Visual Error Patterns in Multi-Modal AI: A Statistical Approach

Ching-Yi Wang

REVIEW 5 major objections 5 minor 13 references

Visual Error Patterns in Multi-Modal AI: A Statistical Approach

T0 review · 5 major / 5 minor · reviewed 2026-08-12 · deepseek-v4-flash

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

Pith's one-line read GPT-4o's classification errors on 75 geometric stimuli are predictable from binary visual-feature labels, with XGBoost reaching AUC 0.85 and '3D' plus 'Pentagon' dominating the feature-importance signal.

desk verdict Small, honest study of GPT-4o's geometric errors, but the headline AUC is uninterpretable until the paper confirms the 'missing face' feature was excluded from the predictors. read the letter →

arxiv 2412.00083 v3 pith:763KA24V submitted 2024-11-27 cs.LG cs.AIcs.CVstat.AP

Ching-Yi Wang This is my paper · ORCID

classification cs.LGcs.AIcs.CVstat.AP

keywords multimodallargelanguagemodelsGPT-4ovisualclassificationerrorsgradientboostingXGBoostfeatureimportancemissingfacegeometricstimuli

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

The paper sets out to show that the systematic mistakes of a multi-modal language model can be mapped statistically, without opening the network. It feeds GPT-4o 75 geometric images tagged with binary features such as 3D, rotation, and missing face, records where the model errs, and then trains four classifiers to predict those errors from the tags. XGBoost, a gradient-boosted tree ensemble, achieves a 5-fold cross-validated AUC of 0.85, and its feature-importance scores single out '3D' and 'Pentagon' as the dominant drivers of missing-face errors. If the result holds, it means a small labeled stimulus set plus an off-the-shelf classifier is enough to expose which visual properties cause a multi-modal model to fail, and to point at targeted fixes such as amodal completion.

What carries the argument

The load-bearing mechanism is an XGBoost binary classifier—gradient-boosted decision trees that iteratively fit new trees to the residual errors of the previous ensemble. It is trained on rows of binary-coded visual-feature indicators for the 75 stimuli and asked to predict a binary outcome: whether GPT-4o made a missing-face error on that stimulus. The classifier's internal gain-based feature importance then ranks which tags carry the predictive signal. The argument leans on two quantitative comparisons: the AUC gap between the best non-linear model (0.85) and the linear baselines (0.79), and the dominance of the '3D' and 'Pentagon' importance values over all other tags.

What would settle it

Run the same 5-fold cross-validation with the 'missing face' column removed from the predictor matrix; if the XGBoost AUC collapses from 0.85 toward 0.5, the headline result is label leakage rather than a discovered error pattern.

Watch

Extended reading notes

Core claim

On its own terms, the paper's central discovery is that GPT-4o's visual classification errors are not random but predictable from simple feature labels. The predictive target is a specific failure mode—whether the model made a missing-face error—and XGBoost predicts it with mean AUC 0.85 over 5-fold cross-validation, above the 0.79 of logistic regression and ridge logistic regression and the 0.80 of random forest. The fitted model's feature importance assigns 33.61% to the 'Pentagon' tag and 33.24% to the '3D' tag, which the paper reads as evidence that depth perception and reconstruction of incomplete structures are the main sources of error. This is taken to support the feature-specific error hypothesis and the broader claim that GPT-4o's bottom-up processing lacks the top-down, amodal-completion machinery humans use to fill in missing parts.

Load-bearing premise

The main result assumes that the 'missing face' tag used to define the prediction target was not also left on the list of features the model was trained on; the paper never says it was removed.

Editorial extensions

If this is right

The same pipeline can rank any multi-modal model's failure-inducing visual features from a small labeled stimulus set, without access to the model's internal activations.
Architectural fixes should target reconstruction of incomplete 3D structures, since missing faces dominate the error signal.
Because non-linear XGBoost clearly beats the linear baselines, future error-prediction studies should use non-linear ensembles rather than regression alone.
The feature analysis points specifically to adding top-down, amodal-completion-style reasoning to multi-modal models.

Reading between the lines

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

If the result survives removal of the 'missing face' feature from the predictor set, the method becomes a cheap, general diagnostic: 75 images and a gradient-boosted classifier are enough to map a model's error surface.
The 'Pentagon' importance may be an artifact of the stimulus design, since pentagonal prisms were the specific 3D shape category with missing faces; a balanced set of 3D shapes would reveal whether the effect is about pentagons or about 3D incompleteness.
Applying the same XGBoost pipeline to a second multi-modal model would test whether the 3D/missing-face pattern is GPT-4o-specific or a general property of current multi-modal models.

Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Desk Editor's Note

read the letter

Two things to know. First, this is a small, honest empirical study: 75 geometric stimuli, one API model, four standard classifiers, and a clear prompt. The observation that GPT-4o misses 63% of 3D stimuli with missing faces is plausible and worth confirming. Second, the headline result — XGBoost predicts those errors with AUC 0.85 — is currently not interpretable because the paper never states whether the 'missing face' indicator was removed from the predictor matrix. Section 2.1 lists 'missing faces' as an input feature; Section 3.2 defines the prediction target as 'missing face errors.' If that feature stayed in X, the AUC is largely a restatement of the label, and the feature-importance plot (which omits 'missing face') becomes unreadable.

What the paper does well: the experimental setup is transparent enough to reproduce privately, the authors report a limitation paragraph, and they did not oversell individual model differences. The error-rate table, despite missing confidence intervals, is a useful snapshot of where a frontier MLLM fails on synthetic geometry.

The soft spots are proportional to their seriousness. The target/predictor overlap is the load-bearing issue, and the fix is trivial: say the indicator was excluded, or show the feature matrix. The second issue is statistical: 75 samples, per-cell counts for the 63.2% rate are unreported, and hyperparameters are tuned on the same 75 cases without nested CV. The AUC SD of 0.02 says nothing about the uncertainty of the error rate itself. Third, the paper claims the feature-importance result supports a bottom-up processing limitation in GPT-4o. That is an interpretive leap: feature importance from a fitted model describes the model's predictors, not the multimodal model's cognitive architecture. The data do not measure top-down vs bottom-up processing.

I find the reader's take fair. The stress-test note is correct about the leakage ambiguity; I'd add that the omission of 'missing face' from the feature-importance list suggests the author may have excluded it, but the paper must say so. The cited literature is reasonable; self-citation is not an issue here.

Who is this for? Someone thinking about benchmark design for MLLMs on geometric reasoning could use this as a cautionary example. It is not a foundational contribution. I would not cite it in my own work, but I'd bring it to a reading group to discuss how easy it is to leak the target into a feature matrix.

Recommendation: this deserves a serious referee, not a desk reject. The empirical core is clear enough that a referee can request the feature matrix, the exclusion statement, and the per-cell counts, and then judge whether the result stands. As submitted, the central claim is conditional on that missing clarification.

Referee Report

5 major / 5 minor

Summary. The paper studies GPT-4o's visual classification errors on 75 geometric stimuli (54 3D and 21 2D) that are annotated with shape and spatial features such as 3D, rotation, missing face, arrangement, and repetition. Error rates are reported by stimulus type and feature (Table 1 and Table 2), and four classifiers (logistic regression, ridge logistic regression, random forest, and XGBoost) are trained to predict the occurrence of "missing face errors" from the image features. XGBoost is reported to achieve the best 5-fold cross-validated AUC of 0.85 (SD = 0.02), and feature importance analysis identifies "3D" and "Pentagon" as dominant predictors. The paper interprets these results as evidence that GPT-4o relies on bottom-up processing and struggles with incomplete 3D structures.

Significance. The paper addresses a practically important question: can statistical models trained on image-feature annotations reveal systematic failure modes of a multimodal large language model? The strongest contribution is the idea of using classifier AUC and feature importance as a diagnostic tool for probing GPT-4o's error patterns, and the paper is transparent about the small dataset in its Discussion. If the reported results are correct, this is a useful proof-of-concept for the broader goal of explaining MLLM limitations. However, the empirical value is currently conditional on resolving the target/predictor overlap, avoiding selection bias in hyperparameter tuning, and disclosing the data and code needed for independent verification. As written, the headline AUC and feature-importance conclusions are too fragile to be accepted as established findings.

major comments (5)

[§2.1, §3.2, Table 2] The prediction target and the predictor vocabulary are not kept distinct. Section 2.1 lists "missing faces" among the spatial attributes used to characterize the stimuli, while Section 3.2 defines the prediction target as "the occurrence of missing face errors." The paper never states explicitly whether the missing-face indicator (or any other error-category feature) was removed from the predictor matrix X. Since Table 2 reports missing-face error rates of 63.2% for 3D stimuli and 25% for 2D stimuli, a retained missing-face indicator would be a near-perfect proxy for the positive class, and the 5-fold CV AUC of 0.85 (SD = 0.02) would be largely a restatement of the label rather than evidence that the model learned error-driving features. Please provide the precise definition of the target (e.g., a binary indicator of whether the error was assigned to the "Missing Face" category, versus a binary correct/incorrect outcome on stimuli containing a missing face), and report whether the missing-face feature was excluded from X. If it was not excluded, re-run the comparison with that feature removed and report the resulting AUC and importance rankings.
[§2.4, Algorithm 1, §2.5] The hyperparameter tuning protocol is not sufficiently specified to rule out selection bias. Algorithm 1 evaluates hyperparameter combinations on a "validation set," but the text does not state how that validation set is formed relative to the reported 5-fold cross-validation. If hyperparameters were tuned on the same 75 samples that are then used to report the mean AUC (and used for model comparison), the reported AUC of 0.85 is optimistically biased and the comparison with the other models is unreliable. Please describe exactly when hyperparameters are selected: if selection is performed on the full dataset before cross-validation, provide nested cross-validation or an outer holdout; if selection is performed inside each training fold, state this explicitly and show the per-fold choices.
[§3.1, Table 2, §3.2] The paper reports error rates such as 63.2% for 3D missing faces and 25% for 2D missing faces without giving the number of stimuli per feature cell or any measure of uncertainty. With only 75 stimuli and multiple overlapping feature categories, the effective sample size for a given feature may be very small, so the stability of these rates, of the SD = 0.02 reported for XGBoost AUC, and of the resulting feature-importance rankings is unverified. Please report per-feature and per-error-cell counts, and add confidence intervals (or exact binomial tests) for the key error-rate differences so the reader can assess whether the differences are meaningful rather than sampling noise.
[§3.3, Figure 4, Discussion] The feature importance analysis is not adequately documented or internally consistent. Section 3.3 reports "Pentagon" (33.61%) and "3D" (33.24%) as dominant and lists small values for "Triangle" (0.06%) and "Rotation" (2.14%), but the Discussion states that "3D," "Circle," and "Pentagon" were the most influential predictors, with "Circle" not appearing in the Section 3.3 list. The paper also does not state which XGBoost importance measure is used (gain, cover, or split count), and Figure 4 is not visible in the submitted text. Please provide the full importance table, the exact importance metric, and reconcile the discrepancy between the Results and Discussion sections, since the feature-importance conclusions are central to the paper's claims.
[Methods and Results] The paper does not release the stimulus dataset, the GPT-4o outputs, the feature matrix, or the analysis code, despite all central claims being empirical and based on only 75 samples. Given the target/feature overlap issue described above, independent verification of the reported AUCs, error rates, and importance values is impossible. Please make the data and code available in a form that allows the predictor matrix and the target construction to be inspected.

minor comments (5)

[Title] The title contains a stray space in "Multi-M odal" on the first page; please correct this typo.
[§2.2, Tables 1 and 2] Tables 1 and 2 present nearly identical material; consider consolidating them into a single table and referencing it consistently from the text.
[§3] The text mentions a "multi-regression analysis" in Section 3, but no regression results are reported anywhere in the paper; please either report the regression output or remove this reference.
[§3.3, Figure 4] Figures 2, 3, and 4 are cited in the text, but the figure images and captions do not appear in the submitted manuscript; please ensure the figures are included and that each is discussed in the body.
[Throughout] Use of terminology is inconsistent: "missing face" and "missing faces" appear interchangeably, and "multi-modal" is sometimes written with and sometimes without a hyphen; please standardize these terms.

Circularity Check

1 steps flagged · score 6.0 of 10

Target feature 'missing faces' doubles as the outcome category 'missing face errors,' so the reported 0.85 AUC may restate the label unless that feature was explicitly excluded from the predictor matrix.

self definitional [Section 2.1 (Dataset features), Section 3.2 (Model Results), and Table 1]
"Each stimulus was characterized by a combination of geometric features, including '3D,' 'Circle,' 'Pentagon,' 'Hexagon,' 'Cube,' and 'Triangle,' alongside spatial attributes such as arrangement patterns, repetition, combination, rotation, and missing faces. ... We evaluated the predictive performance of four statistical models—Logistic Regression, Ridge Logistic Regression, Random Forest, and Gradient Boosting (XGBoost)—to predict the occurrence of missing face errors. ... Missing Face Significant error in 3D (63.2%) and moderate error in 2D tasks (25%)."

The outcome 'missing face errors' is defined by the same spatial attribute 'missing faces' that Section 2.1 lists among the stimulus features. If that attribute is included in the predictor matrix, the target is nested inside a predictor: a stimulus can have a missing-face error only if it has the 'missing face' attribute, and Table 1 shows the attribute alone separates a large share of the class (63.2% in 3D, 25% in 2D). A model could therefore achieve high AUC by thresholding the missing-face indicator or its correlates; the 0.85 cross-validated AUC would then be partly a restatement of the label rather than evidence that the model learned meaningful feature interactions.

full rationale

The paper's central claim is that XGBoost predicts GPT-4o's 'missing face errors' with AUC 0.85 and that feature importance identifies '3D' and 'Pentagon' as dominant predictors. The most load-bearing circularity risk is that the target category is constructed from the same feature vocabulary as the predictors: Section 2.1 lists 'missing faces' as a spatial attribute, and Section 3.2 predicts 'the occurrence of missing face errors.' Because the paper does not state that the missing-face indicator was excluded from the predictor matrix, the high AUC is at least partly a restatement of the label. This is a genuine partial circularity, not merely a weak-assumption concern, because the quoted definitions make the overlap explicit and the reported error rates quantify how much separation the attribute alone provides. The self-citations in the reference list (e.g., [7], [10]) are not load-bearing; they support background claims about AI transparency and MLLM development, so they do not contribute to circularity. The scoring reflects that the outcome still contains independent error variability (the missing-face error rates are 63.2% and 25%, not 100%), so the AUC is not entirely forced; however, the absence of any statement that the missing-face attribute was removed leaves the central prediction partially defined by its own input.

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

The central claim rests on unverified data-handling assumptions rather than new theoretical entities. The key burdens are the possible target-predictor overlap for missing-face errors, the lack of a human baseline, and the small-sample stability of the reported AUC and error rates.

free parameters (3)

XGBoost hyperparameters (learning rate, max depth, number of estimators, subsample ratio) = not reported in final model
Grid-searched on 75 samples in 5-fold CV without nested resampling; can inflate reported AUC.
Random forest hyperparameters (number of trees, max depth, min samples split) = not reported
Tuned on the same small dataset; no final values are given.
Ridge logistic regression regularization strength = not reported
L2 penalty is mentioned but the chosen strength is not reported.

assumptions (4)

domain assumption GPT-4o's visual processing is predominantly bottom-up.
Section 1 and the Discussion invoke this to explain errors, but the study never measures or manipulates processing mechanism.
domain assumption Human top-down processing, including amodal completion, is the correct reference standard.
Section 1 and 4 claim humans resolve missing shapes via top-down processes, but no human data are collected on these stimuli.
domain assumption The hand-coded feature labels and error categories are reliable.
Sections 2.1 and 2.2 describe binary coding of features and errors without inter-rater reliability or a codebook.
domain assumption 5-fold cross-validation on 75 samples yields stable AUC estimates.
Section 2.5 uses 5-fold CV; with n=75 and binary targets, fold-level AUC can be unstable even if the mean looks high.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Visual Error Patterns in Multi-Modal AI: A Statistical Approach." pith.science (2026). https://pith.science/paper/763KA24V

@misc{pith2026241200083,
  author       = {Pith},
  title        = {Pith review of: Visual Error Patterns in Multi-Modal AI: A Statistical Approach},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/763KA24V}},
  note         = {Machine review of arXiv:2412.00083}
}

read the original abstract

Multi-modal large language models (MLLMs), such as GPT-4o, excel at integrating text and visual data but face systematic challenges when interpreting ambiguous or incomplete visual stimuli. This study leverages statistical modeling to analyze the factors driving these errors, using a dataset of geometric stimuli characterized by features like 3D, rotation, and missing face/side. We applied parametric methods, non-parametric methods, and ensemble techniques to predict classification errors, with the non-linear gradient boosting model achieving the highest performance (AUC=0.85) during cross-validation. Feature importance analysis highlighted difficulties in depth perception and reconstructing incomplete structures as key contributors to misclassification. These findings demonstrate the effectiveness of statistical approaches for uncovering limitations in MLLMs and offer actionable insights for enhancing model architectures by integrating contextual reasoning mechanisms.

Figures

Figures reproduced from arXiv: 2412.00083 by the authors.

**Figure 3.** ROC curves with AUC highlighted for XGBoost [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗

**Figure 4.** Bar plot of Feature Importance for XGBoost Model. [PITH_FULL_IMAGE:figures/full_fig_p007_4.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

13 extracted references · 7 canonical work pages

[1]

A Comprehensive Review of Multimodal Large Language Models: Performance and Challenges Across Different Tasks,

J. Wang, H. Jiang, Y . Liu, C. Ma, X. Zhang, Y . Pan, M. Liu, P. Gu, S. Xia, W. Li, Y . Zhang, Z. Wu, Z. Liu, T. Zhong, B. Ge, T. Zhang, N. Qiang, X. Hu, X. Jiang, X. Zhang, W. Zhang, D. Shen, T. Liu, and S. Zhang, “A Comprehensive Review of Multimodal Large Language Models: Performance and Challenges Across Different Tasks,” 2023

work page 2023
[2]

LLM-Select: Feature Selection with Large Language Models,

D. P. Jeong, Z. C. Lipton, and P. Ravikumar, “LLM-Select: Feature Selection with Large Language Models,” arXiv:2407.02694v1 [cs.LG], 2024

arXiv 2024
[3]

InterPreT: Interactive Predicate Learning from Language Feedback for Generalizable Task Planning,

M. Han, Y . Zhu, S.-C. Zhu, Y . N. Wu, and Y . Zhu, “InterPreT: Interactive Predicate Learning from Language Feedback for Generalizable Task Planning,”arXiv preprint arXiv:2405.19758, 2024

arXiv 2024
[4]

Mini-Omni2: Towards Open-source GPT-4o Model with Vision, Speech and Duplex,

Z. Xie and C. Wu, “Mini-Omni2: Towards Open-source GPT-4o Model with Vision, Speech and Duplex,”arXiv preprint, DOI: 10.48550/arXiv.2410.11190, Oct. 2024

arXiv doi:10.48550/arxiv.2410.11190 2024
[5]

J. J. Gibson, The Senses Considered as Perceptual Systems, Houghton Mifflin, 1966

work page 1966
[6]

R. L. Gregory, Eye and Brain: The Psychology of Seeing, 3rd ed., McGraw-Hill, 1974

work page 1974
[7]

(In)Accuracy of Human-Generated Correlations in a Scatterplot Drawing Task,

L. Cui, C. Wang, Y . Wang, P. Li, M. Kini, and Z. Liu, “(In)Accuracy of Human-Generated Correlations in a Scatterplot Drawing Task,” Proceedings of the 46th Annual Meeting of the Cognitive Science Society , 2024. [Online]. Available: https://escholarship.org/uc/item/49m756r5

work page 2024
[8]

Emergent analogical reasoning in large language models,

T. Webb, K. J. Holyoak, and H. Lu, “Emergent analogical reasoning in large language models,”Nature Human Behaviour, vol. 7, no. 9, pp. 1526–1541, Sep. 2023. DOI: 10.1038/s41562-023-01663-5

work page doi:10.1038/s41562-023-01663-5 2023

Show all 13 references

[9]

MM-LLMs: Recent Advances in MultiModal Large Language Models,

D. Zhang, Y . Yu, C. Li, J. Dong, D. Su, C. Chu, and D. Yu, “MM-LLMs: Recent Advances in MultiModal Large Language Models,” arXiv preprint arXiv:2401.13601, Jan. 2024

2024 arXiv
[10]

Bar Tip Limit Error and Characteristics of Drawn Data Distributions on Bar Graphs,

L. Cui, C. Wang, Y . Wang, P. Li, M. Kini, and Z. Liu, “Bar Tip Limit Error and Characteristics of Drawn Data Distributions on Bar Graphs,” Proceedings of the 46th Annual Meeting of the Cognitive Science Society, 2024. [Online]. Available: https://escholarship.org/uc/item/33r7b30j

2024
[11]

HULLMI: Human vs. LLM Identification with Explainability,

P. D. Joshi, S. Pocker, R. A. Dandekar, and R. Dandekar, “HULLMI: Human vs. LLM Identification with Explainability,”arXiv preprint arXiv:2409.04808, Sep. 2024. DOI: 10.48550/arXiv.2409.04808

2024 doi arXiv
[12]

Random forest versus logistic regression: a large-scale benchmark experiment,

R. Couronné, P. Probst, and A.-L. Boulesteix, “Random forest versus logistic regression: a large-scale benchmark experiment,” BMC Bioinformatics, vol. 19, no. 270, 2018. DOI: 10.1186/s12859-018-2264-5

2018 doi
[13]

Feature Importance versus Feature Influence and What It Signifies for Explainable AI,

K. Främling, “Feature Importance versus Feature Influence and What It Signifies for Explainable AI,” Pro- ceedings of the 1st World Conference on eXplainable Artificial Intelligence, Lissabon, 26–28 July, 2023. DOI: 10.48550/arXiv.2308.03589. 9

2023 doi arXiv

Pith tools

Reviewed August 12, 2026 · model on record in the stance chip above.

[1] [1]

A Comprehensive Review of Multimodal Large Language Models: Performance and Challenges Across Different Tasks,

J. Wang, H. Jiang, Y . Liu, C. Ma, X. Zhang, Y . Pan, M. Liu, P. Gu, S. Xia, W. Li, Y . Zhang, Z. Wu, Z. Liu, T. Zhong, B. Ge, T. Zhang, N. Qiang, X. Hu, X. Jiang, X. Zhang, W. Zhang, D. Shen, T. Liu, and S. Zhang, “A Comprehensive Review of Multimodal Large Language Models: Performance and Challenges Across Different Tasks,” 2023

work page 2023

[2] [2]

LLM-Select: Feature Selection with Large Language Models,

D. P. Jeong, Z. C. Lipton, and P. Ravikumar, “LLM-Select: Feature Selection with Large Language Models,” arXiv:2407.02694v1 [cs.LG], 2024

arXiv 2024

[3] [3]

InterPreT: Interactive Predicate Learning from Language Feedback for Generalizable Task Planning,

M. Han, Y . Zhu, S.-C. Zhu, Y . N. Wu, and Y . Zhu, “InterPreT: Interactive Predicate Learning from Language Feedback for Generalizable Task Planning,”arXiv preprint arXiv:2405.19758, 2024

arXiv 2024

[4] [4]

Mini-Omni2: Towards Open-source GPT-4o Model with Vision, Speech and Duplex,

Z. Xie and C. Wu, “Mini-Omni2: Towards Open-source GPT-4o Model with Vision, Speech and Duplex,”arXiv preprint, DOI: 10.48550/arXiv.2410.11190, Oct. 2024

arXiv doi:10.48550/arxiv.2410.11190 2024

[5] [5]

J. J. Gibson, The Senses Considered as Perceptual Systems, Houghton Mifflin, 1966

work page 1966

[6] [6]

R. L. Gregory, Eye and Brain: The Psychology of Seeing, 3rd ed., McGraw-Hill, 1974

work page 1974

[7] [7]

(In)Accuracy of Human-Generated Correlations in a Scatterplot Drawing Task,

L. Cui, C. Wang, Y . Wang, P. Li, M. Kini, and Z. Liu, “(In)Accuracy of Human-Generated Correlations in a Scatterplot Drawing Task,” Proceedings of the 46th Annual Meeting of the Cognitive Science Society , 2024. [Online]. Available: https://escholarship.org/uc/item/49m756r5

work page 2024

[8] [8]

Emergent analogical reasoning in large language models,

T. Webb, K. J. Holyoak, and H. Lu, “Emergent analogical reasoning in large language models,”Nature Human Behaviour, vol. 7, no. 9, pp. 1526–1541, Sep. 2023. DOI: 10.1038/s41562-023-01663-5

work page doi:10.1038/s41562-023-01663-5 2023