 | 0.58 | 2.08 |
| Logit-lens localiser | 0.15 | 0.30 | 0.74 | 2.13 |
| Table R4. Localiser efficiency on the same machine, Qwen3-VL-8B-Instruct. f and b denote forward and backward passes. |
| Localiser | Passes per picture | Seconds per picture | One-time fitting |
|---|---|---|---|
| V-Write | 8 f+b | 0.4 | none |
| Logit-lens object | 1 f | 0.4 | none |
| Tuned-lens object | 1 f | 0.4 | tuned lens, 7.9 h |
| V-Read object | 1 f | 0.4 | V-Read, 6.9 min |
| LatentLens | 1 f + nearest-neighbour search | 0.4 | caption store, 0.6 min |
| KernelSHAP, 16 groups | 2,080 f | 49 | none |
| KernelSHAP, token level | 10,000 f | 243 | none |
We also kindly invite the reviewer to consider our theoretical results in Section 4 (proofs in Appendix H), which we hope further clarify the method.
W4. Overlap with [4]
We thank the reviewer for pointing us to [4]. We agree that a dissociation between internally encoded visual information and the model's final answer has been studied previously. We now cite and discuss [4] in Section 7.2 and Appendix F, and we revised our wording to make clear that we do not claim this general finding as new.
In a related text–image conflict setting on SimonSays, we find that the visual fact remains highly decodable by a probe (AUROC 0.98 on both Qwen models), while answers follow the picture on only 0.53 and 0.49 of conflict records, respectively. Our additional analyses examine the answer readout and the effects of restoring states at different token positions. On Qwen3-VL-8B-Instruct, V-Read's AUROC for the visual fact decreases from 0.95 at mid-depth to 0.55 at the output, despite high probe decodability (Appendix Table 32). At the tested depths, restoring the answer-position state recovers more of the measured conflicting-sentence effect than restoring image-token states. These results are consistent with visual information remaining decodable while the final answer increasingly reflects the conflicting cue.
We also study related cue–behaviour gaps in additional settings, including authority attribution, evaluation cues and sandbagging instructions, and the carrier finding replicates on gemma-4-31B-it, a larger model from another family (probe 0.82–0.86 over six seeds, answer 0.57). We position these as extensions to different cue types, alongside the prior findings of [4].
Table R5 (Appendix Table 32). Text–image conflict on nuScenes (900 records per model). Follows picture gives the share without and with a conflicting sentence. Probe values are AUROC from mid-depth to output; V-Read gives AUROC on conflict records. The last two rows give the share of the cue's effect recovered by restoring image-token states at depth 0.25 or the answer-position state at depth 0.75. Arrows indicate mid-depth and output; ranges summarise the evaluated depths. The recovery values apply to the specified interventions and layers.
| Measurement | Qwen3-VL-8B | Qwen3.5-27B | gemma-4-31B-it |
|---|---|---|---|
| Follows picture, no cue / conflict | 0.93 / 0.53 | 0.91 / 0.49 | 0.97 / 0.89 |
| Visual-fact probe AUROC | 0.98 | 0.98–0.99 | 0.99–1.00 |
| V-Read AUROC across depth | 0.95 → 0.55 | – | 0.90–0.98 |
| Sentence: image tokens / answer position | 0.00 / 0.73 | −0.01 / 0.40 | – |
| Sign carrier: image tokens / answer position | 0.99 / 0.79 | 1.00 / 0.47 | – |
Response to Reviewer ngeH (3/3)
W5. Qualitative comparisons
We agree that the logit lens is a weak early- and mid-depth reference. Figure 3 previously showed only the logit lens, and its redrawn summary panel and caption now include the tuned lens. Table R6 extends the same test to every reader. Other comparisons with stronger readers include:
- Figure 31: before half depth, V-Read reads 0.56–0.67, while the tuned, Jacobian and logit lenses are near or below chance.
- Figure 30: hallucinated vs. true answers, AUROC 0.95 for V-Read vs. 0.79 for the Jacobian lens at half depth.
- Figure 29: object names read from image patches, where V-Read peaks at 33% at one-fifth depth.
- Figure 28: fitting takes 412 s for V-Read, 28,302 s for the tuned lens and 9,313 s for the Jacobian lens.
- Table 2 and Appendix Tables 22, 23, 24 and 25: within-question, matched-time and early-depth comparisons with the tuned lens, Jacobian lens and Patchscopes. Figure 2 compares V-Write's hit@1 with attribution baselines.
Table R6. Figure 3's test with every reader: the share of matched pairs where V-Read / the other reader shows the cue's effect first (sign test). gemma-4-31B-it uses 19 pairs and a tuned lens trained on 31× V-Read's tokens; –: not fitted.
| Model (pairs) | vs tuned lens | vs Jacobian lens | vs logit lens | vs Patchscopes |
|---|---|---|---|---|
| Qwen3-VL-8B-Instruct (43) | 49% / 21% (p = 0.04) | 44% / 23% | 40% / 23% | 12% / 51% |
| Qwen3.5-27B (43) | 26% / 26% | 19% / 14% | 28% / 19% | 12% / 65% |
| gemma-4-31B-it (19) | 47% / 5% (p = 0.02) | – | 53% / 26% | 16% / 47% |
V-Read shows the cue's effect earlier than the tuned lens on Qwen3-VL-8B-Instruct and gemma-4-31B-it, and is level with it on Qwen3.5-27B. Patchscopes shows it earlier on both Qwen models, but it re-runs the later blocks for every reading. Figure 3 uses the 43 pairs read by the paper's held-out V-Read maps; with refitted maps covering all 53 Qwen3-VL-8B-Instruct pairs, the lead over the tuned lens is smaller and not significant (36% vs. 21%, p = 0.20). We will add Figure 3 examples for every reader in the appendix.
Q1. Why can V-Read differ from the tuned lens?
V-Read and the tuned lens use the same family of affine maps but differ in three ways:
- V-Read regresses onto the final residual state, while the tuned lens matches the final token distribution by KL.
- V-Read is solved exactly by ridge regression in one pass, while the tuned lens is trained by SGD from the identity.
- V-Read adds a per-block output gain chosen by KL.
Table R7 changes one of these differences at a time.
Table R7 (Appendix Table 3). V-Read and tuned-lens fitting variants (Qwen3-VL-8B-Instruct; AUROC averaged over the mid-depth window used in Table 2; 900 fitting prompts and 16 blocks). Lower held-out KL is better.
| Variant | AUROC | Held-out KL |
|---|---|---|
| Tuned lens, 1.1M tokens | 0.829 | 3.06 |
| Tuned lens, full 65.7M-token schedule | 0.828 | – |
| V-Read objective, solved by SGD | 0.831 | 12.17 |
| V-Read, closed form | 0.892 | 4.13 |
| Tuned lens initialised at V-Read | 0.893 | 2.89 |
Solving V-Read's own objective by SGD loses the gain, while starting the tuned lens from V-Read's solution keeps it, and training the tuned lens 60× longer from the identity does not close the gap. The difference therefore comes from the exact solve, since SGD from the identity does not reach the best map in the family. The initialised tuned lens has a lower held-out KL than V-Read, so V-Read's advantage lies in reading the answer rather than in matching the full output distribution.
Q2. Why gradient-based V-Write instead of tuned-lens localisation?
A tuned-lens reading at an image-token position shows which vocabulary concepts can be read from that token. V-Write instead targets a specific decision of the model. It measures how the answer score changes with the intermediate image-token states as the remaining layers run, using integrated gradients from the picture's mean state. A patch can be readable without influencing the answer, so we test both kinds of method by intervening on image-token states and measuring the change in the answer score.
On the same 888 COCO photographs (Table R1), V-Write finds the object more often than the best lens localiser (hit@1 0.50 vs. 0.38), agrees much better with measured region patching (Spearman 0.47 vs. 0.03), and its top 20% of tokens remove more of the answer (3.84 vs. 1.66). An object-readable patch is therefore not necessarily the patch the answer depends on. We added this comparison to Section 7.1 and Appendix D.1.
We thank the reviewer again for the detailed review, which led to substantial changes in the paper, including full-resolution figures, a validation summary in Section 5, direct comparisons with lens-based localisers and the tuned lens, and a discussion of [4]. If our responses address the reviewer's concerns, we would be grateful if they would consider revisiting their assessment, and we are happy to answer any further questions.
Response to Reviewer ngeH (1/3)
We thank the reviewer for their detailed feedback. Below we address the concerns and questions with clarifications, manuscript revisions and additional experiments. Changes in the revised PDF are shown in blue. References [1–4] follow the numbering in the review, and table and appendix numbers refer to the revised paper.
W1. Presentation and formatting
Thank you for pointing out these presentation issues. We had to compress the PDF at the last moment to meet the file-size limit, which degraded the figures, and we could not fix it due to time constraints. We again apologise for the excessive negative
\vspace. We have removed those commands, restored full-resolution figures and redrawn Figures 3 and 4 for clarity. Figure 3 now also compares V-Read with the tuned lens; Figure 4 labels training and held-out watermark placements. We welcome further suggestions. All plots are produced by our code directly from the result files or created in Canva, no figure is completely AI-generated, the low quality was because of compression which we have fixed by converting eveything to pdf instead of pngWe agree that parts of the results section listed numbers without saying what they mean; to convey the key messages, we revised Section 7.1 to clarify V-Write's question-specific attributions, the effect of removing top-ranked tokens, and V-Read's intermediate readings and hallucination results. Section 7.2 now clarifies representing versus following a sandbagging instruction and the visual-password results. We also clarified the cue–behaviour direction comparison: on Qwen3.5-27B, authority and test-look directions are no closer to their behaviours than shuffled controls (0.26 and 0.33 vs. 0.32 and 0.37), while a harmful prompt lies along refusing (0.81 vs. 0.47; Appendix Figure 38 and Table 27).
W2. SimonSays validation
We agree that the validation procedures should be clearer in the main paper more than they are now. Section 5 now summarises them, and Appendix B.11 gives details. Claude helped draft prompts, reference answers and scene descriptions; validation included human inspection and an independent model.
The full validation results are reported in Appendix Tables 11–13.
W3. Novelty and comparison with lens-based localisation
We agree that V-Read uses the same family of affine maps as the tuned lens, and we do not claim the family as new. What we add is an exact closed-form fit of that map, which reads the answer better than the tuned lens and fits in minutes instead of hours (Q1), and its use at the VLM answer position with matched image edits. Prior visual-token localisation and intervention work [1–3] is now discussed in related work. Neo et al. and Jiang et al. analyse visual representations and interventions; LatentLens describes visual tokens using contextualised text. Appendix Figure 29 compares patch vocabulary reading, distinct from the direct V-Write localisation comparison now in Section 7.1 and Appendix D.1.
New direct comparison. On 888 COCO photographs with Qwen3-VL-8B-Instruct, V-Write leads the tested lens localisers on the reported metrics. The methods target different quantities; the comparison is specific to this model, dataset and intervention.