arxiv: 2604.20191 · v2 · submitted 2026-04-22 · 💻 cs.CV · cs.AI· cs.RO

Recognition: unknown

From Scene to Object: Text-Guided Dual-Gaze Prediction

Heye Huang, Jianqiang Wang, Jinhao Li, Qingwen Meng, Yanbo Jiang, Yiqian Tu, Zehong Ke, Zhiyuan Liu

Pith reviewed 2026-05-10 00:58 UTC · model grok-4.3

classification 💻 cs.CV cs.AIcs.RO

keywords driver attention predictionobject-level gazevision-language modelsdual-branch architecturetext-guided predictionautonomous drivingattention heatmapsSAM integration

0 comments

The pith

A new dual-branch VLM framework predicts driver attention at the object level using text guidance rather than scene-level heatmaps.

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

The paper seeks to overcome the lack of fine-grained object annotations in driver attention datasets, which prevents vision-language models from performing reliable text-grounded reasoning. It does this by first building an object-level dataset through integration of a multimodal LLM with SAM3 to convert global gaze maps into precise per-object masks. The resulting DualGaze-VLM then uses semantic query states to dynamically gate and anchor visual features according to driving intent. If successful, this produces attention predictions that align better with human cognition in safety-critical situations and generate heatmaps humans accept as authentic.

Core claim

The paper claims that integrating a multimodal LLM with SAM3 under cross-validation creates reliable object-level ground truth masks from scene-level data, and that a dual-branch architecture extracting hidden states from semantic queries and applying a Condition-Aware SE-Gate for feature modulation enables precise intent-driven spatial anchoring, yielding up to 17.8 percent gains in similarity metrics on the W3DA benchmark and 88.22 percent authenticity in human visual Turing tests.

What carries the argument

The Condition-Aware SE-Gate, which takes semantic query hidden states to dynamically modulate visual features for intent-driven object anchoring.

If this is right

Object-level gaze data enables VLMs to perform semantic reasoning without text-vision decoupling.
The dual-branch design supports precise anchoring of attention to specific objects in safety-critical driving scenes.
Generated heatmaps pass human authenticity checks at high rates, indicating they capture rational cognitive priors.
The data construction pipeline can be applied to convert existing scene-level datasets into object-level ones.
Improved metrics in safety scenarios suggest better support for human-like decision making in autonomous driving.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

The approach could generalize to other text-conditioned vision tasks where scene-level labels currently limit model precision.
Object-level attention maps might serve as more interpretable priors for downstream planning modules in robotics.
If the data pipeline proves robust, it could reduce reliance on expensive manual object annotations in attention research.
Testing the same architecture on non-driving gaze datasets would reveal whether the dual-branch mechanism is domain-specific.

Load-bearing premise

That multimodal LLMs plus SAM3 under cross-validation can generate accurate object-level masks that remove annotation hallucinations and serve as trustworthy training targets.

What would settle it

A new driver attention dataset with independently verified object-level masks where DualGaze-VLM fails to exceed prior scene-level models in spatial alignment metrics.

Figures

Figures reproduced from arXiv: 2604.20191 by Heye Huang, Jianqiang Wang, Jinhao Li, Qingwen Meng, Yanbo Jiang, Yiqian Tu, Zehong Ke, Zhiyuan Liu.

**Figure 1.** Figure 1: The overall architecture of the G-W3DA dataset generation pipeline and the proposed DualGaze-VLM framework. The left panel illustrates the [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗

**Figure 2.** Figure 2: Comparison of dataset annotation paradigms. [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗

**Figure 4.** Figure 4: Visualization of the strong center-bias prior via aggregated ground [PITH_FULL_IMAGE:figures/full_fig_p007_4.png] view at source ↗

**Figure 3.** Figure 3: Detailed architecture of the Shared Progressive Cognitive Decoder. [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗

**Figure 5.** Figure 5: Qualitative evaluation of dataset refinement and attention decoupling. (a) Dataset Refinement: Two comparative cases demonstrating the precision of our semantic parsing. The top case shows G-W3DA precisely anchoring attention to the specific “black hatchback car” missed by raw ambiguous annotations. The bottom case illustrates the successful recovery of secondary cognitive targets with weaker attention den… view at source ↗

**Figure 6.** Figure 6: Diversity of driving scenarios in the W3DA benchmark. Within each dataset block, the top row displays the original raw driving scenes, while the bottom row illustrates the corresponding ground-truth attention heatmaps overlaid on the images. The benchmark encompasses diverse conditions: (a) BDDA (safety-critical), (b) DADA (accident scenarios), as well as (c) DR(eye)VE and (d) LBW (normal driving). car”, w… view at source ↗

**Figure 7.** Figure 7: Qualitative comparison of global attention prediction. Our method generates attention maps that are more consistent with ground truth than FSDAM, with better coverage of multiple hazard-relevant regions, stronger sensitivity to pedestrians and other vulnerable road users, and more compact responses with less diffuse activation. It also clearly mitigates the center-bias of the baseline by shifting attention… view at source ↗

**Figure 8.** Figure 8: Confusion Matrix of the human evaluation test. The matrix illustrates [PITH_FULL_IMAGE:figures/full_fig_p012_8.png] view at source ↗

read the original abstract

Interpretable driver attention prediction is crucial for human-like autonomous driving. However, existing datasets provide only scene-level global gaze rather than fine-grained object-level annotations, inherently failing to support text-grounded cognitive modeling. Consequently, while Vision-Language Models (VLMs) hold great potential for semantic reasoning, this critical data limitations leads to severe text-vision decoupling and visual-bias hallucinations. To break this bottleneck and achieve precise object-level attention prediction, this paper proposes a novel dual-branch gaze prediction framework, establishing a complete paradigm from data construction to model architecture. First, we construct G-W3DA, a object-level driver attention dataset. By integrating a multimodal large language model with the Segment Anything Model 3 (SAM3), we decouple macroscopic heatmaps into object-level masks under rigorous cross-validation, fundamentally eliminating annotation hallucinations. Building upon this high-quality data foundation, we propose the DualGaze-VLM architecture. This architecture extracts the hidden states of semantic queries and dynamically modulates visual features via a Condition-Aware SE-Gate, achieving intent-driven precise spatial anchoring. Extensive experiments on the W3DA benchmark demonstrate that DualGaze-VLM consistently surpasses existing state-of-the-art (SOTA) models in spatial alignment metrics, notably achieving up to a 17.8% improvement in Similarity (SIM) under safety-critical scenarios. Furthermore, a visual Turing test reveals that the attention heatmaps generated by DualGaze-VLM are perceived as authentic by 88.22% of human evaluators, proving its capability to generate rational cognitive priors.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

This paper adds a new object-level driver attention dataset and a text-guided dual-branch VLM, but the dataset masks lack any reported accuracy metrics against human labels.

read the letter

The main takeaway is a new dataset called G-W3DA that breaks scene-level gaze into object masks, paired with DualGaze-VLM that routes text queries through a Condition-Aware SE-Gate to steer visual features. The approach targets a clear gap in driving attention work, where global heatmaps make it hard to link gaze to specific objects or language descriptions. The dual-branch design and gate mechanism look like a practical way to inject semantic intent without overhauling the whole backbone. They also show gains on W3DA, including the 17.8% SIM lift in safety-critical cases and 88% human approval in the Turing test, which at least suggests the outputs feel more natural than prior models. The dataset pipeline using MLLM plus SAM3 under cross-validation is the part that stands out as not directly copied from earlier papers. The soft spot sits in the data quality claim. The paper asserts that the MLLM-SAM3 process removes hallucinations and supplies reliable object masks, yet it gives no IoU numbers, precision figures, or agreement scores against human annotators. Without those checks, any downstream performance numbers or Turing results could partly reflect noise in the labels rather than true model improvement. The abstract also omits baseline details and error bars, so the size of the advance is still hard to pin down. This is useful for groups working on interpretable attention for autonomous driving or on VLMs that need fine-grained spatial grounding. A reader who already runs gaze experiments on driving data would get the most from the dataset release and the architecture tweaks. The work is coherent enough on its own terms to go to referees, even with the validation gap. I would send it for review and ask specifically for mask accuracy stats and full ablation tables in the next round.

Referee Report

2 major / 1 minor

Summary. The paper claims to address limitations in scene-level driver gaze datasets by constructing G-W3DA, an object-level attention dataset via MLLM + SAM3 integration under cross-validation to eliminate hallucinations, and introduces DualGaze-VLM, a dual-branch VLM architecture using semantic query states and a Condition-Aware SE-Gate for text-guided object-level prediction. It reports up to 17.8% SIM improvement over SOTA on W3DA and 88.22% human authenticity in a visual Turing test.

Significance. If the object-level masks prove accurate and the gains hold with proper controls, the work could meaningfully advance text-grounded, interpretable attention modeling for autonomous driving by shifting from global heatmaps to object-centric cognitive priors, with the Turing test providing useful human validation.

major comments (2)

[§3 (Dataset Construction, implied)] §3 (Dataset Construction, implied): The assertion that MLLM + SAM3 under 'rigorous cross-validation' 'fundamentally eliminat[es] annotation hallucinations' to yield reliable object-level masks for G-W3DA is load-bearing for all downstream claims, yet no quantitative validation (IoU, precision, recall, or inter-annotator agreement vs. human ground truth) or failure-case analysis is referenced. Without these, residual semantic errors could explain the reported 17.8% SIM gain and 88.22% Turing-test result rather than the proposed architecture.
[Experimental section (implied)] Experimental section (implied): The abstract states that DualGaze-VLM 'consistently surpasses existing state-of-the-art models' with a specific 17.8% SIM improvement under safety-critical scenarios, but provides no information on the exact baselines, dataset splits, number of runs, error bars, or statistical significance tests. This absence prevents assessment of whether the central performance claim is robust.

minor comments (1)

[Abstract] Abstract: 'this critical data limitations leads' contains a subject-verb agreement error and should read 'this critical data limitation leads'.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the detailed and constructive feedback. We address the two major comments point by point below, agreeing that additional quantitative details are needed to strengthen the claims. We will incorporate the suggested revisions in the next version of the manuscript.

read point-by-point responses

Referee: The assertion that MLLM + SAM3 under 'rigorous cross-validation' 'fundamentally eliminat[es] annotation hallucinations' to yield reliable object-level masks for G-W3DA is load-bearing for all downstream claims, yet no quantitative validation (IoU, precision, recall, or inter-annotator agreement vs. human ground truth) or failure-case analysis is referenced. Without these, residual semantic errors could explain the reported 17.8% SIM gain and 88.22% Turing-test result rather than the proposed architecture.

Authors: We agree that quantitative validation metrics are essential to substantiate the reliability of G-W3DA. While Section 3 describes the cross-validation procedure with MLLM and SAM3 to detect and correct hallucinations, we did not report numerical results such as IoU, precision, recall, or inter-annotator agreement against human annotations, nor a dedicated failure-case analysis. This omission weakens the load-bearing claim. In the revision, we will add a new subsection (3.3) presenting these metrics computed on a held-out human-annotated subset, along with examples of detected hallucinations and corrections. We expect these additions to confirm that residual errors do not account for the performance gains. revision: yes
Referee: The abstract states that DualGaze-VLM 'consistently surpasses existing state-of-the-art models' with a specific 17.8% SIM improvement under safety-critical scenarios, but provides no information on the exact baselines, dataset splits, number of runs, error bars, or statistical significance tests. This absence prevents assessment of whether the central performance claim is robust.

Authors: The experimental section (Section 4) specifies the W3DA benchmark, the safety-critical scenario subset, and the compared SOTA baselines (including their original implementations). However, we acknowledge the absence of explicit details on the number of runs, error bars, and statistical tests, which limits assessment of robustness. We will revise Section 4.2 and the abstract to include: (i) the precise train/test splits, (ii) mean and standard deviation over 5 independent runs, and (iii) p-values from paired t-tests against the strongest baseline. These changes will make the 17.8% SIM improvement claim fully verifiable. revision: yes

Circularity Check

0 steps flagged

No circularity; empirical claims rest on novel dataset and architecture

full rationale

The paper introduces G-W3DA via MLLM+SAM3 integration with cross-validation for object-level masks, then DualGaze-VLM with semantic query hidden states and Condition-Aware SE-Gate modulation. Performance is reported via experiments (17.8% SIM gain, 88.22% Turing test) on W3DA benchmark. No equations, derivations, or predictions appear that reduce by construction to fitted inputs or self-citations. Central claims derive from new data construction and model design rather than self-referential loops, rendering the chain self-contained.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract-only information prevents identification of specific free parameters, axioms, or invented entities; the work introduces a new dataset and architecture but supplies no equations or implementation details for auditing.

pith-pipeline@v0.9.0 · 5605 in / 1157 out tokens · 54109 ms · 2026-05-10T00:58:38.498472+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

35 extracted references · 5 canonical work pages · 1 internal anchor

[1]

Hammerdrive: A task- aware driving visual attention model,

P. V . Amadori, T. Fischer, and Y . Demiris, “Hammerdrive: A task- aware driving visual attention model,”IEEE Transactions on Intelligent Transportation Systems, vol. 23, no. 6, pp. 5573–5585, 2021

2021
[2]

Dada-2000: Can driving accident be predicted by driver attentionƒ analyzed by a benchmark,

J. Fang, D. Yan, J. Qiao, J. Xue, H. Wang, and S. Li, “Dada-2000: Can driving accident be predicted by driver attentionƒ analyzed by a benchmark,” in2019 IEEE Intelligent Transportation Systems Confer- ence (ITSC). IEEE, 2019, pp. 4303–4309

2000
[3]

Predicting driver attention in critical situations,

Y . Xia, D. Zhang, J. Kim, K. Nakayama, K. Zipser, and D. Whitney, “Predicting driver attention in critical situations,” inAsian conference on computer vision. Springer, 2018, pp. 658–674

2018
[4]

Dr (eye) ve: a dataset for attention-based tasks with applications to autonomous and assisted driving,

S. Alletto, A. Palazzi, F. Solera, S. Calderara, and R. Cucchiara, “Dr (eye) ve: a dataset for attention-based tasks with applications to autonomous and assisted driving,” inProceedings of the ieee conference on computer vision and pattern recognition workshops, 2016, pp. 54–60

2016
[5]

LLaDA 1.5: Variance-Reduced Preference Optimization for Large Language Diffusion Models

F. Zhu, R. Wang, S. Nie, X. Zhang, C. Wu, J. Hu, J. Zhou, J. Chen, Y . Lin, J.-R. Wenet al., “Llada 1.5: Variance-reduced preference optimization for large language diffusion models,”arXiv preprint arXiv:2505.19223, 2025

work page internal anchor Pith review arXiv 2025
[6]

Simlingo: Vision-only closed-loop autonomous driving with language-action alignment,

K. Renz, L. Chen, E. Arani, and O. Sinavski, “Simlingo: Vision-only closed-loop autonomous driving with language-action alignment,” in Proceedings of the Computer Vision and Pattern Recognition Confer- ence, 2025, pp. 11 993–12 003

2025
[8]

A deep multi-level network for saliency prediction,

M. Cornia, L. Baraldi, G. Serra, and R. Cucchiara, “A deep multi-level network for saliency prediction,” in2016 23rd International Conference on Pattern Recognition (ICPR). IEEE, 2016, pp. 3488–3493. 13

2016
[9]

Where does the driver look? top- down-based saliency detection in a traffic driving environment,

T. Deng, K. Yang, Y . Li, and H. Yan, “Where does the driver look? top- down-based saliency detection in a traffic driving environment,”IEEE Transactions on Intelligent Transportation Systems, vol. 17, no. 7, pp. 2051–2062, 2016

2051
[10]

Predicting the driver’s focus of attention: The dr (eye) ve project,

A. Palazzi, D. Abati, F. Solera, R. Cucchiaraet al., “Predicting the driver’s focus of attention: The dr (eye) ve project,”IEEE transactions on pattern analysis and machine intelligence, vol. 41, no. 7, pp. 1720– 1733, 2018

2018
[11]

Medirl: Predicting the visual attention of drivers via maximum entropy deep inverse reinforcement learning,

S. Baee, E. Pakdamanian, I. Kim, L. Feng, V . Ordonez, and L. Barnes, “Medirl: Predicting the visual attention of drivers via maximum entropy deep inverse reinforcement learning,” inProceedings of the IEEE/CVF international conference on computer vision, 2021, pp. 13 178–13 188

2021
[12]

Scout+: Towards practical task-driven drivers’ gaze prediction,

I. Kotseruba and J. K. Tsotsos, “Scout+: Towards practical task-driven drivers’ gaze prediction,” in2024 IEEE Intelligent Vehicles Symposium (IV). IEEE, 2024, pp. 1927–1932

2024
[13]

Fblnet: Feedback loop network for driver attention prediction,

Y . Chen, Z. Nan, and T. Xiang, “Fblnet: Feedback loop network for driver attention prediction,” inProceedings of the IEEE/CVF interna- tional conference on computer vision, 2023, pp. 13 371–13 380

2023
[14]

Behavior- aware knowledge-embedded model for driver attention prediction,

Y . Zhou, C. Gou, Z. Guo, Y . Cheng, and H. J. Chang, “Behavior- aware knowledge-embedded model for driver attention prediction,”IEEE Transactions on Circuits and Systems for Video Technology, 2025

2025
[15]

Unsupervised self-driving attention prediction via uncertainty mining and knowledge embedding,

P. Zhu, M. Qi, X. Li, W. Li, and H. Ma, “Unsupervised self-driving attention prediction via uncertainty mining and knowledge embedding,” inProceedings of the IEEE/CVF international conference on computer vision, 2023, pp. 8558–8568

2023
[16]

Salm 2: An extremely lightweight saliency mamba model for real-time cognitive awareness of driver attention,

C. Zhao, W. Mu, X. Zhou, W. Liu, F. Yan, and T. Deng, “Salm 2: An extremely lightweight saliency mamba model for real-time cognitive awareness of driver attention,” inProceedings of the AAAI Conference on Artificial Intelligence, vol. 39, no. 2, 2025, pp. 1647–1655

2025
[18]

arXiv preprint arXiv:2512.24331 (2025) 4

W. Wei, Z. Luo, L. Feng, and V . E. Liong, “Spatial-aware vision lan- guage model for autonomous driving,”arXiv preprint arXiv:2512.24331, 2025

work page arXiv 2025
[20]

How do drivers allocate their potential attention? driving fixation prediction via convolu- tional neural networks,

T. Deng, H. Yan, L. Qin, T. Ngo, and B. Manjunath, “How do drivers allocate their potential attention? driving fixation prediction via convolu- tional neural networks,”IEEE Transactions on Intelligent Transportation Systems, vol. 21, no. 5, pp. 2146–2154, 2019

2019
[21]

Maad: A model and dataset for

D. Gopinath, G. Rosman, S. Stent, K. Terahata, L. Fletcher, B. Argall, and J. Leonard, “Maad: A model and dataset for” attended awareness” in driving,” inProceedings of the IEEE/CVF International Conference on Computer Vision, 2021, pp. 3426–3436

2021
[22]

Gaze360: Physically unconstrained gaze estimation in the wild,

P. Kellnhofer, A. Recasens, S. Stent, W. Matusik, and A. Torralba, “Gaze360: Physically unconstrained gaze estimation in the wild,” in Proceedings of the IEEE/CVF international conference on computer vision, 2019, pp. 6912–6921

2019
[23]

Do low birth weight infants not see eyes? face recognition in infancy,

M. Yamamoto, Y . Konishi, I. Kato, K. Koyano, S. Nakamura, T. Nishida, and T. Kusaka, “Do low birth weight infants not see eyes? face recognition in infancy,”Brain and Development, vol. 43, no. 2, pp. 186– 191, 2021

2021
[24]

Squeeze-and-excitation networks,

J. Hu, L. Shen, and G. Sun, “Squeeze-and-excitation networks,” in2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2018, pp. 7132–7141

2018
[25]

Learning to predict where humans look,

T. Judd, K. Ehinger, F. Durand, and A. Torralba, “Learning to predict where humans look,” in2009 IEEE 12th International Conference on Computer Vision, 2009, pp. 2106–2113

2009
[26]

Analysis of scores, datasets, and models in visual saliency prediction,

A. Borji, H. R. Tavakoli, D. N. Sihite, and L. Itti, “Analysis of scores, datasets, and models in visual saliency prediction,” in2013 IEEE International Conference on Computer Vision, 2013, pp. 921–928

2013
[27]

A model of saliency-based visual at- tention for rapid scene analysis,

L. Itti, C. Koch, and E. Niebur, “A model of saliency-based visual at- tention for rapid scene analysis,”IEEE Transactions on pattern analysis and machine intelligence, vol. 20, no. 11, pp. 1254–1259, 1998

1998
[28]

Exploiting the gbvs for saliency aware gaze heatmaps,

D. Geisler, D. Weber, N. Castner, and E. Kasneci, “Exploiting the gbvs for saliency aware gaze heatmaps,” inACM Symposium on Eye Tracking Research and Applications, 2020, pp. 1–5

2020
[29]

Deep gaze i: Boosting saliency prediction with feature maps trained on imagenet,

M. K ¨ummerer, L. Theis, and M. Bethge, “Deep gaze i: Boosting saliency prediction with feature maps trained on imagenet,”arXiv preprint arXiv:1411.1045, 2014

work page arXiv 2014
[30]

Deepgaze iie: Calibrated prediction in and out-of-domain for state-of-the-art saliency modeling,

A. Linardos, M. K ¨ummerer, O. Press, and M. Bethge, “Deepgaze iie: Calibrated prediction in and out-of-domain for state-of-the-art saliency modeling,” inProceedings of the ieee/cvf international conference on computer vision, 2021, pp. 12 919–12 928

2021
[31]

Visual saliency prediction using a mixture of deep neural networks,

S. F. Dodge and L. J. Karam, “Visual saliency prediction using a mixture of deep neural networks,”IEEE Transactions on Image Processing, vol. 27, no. 8, pp. 4080–4090, 2018

2018
[32]

Fbnet: Feedback-recursive cnn for saliency detection,

G. Ding, N. ˙Imamo˘glu, A. Caglayan, M. Murakawa, and R. Nakamura, “Fbnet: Feedback-recursive cnn for saliency detection,” in2021 17th International Conference on Machine Vision and Applications (MVA). IEEE, 2021, pp. 1–5

2021
[33]

A convnet for the 2020s,

Z. Liu, H. Mao, C.-Y . Wu, C. Feichtenhofer, T. Darrell, and S. Xie, “A convnet for the 2020s,” inProceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2022, pp. 11 976–11 986

2022
[34]

Erfnet: Effi- cient residual factorized convnet for real-time semantic segmentation,

E. Romera, J. M. Alvarez, L. M. Bergasa, and R. Arroyo, “Erfnet: Effi- cient residual factorized convnet for real-time semantic segmentation,” IEEE Transactions on Intelligent Transportation Systems, vol. 19, no. 1, pp. 263–272, 2017

2017
[35]

Gazexplain: Learning to predict natu- ral language explanations of visual scanpaths,

X. Chen, M. Jiang, and Q. Zhao, “Gazexplain: Learning to predict natu- ral language explanations of visual scanpaths,” inEuropean Conference on Computer Vision. Springer, 2024, pp. 314–333

2024
[36]

Where, what, why: Towards explainable driver attention prediction,

Y . Zhou, J. Tang, X. Xiao, Y . Lin, L. Liu, Z. Guo, H. Fei, X. Xia, and C. Gou, “Where, what, why: Towards explainable driver attention prediction,” inProceedings of the IEEE/CVF International Conference on Computer Vision, 2025, pp. 2675–2685

2025
[37]

Fsdam: Few- shot driving attention modeling via vision-language coupling,

K. Hamid, C. Cui, K. A. Akbar, Z. Wang, and N. Liang, “Fsdam: Few- shot driving attention modeling via vision-language coupling,”arXiv preprint arXiv:2511.12708, 2025

work page arXiv 2025
[38]

A. M. Turing,Computing Machinery and Intelligence. Dordrecht: Springer Netherlands, 2009, pp. 23–65. [Online]. Available: https: //doi.org/10.1007/978-1-4020-6710-5 3

work page doi:10.1007/978-1-4020-6710-5 2009