Pith. sign in

REVIEW 4 major objections 5 minor 3 cited by

Concept Bottleneck Large Language Models

T0 review · 4 major / 5 minor · reviewed 2026-08-11 · deepseek-v4-flash

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

Pith's one-line read Inserting a concept bottleneck layer into pretrained LLMs gives near-black-box classification and generation quality while exposing interpretable, steerable concepts.

desk verdict Real engineering contribution with a weaker-than-advertised interpretability claim; worth reviewing, but the class-conditioned pseudo-labels and contradictory perplexity numbers need addressing. read the letter →

arxiv 2412.07992 v4 pith:M3TXNTXK submitted 2024-12-11 cs.CL cs.LG

Chung-En Sun , Tuomas Oikarinen , Berk Ustun , Tsui-Wei Weng This is my paper · ORCID
classification cs.CLcs.LG
keywords conceptbottleneckmodelsinterpretablemachinelearninglargelanguagetextclassificationgenerationsteeringunlearningtoxicityreduction
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

This paper proposes Concept Bottleneck Large Language Models (CB-LLMs), a way to turn a pretrained black-box LLM into an inherently interpretable model by inserting a human-interpretable concept bottleneck layer before the prediction layer. For text classification, the authors claim CB-LLMs match or slightly beat fine-tuned black-box models on SST2, Yelp, AGnews, and DBpedia while producing explicit concept-based explanations that human raters judge more faithful than prior interpretable-text models. For text generation, they claim a hybrid design with an adversarially disentangled unsupervised layer lets CB-LLMs detect concepts, steer generation toward a chosen concept, and reduce toxicity, with generation quality and perplexity close to a fine-tuned black-box Llama3-8B. The paper concludes that interpretability need not come with the usual accuracy cost: the reported gap to black-box LLMs is within 1 percent, and the interpretable neurons enable concept unlearning and controllable responses.

What carries the argument

The load-bearing object is the Concept Bottleneck Layer (CBL): a linear layer whose k neurons are each trained to reproduce a named concept score, so that each neuron's activation has a human-readable meaning and its contribution to the final prediction is a linear weight times the neuron activation. Concept scores are produced by Automatic Concept Scoring (ACS), which uses cosine similarity between a sentence embedding of the text and embeddings of ChatGPT-generated concept phrases, and by Automatic Concept Correction (ACC), which uses the ground-truth class label to zero out scores for concepts belonging to other classes. For generation, the machinery adds an unsupervised parallel layer and an adversarial training module: a linear classifier tries to predict concepts from the unsupervised layer's features while the unsupervised layer is trained to maximize entropy of that classifier, forcing concept information to live only in the interpretable CBL and making neuron interventions effective.

What would settle it

Swap the ChatGPT concept lists between two classes before training a CB-LLM, keeping everything else identical; if the resulting model still achieves the same accuracy and human raters still find its neuron explanations equally faithful, then the concept names are not semantically load-bearing and the interpretability claim fails. A more direct test would replace the automatically generated concept scores with human-annotated concept labels on the same dataset and check whether neuron-concept alignment and explanation faithfulness improve.

Watch

Extended reading notes

Core claim

The central claim is that a pretrained LLM can be made intrinsically interpretable without sacrificing task performance by training a Concept Bottleneck Layer (CBL) whose neurons are forced to align with automatically generated, human-readable concepts. Concept lists come from a single ChatGPT query per class; each text's concept scores are computed as cosine similarities between the text embedding and concept embeddings; an Automatic Concept Correction step zeroes out scores whose concepts belong to a different class than the sample's ground-truth label; and the backbone plus CBL are trained to reproduce these corrected scores before a sparse linear layer maps the ReLU'd activations to class logits. In the generation setting, the paper claims the same CBL structure plus an unsupervised parallel layer trained adversarially to remove concept information yields a model whose interpretable neurons can detect the topic or toxicity of a prompt, whose generation can be steered by setting neuron activations, and whose output quality is essentially on par with a black-box fine-tuned model. On the four classification datasets, CB-LLMs with ACC reach 0.94-0.99 accuracy, matching or occasionally exceeding the fine-tuned black-box baseline; in generation, concept detection accuracy is within 1 percent of a directly fine-tuned concept classifier, steerability scores range from 0.76 to 0.95, and perplexity is comparable to the black-box model.

Load-bearing premise

The framework's interpretability rests on the assumption that ChatGPT-generated concept lists and text-similarity scores from an embedding model faithfully capture the human concepts that actually drive each text; if these scores misrepresent human concepts, the named neurons are not truly aligned with their labels and the faithfulness, detection, and steering claims collapse.

Editorial extensions

If this is right

  • CB-LLMs (classification) reach the accuracy of fine-tuned black-box models on SST2, Yelp, AGnews, and DBpedia, with ACC closing the gap by about 3.5 percent on average.
  • CB-LLMs (generation) can detect the concept of a prompt from its neuron activations, with accuracy within 1 percent of a directly fine-tuned concept classifier.
  • Setting a target concept neuron to a high activation steers generation toward that concept; steerability scores range from 0.76 to 0.95, far above a model trained without adversarial disentanglement.
  • The framework supports concept unlearning: deactivating a neuron or removing its weights shifts predictions in the intended direction, e.g., 79 percent of reviews that flipped after unlearning 'overpriced' strongly entail that concept.
  • Toxicity detection in a chatbot variant reaches 0.9996 accuracy, and users can steer responses between benign and toxic by intervening on four interpretable neurons.

Reading between the lines

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

  • If the concept labels are derived from class-level ChatGPT lists and sentence-embedding similarities, then for richer or open-ended generation tasks the 'concepts' are likely to be coarse and surface-level; extending the approach to concept sets built from domain ontologies or human annotation would test whether the interpretability holds beyond class-like categories.
  • The adversarial disentanglement mechanism is a general recipe: it could be applied to any hybrid bottleneck architecture to prevent an unsupervised residual path from silently carrying the information that the interpretable path is supposed to own, and its success suggests information leakage is the main threat to bottleneck faithfulness.
  • A natural next experiment is to measure whether the concept bottleneck's explanations support human debugging in practice, e.g., whether users who can intervene on neurons achieve higher task accuracy than users of a black-box model with post-hoc explanations, which the paper does not directly test.
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

4 major / 5 minor

Summary. The paper introduces Concept Bottleneck Large Language Models (CB-LLMs), a framework that inserts a human-interpretable concept bottleneck layer into pretrained LLMs for both text classification and text generation. For classification, concepts are generated with ChatGPT, scored by sentence-embedding similarity, corrected using ground-truth class labels (ACC), and used to train a bottleneck layer whose activations feed a sparse linear classifier. For generation, a hybrid design combines the concept bottleneck with an unsupervised layer, trained with an adversarial disentanglement objective to keep concept information out of the unsupervised path, enabling concept detection, steering, and toxicity reduction. Experiments on SST2, YelpP, AGnews, and DBpedia report classification accuracy near or above fine-tuned black-box models, human faithfulness ratings above a TBM&C3M baseline, and generation accuracy, steerability, and perplexity close to a fine-tuned Llama3-8B. The paper claims to be the first CBM that scales to large text classification and text generation while providing intrinsic interpretability.

Significance. If the central interpretability claim is valid, this is a significant step toward intrinsically interpretable LLMs. The method is fully automatic, avoids per-sample LLM labeling at inference, and scales to datasets with hundreds of thousands of examples. The paper also contributes a novel adversarial disentanglement mechanism for hybrid CBMs in generation, and it provides multiple case studies (concept unlearning, toxicity detection and steering) with concrete validation of one unlearning behavior via an NLI model. The authors released code, and the appendix contains substantial additional results across backbones and datasets. The main weakness is that the evidence for the faithfulness of the learned neurons is incomplete: the generation concept-detection metric is circular, and the ACC target construction may allow neurons to encode class membership rather than semantic concepts. These issues are load-bearing because the paper's headline contribution is 'faithful interpretability.'

major comments (4)
  1. [§4.2, Eq. (5), Table 5] The reported 'concept detection accuracy' in the generation setting is circular. The concept labels y_c used in Eq. (5) are simply the dataset class labels (as stated in §4.2, e.g., world, sport, business, and technology news for AGnews), and the CBL is trained with cross-entropy to predict these labels. The 'Accuracy' row in Table 5 is therefore the test accuracy of the same classifier that defines the concept-detection objective; high numbers only show that the CBL memorizes the training objective, not that the neurons detect concepts in a way that is independent of the class label. This metric does not support the conclusion that 'the interpretable neurons behave as expected.' Please evaluate concept detection on concepts that are not identical to the class labels, or provide human-annotated concept-presence labels on held-out examples to measure alignment.
  2. [§3.1, Eq. (2)] Automatic Concept Correction (ACC) zeros out the concept score for every concept whose associated class differs from the sample's ground-truth class. As a result, the training target for each neuron is class-conditional: a neuron can achieve high similarity in Eq. (3) by learning a class-membership detector rather than the semantic content of its named concept. The human faithfulness study in Task 1 (Appendix A.7) only presents the top-activated samples for each neuron; since cross-class samples have zero target by construction, those top samples are almost guaranteed to come from the neuron's own class, so high ratings in Table 3 do not rule out class-membership coding. The NLI check for the 'overpriced' concept in §3.3 is encouraging but covers one concept only. To support the central interpretability claim, the paper should test neurons on samples from other classes (e.g., how strongly a 'sports' neuron activates on business or world news), or validate a random sample of neurons against concept-level human labels.
  3. [§3.2 Setup, Table 2, Tables 3-4] The TBM&C3M baseline is re-implemented with only 1,000 labeled samples per dataset (labeled by Llama3-8B-Instruct), because the original methods' per-sample LLM labeling is cost-prohibitive. This means the accuracy and faithfulness comparisons in Tables 2-4 are against a heavily data-restricted approximation of TBM and C3M, not against the methods as originally proposed. The text's claim that CB-LLMs 'significantly outperforming TBM&C3M' (Section 3.1, Step 3) is therefore not established on equal footing. Please train the baseline on the full training sets, or at least on matched subsets with the same number of labeled examples as used for CB-LLMs, and clearly state the training data size in the table captions.
  4. [Table 5, §4.2 'Generation Quality', §5] The conclusion that CB-LLMs 'achieve performance nearly on par with black-box LLMs (within 1% gap)' is not supported for generation quality. The perplexity values in Table 5 show large differences: 116.22 vs 84.70 on SST2 (about 37% worse) and 13.03 vs 6.62 on YelpP (about 97% worse), with AGnews also worse (18.25 vs 12.52). The text dismisses these differences by saying 'a small difference in perplexity would not affect generation quality,' but these are not small differences. Please report confidence intervals or a statistical test for perplexity, and revise the generation-quality claim accordingly; the accuracy and steerability results can still stand, but the 'nearly on par' phrasing should be qualified to reflect the actual generation quality gap.
minor comments (5)
  1. [§3.2] The heading 'Efficency' contains a typo; it should be 'Efficiency.'
  2. [Appendix A.7, Table 3] The human study description says a random baseline was evaluated for Task 1, but the random baseline numbers are not reported in Table 3; please include them for context, since they are essential for interpreting the absolute activation-faithfulness ratings.
  3. [Section 2, Related Work] The text says TBM requires multiple queries to GPT-4 for each text sample, but the re-implementation used Llama3-8B-Instruct on 1,000 samples; please clarify that the quoted cost and the implemented baseline are different.
  4. [Eq. (6)] The symbol '∥' is used for concatenation but is not defined in the text; please define it the first time it appears.
  5. [Eq. (3)] The 'cos cubed' similarity is mentioned but not defined; please provide the formula or a citation to the source (Oikarinen et al., 2023).

Circularity Check

2 steps flagged · score 6.0 of 10

Generation 'concept detection' restates the classification objective; classification concept targets are class-conditioned by ACC, so the faithfulness claim is partly built from the label.

  1. fitted input called prediction [Section 4.2, Eq. 5, Table 5]
    "We use the labels of these datasets as concept labels directly (e.g., for AGnews, the concepts will be world, sport, business, and technology news), which allow us to calculate the concept accuracy in Table 5. ... Concept detection involves identifying the concepts in the prompt by extracting the interpretable neurons with the highest activation in the CBL ... the accuracy is calculated as the proportion of correctly aligned cases."

    The concept loss in Eq. 5 is CE(f+CBL(...), y_c), and the setup paragraph defines y_c as the dataset class label. The 'concept detection accuracy' in Table 5 is therefore the test accuracy of a model trained with cross-entropy to predict those same class labels. Calling this 'concept detection' is renaming the classification objective: the reported metric does not test whether neurons encode human-interpretable concepts beyond the class labels that were used as training targets. The claim that interpretable neurons enable precise concept detection reduces, by construction, to the claim that the CBL classifies the training labels.

  2. self definitional [Section 3.1, Steps 3-4, Eq. 2-3; Section 3.2, Task 1]
    "SACC_c(x)_i = Sc(x)_i, if S_c(x)_i > 0, M(c_i)=y; 0 otherwise ... To force the k neurons in the CBL learn the concepts, we maximize the similarity between f_CBL(f_LM(x)) and S_ACC_c(x)."

    Eq. 2 defines the CBL training target for every concept neuron as zero unless the concept's associated class equals the sample's ground-truth label y. The neurons are therefore trained to reproduce a class-conditional target, so their activations are optimized to encode class membership rather than concept semantics independently. In the activation-faithfulness evaluation, workers see the top-activated samples for each neuron; because cross-class targets are zeroed out by Eq. 2, these samples are drawn almost entirely from the neuron's associated class, so part of the apparent concept alignment is manufactured by the label-conditioned target. The human ratings and the NLI-based unlearning check remain external evidence, which is why this is only a partial circularity.

full rationale

The generation experiment is the clearest circular step: concept labels are the dataset class labels, the CBL is trained with cross-entropy on those labels (Eq. 5), and then the test accuracy of that same classifier is reported as 'concept detection accuracy' (Table 5). That metric is the training objective under a new name rather than independent evidence of concept-level understanding. The classification pipeline is partially circular as well: ACC (Eq. 2) constructs the concept scoring target from the ground-truth class label, so the neuron activations that later serve as explanations are trained to encode class membership; the top-activated-sample faithfulness evaluation is therefore conditioned on the label by construction. However, the paper does include external validations — MTurk human faithfulness ratings (Task 1 and Task 2) and an NLI-based check of the 'overpriced' unlearning example — which give the central interpretability claim some independent content. No load-bearing self-citation chain was found; citations to the authors' prior work [9, 10, 16] are design borrowings (e.g., the cos-cubed similarity) rather than forced premises. Overall, the score reflects one prediction that reduces to its training objective and one target construction that builds the class label into the concept representation, but not a fully circular derivation.

Assumptions & free parameters 7 free parameters · 5 assumptions · 0 invented entities

No new physical or conceptual entities are introduced; the architecture components (CBL, unsupervised layer, adversarial classifier) are standard ML modules, and the "interpretable neurons" are ordinary neurons trained to match concept scores.

free parameters (7)
  • classification regularization weight λ = 0.0007
    Set by hand for the elastic-net penalty in Eq. (4); affects sparsity and accuracy.
  • elastic-net mixing α = 0.99
    Set by hand in Eq. (4); controls L1 vs L2 regularization.
  • concept set size k (classification) = 208 (SST2), 248 (YelpP), 216 (AGnews), 476 (DBpedia)
    Number of neurons in the CBL is determined by ChatGPT prompt output ("List 100..." or "List 50..."); not fitted but is a design choice affecting capacity.
  • unsupervised layer size u (generation) = not stated
    The width of the parallel unsupervised layer in the hybrid CBM is never reported; it is a critical capacity hyperparameter for the generation experiments.
  • intervention value for steering = 100
    Setting the target concept neuron to 100 and others to 0 in Section 4.2; a hand-chosen amplitude for the steerability evaluation.
  • nonnegative threshold in ACC = >0
    Eq. (2) keeps concept scores only if they are positive and belong to the true class; the 0 threshold is a design choice.
  • concept labels for generation = dataset class labels (2, 2, 4, 14)
    Section 4.2 uses dataset labels directly as concept targets for the CBL, which simplifies but trivializes concept detection.
assumptions (5)
  • domain assumption ChatGPT-generated concept lists are sufficient and appropriate for each classification task
    Section 3.1 Step 1: the concept set C is generated by prompting ChatGPT with a few in-context examples per class; if the generated list misses key concepts, the bottleneck cannot represent them.
  • domain assumption Sentence-embedding cosine similarity is a valid proxy for concept presence
    Eq. (1) defines concept scores as dot products of all-mpnet-base-v2 embeddings; the paper provides no quantitative validation that this similarity aligns with human judgments; ACC partially patches this with labels.
  • ad hoc to paper Ground-truth class labels can be used to correct concept scores (ACC)
    Eq. (2) zeroes out all concepts not belonging to the sample's class and all negative scores; this is a supervised correction that depends on label availability and biases the concept targets toward the class structure.
  • ad hoc to paper Dataset labels are sufficient as concept labels for the generation task
    Section 4.2 states "We use the labels of these datasets as concept labels directly"; this equates broad classes with concepts, which makes concept detection a classification measure.
  • domain assumption The adversarial training module fully removes concept information from the unsupervised layer
    Module 2 (Eqs. 7-8) is assumed to disentangle concept information; the paper does not report the adversarial classifier's final accuracy, so the degree of disentanglement is not directly verified.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Concept Bottleneck Large Language Models." pith.science (2026). https://pith.science/paper/M3TXNTXK

@misc{pith2026241207992,
  author       = {Pith},
  title        = {Pith review of: Concept Bottleneck Large Language Models},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/M3TXNTXK}},
  note         = {Machine review of arXiv:2412.07992}
}
read the original abstract

We introduce Concept Bottleneck Large Language Models (CB-LLMs), a novel framework for building inherently interpretable Large Language Models (LLMs). In contrast to traditional black-box LLMs that rely on limited post-hoc interpretations, CB-LLMs integrate intrinsic interpretability directly into the LLMs -- allowing accurate explanations with scalability and transparency. We build CB-LLMs for two essential NLP tasks: text classification and text generation. In text classification, CB-LLMs is competitive with, and at times outperforms, traditional black-box models while providing explicit and interpretable reasoning. For the more challenging task of text generation, interpretable neurons in CB-LLMs enable precise concept detection, controlled generation, and safer outputs. The embedded interpretability empowers users to transparently identify harmful content, steer model behavior, and unlearn undesired concepts -- significantly enhancing the safety, reliability, and trustworthiness of LLMs, which are critical capabilities notably absent in existing models. Our code is available at https://github.com/Trustworthy-ML-Lab/CB-LLMs.

Figures

Figures reproduced from arXiv: 2412.07992 by the authors.

Figure 1
Figure 1. The overview of CB-LLMs (classification). The pipeline consists of five steps: (1) Generate concept set via querying ChatGPT. (2) Automatically label the samples with sentence embedding models. (3) Fix the incorrect concept labels. (4) Train backbone LLM and CBL with the concept labels. (5) Train a linear layer on top of the CBL to make the class predictions. between the text x and concept cj . These entries represe… view at source ↗
Figure 2
Figure 2. An example of concept unlearning. This example is initially classified as negative due to [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗
Figure 3
Figure 3. There are two modules: CB-LLM training and adversarial training for disentangling. [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figures from the paper (14 more)
Figure 3
Figure 3. Figure 3: The overview of CB-LLMs (generation). The training has two modules: (1) the main module for concept and token learning, and (2) the ADV training module, which prevents the unsupervised layer from encoding concept-related information, improving steerability. Module 1: C…
Figure 4
Figure 4. Figure 4: An example of how neurons in CB-LLMs (generation) detect the concepts. A deeper color means higher neuron activations. Steerability. An interesting application of CB-LLMs (generation) is steering generation by inter￾vening the activations of the neurons in CBL, as thes…
Figure 5
Figure 5. Figure 5: An example of toxicity detection and successful steering the generation via [PITH_FULL_IMAGE:figures/full_fig_p010_5.png]
Figure 6
Figure 6. Figure 6: The visualization of how the interpretable neurons in CB-LLM trained with SST2 connect [PITH_FULL_IMAGE:figures/full_fig_p015_6.png]
Figure 7
Figure 7. Figure 7: The visualization of how the interpretable neurons in CB-LLM trained with Yelp connect [PITH_FULL_IMAGE:figures/full_fig_p015_7.png]
Figure 8
Figure 8. Figure 8: The visualization of how the interpretable neurons in CB-LLM trained with AGnews [PITH_FULL_IMAGE:figures/full_fig_p016_8.png]
Figure 9
Figure 9. Figure 9: The visualization of how the interpretable neurons in CB-LLM trained with DBpedia [PITH_FULL_IMAGE:figures/full_fig_p017_9.png]
Figure 10
Figure 10. Figure 10: Another example of concept unlearning. This example is initially classified as negative [PITH_FULL_IMAGE:figures/full_fig_p018_10.png]
Figure 11
Figure 11. Figure 11: The interface for task 1 — Activation faithfulness. [PITH_FULL_IMAGE:figures/full_fig_p030_11.png]
Figure 12
Figure 12. Figure 12: The interface for task 2 — Contribution faithfulness. [PITH_FULL_IMAGE:figures/full_fig_p031_12.png]
Figure 13
Figure 13. Figure 13: The visualization of how the interpretable neurons in CB-LLM trained with AGnews [PITH_FULL_IMAGE:figures/full_fig_p039_13.png]
Figure 14
Figure 14. Figure 14: The visualization of how the interpretable neurons in CB-LLM trained with DBpedia [PITH_FULL_IMAGE:figures/full_fig_p040_14.png]
Figure 15
Figure 15. Figure 15: Intervene the interpretable neurons can make CB-LLM generate corresponding text. [PITH_FULL_IMAGE:figures/full_fig_p041_15.png]
Figure 16
Figure 16. Figure 16: Another example of toxicity detection and reduction through steering the generation via [PITH_FULL_IMAGE:figures/full_fig_p041_16.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 3 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Multi-Domain Explainability of Preferences

    cs.CL 2025-05 conditional novelty 6.0 of 10

    A concept-discovery plus hierarchical regression pipeline explains human, LLM-judge, and reward-model preferences at local and global levels across eight domains.

  2. Wanting to Be Understood Explains the Meta-Problem of Consciousness

    q-bio.NC 2025-06 conditional novelty 5.0 of 10

    A social motivation to be understood, combined with the severe bandwidth limit of language, explains why conscious experience feels ineffable and why the hard problem of consciousness persists.

  3. Beyond Explainable AI (XAI): An Overdue Paradigm Shift and Post-XAI Research Directions

    cs.CY 2026-02 unverdicted novelty 4.0 of 10

    Current XAI methods for DNNs and LLMs rest on paradoxes and false assumptions that demand a paradigm shift to verification protocols, scientific foundations, context-aware design, and faithful model analysis rather th...

Reference graph

Works this paper leans on

135 extracted references · 80 canonical work pages · cited by 3 Pith papers

  1. [1]

    Llama 3 model card

    AI@Meta. Llama 3 model card. 2024. URL https://github.com/meta-llama/ llama3/blob/main/MODEL_CARD.md

    2024

  2. [2]

    Concept bottleneck generative models

    Aya Abdelsalam Ismail, Julius Adebayo, Héctor Corrada Bravo, Stephen Ra, and Kyunghyun Cho. Concept bottleneck generative models. InICLR, 2024

    2024

  3. [3]

    Classification with conceptual safeguards

    Hailey Joren, Charles Thomas Marx, and Berk Ustun. Classification with conceptual safeguards. InThe Twelfth International Conference on Learning Representations, 2024. URL https: //openreview.net/forum?id=t8cBsT9mcg

    2024

  4. [4]

    Cai, James Wexler, Fernanda B

    Been Kim, Martin Wattenberg, Justin Gilmer, Carrie J. Cai, James Wexler, Fernanda B. Viégas, and Rory Sayres. Interpretability beyond feature attribution: Quantitative testing with concept activation vectors (TCA V). InICML, 2018

    2018

  5. [5]

    Concept bottleneck models

    Pang Wei Koh, Thao Nguyen, Yew Siang Tang, Stephen Mussmann, Emma Pierson, Been Kim, and Percy Liang. Concept bottleneck models. InICML, 2020

    2020

  6. [6]

    Mendes, Sebastian Hellmann, Mohamed Morsey, Patrick van Kleef, S

    Jens Lehmann, Robert Isele, Max Jakob, Anja Jentzsch, Dimitris Kontokostas, Pablo N. Mendes, Sebastian Hellmann, Mohamed Morsey, Patrick van Kleef, S. Auer, and Christian Bizer. Dbpedia - a large-scale, multilingual knowledge base extracted from wikipedia.Semantic Web, 2015

    2015

  7. [7]

    Roberta: A robustly optimized BERT pretraining approach.CoRR, 2019

    Yinhan Liu, Myle Ott, Naman Goyal, Jingfei Du, Mandar Joshi, Danqi Chen, Omer Levy, Mike Lewis, Luke Zettlemoyer, and Veselin Stoyanov. Roberta: A robustly optimized BERT pretraining approach.CoRR, 2019

    2019

  8. [8]

    Interpretable-by-design text classification with iteratively generated concept bottleneck

    Josh Magnus Ludan, Qing Lyu, Yue Yang, Liam Dugan, Mark Yatskar, and Chris Callison- Burch. Interpretable-by-design text classification with iteratively generated concept bottleneck. CoRR, 2023

    2023

Show all 135 references
  1. [9]

    Oikarinen and Tsui-Wei Weng

    Tuomas P. Oikarinen and Tsui-Wei Weng. Clip-dissect: Automatic description of neuron representations in deep vision networks. InICLR, 2023

    2023

  2. [10]

    Oikarinen, Subhro Das, Lam M

    Tuomas P. Oikarinen, Subhro Das, Lam M. Nguyen, and Tsui-Wei Weng. Label-free concept bottleneck models. InICLR, 2023

    2023

  3. [11]

    Long Ouyang, Jeffrey Wu, Xu Jiang, Diogo Almeida, Carroll L. Wainwright, Pamela Mishkin, Chong Zhang, Sandhini Agarwal, Katarina Slama, Alex Ray, John Schulman, Jacob Hilton, Fraser Kelton, Luke Miller, Maddie Simens, Amanda Askell, Peter Welinder, Paul F. Christiano, Jan Leik...

    2022

  4. [12]

    Language models are unsupervised multitask learners

    Alec Radford, Jeff Wu, Rewon Child, David Luan, Dario Amodei, and Ilya Sutskever. Language models are unsupervised multitask learners. 2019

    2019

  5. [13]

    Learning transferable visual models from natural language supervision

    Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agar- wal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, Gretchen Krueger, and Ilya Sutskever. Learning transferable visual models from natural language supervision. InICML, 2021

    2021

  6. [14]

    Discover-then-name: Task- agnostic concept bottlenecks via automated concept discovery

    Sukrut Rao, Sweta Mahajan, Moritz Böhle, and Bernt Schiele. Discover-then-name: Task- agnostic concept bottlenecks via automated concept discovery. InECCV, 2024

    2024

  7. [15]

    Manning, Andrew Ng, and Christopher Potts

    Richard Socher, Alex Perelygin, Jean Wu, Jason Chuang, Christopher D. Manning, Andrew Ng, and Christopher Potts. Recursive deep models for semantic compositionality over a sentiment treebank. InEMNLP, 2013

    2013

  8. [16]

    Vlg-cbm: Training concept bottleneck models with vision-language guidance

    Divyansh Srivastava, Ge Yan, and Tsui-Wei Weng. Vlg-cbm: Training concept bottleneck models with vision-language guidance. InNeurIPS, 2024. 11 Published as a conference paper at ICLR 2025

    2024

  9. [17]

    Interpreting pretrained language models via concept bottlenecks.CoRR, 2023

    Zhen Tan, Lu Cheng, Song Wang, Yuan Bo, Jundong Li, and Huan Liu. Interpreting pretrained language models via concept bottlenecks.CoRR, 2023

    2023

  10. [18]

    Huggingface’s transformers: State-of-the-art natural language processing.CoRR, 2019

    Thomas Wolf, Lysandre Debut, Victor Sanh, Julien Chaumond, Clement Delangue, Anthony Moi, Pierric Cistac, Tim Rault, Rémi Louf, Morgan Funtowicz, and Jamie Brew. Huggingface’s transformers: State-of-the-art natural language processing.CoRR, 2019

    2019

  11. [19]

    Energy-based concept bottleneck models: Unifying prediction, concept intervention, and probabilistic interpretations

    Xinyue Xu, Yi Qin, Lu Mi, Hao Wang, and Xiaomeng Li. Energy-based concept bottleneck models: Unifying prediction, concept intervention, and probabilistic interpretations. InICLR, 2024

    2024

  12. [20]

    Language in a bottle: Language model guided concept bottlenecks for interpretable image classification

    Yue Yang, Artemis Panagopoulou, Shenghao Zhou, Daniel Jin, Chris Callison-Burch, and Mark Yatskar. Language in a bottle: Language model guided concept bottlenecks for interpretable image classification. InCVPR, 2023

    2023

  13. [21]

    Post-hoc concept bottleneck models

    Mert Yüksekgönül, Maggie Wang, and James Zou. Post-hoc concept bottleneck models. In ICLR, 2023

    2023

  14. [22]

    Unappetizing food

    Xiang Zhang, Junbo Jake Zhao, and Yann LeCun. Character-level convolutional networks for text classification. InNeurIPS, 2015. 12 Published as a conference paper at ICLR 2025 Table of Contents A Appendix: CBLLMs — classification case 13 A.1 Performance of CB-LLMs using GPT2 an...

    2015

  15. [23]

    the humor is hinged on the belief that knees in the crotch , elbows in the face and spit in the eye are inherently funny

  16. [24]

    it ’s laughing at us

  17. [25]

    but mostly the humor is of the sweet , gentle and oc- casionally cloying kind that has become an iranian specialty

    there are a few stabs at absurdist comedy ... but mostly the humor is of the sweet , gentle and oc- casionally cloying kind that has become an iranian specialty

  18. [26]

    occasionally funny , always very colorful and enjoy- ably overblown in the traditional almodóvar style

  19. [27]

    SST2 Neuron 170: Great chemistry between actors

    hilarious , acidic brit comedy . SST2 Neuron 170: Great chemistry between actors

  20. [28]

    hugh grant and sandra bullock are two such likeable actors

  21. [29]

    binoche and magimel are perfect in these roles

  22. [30]

    makes s&m seem very romantic , and maggie gyl- lenhaal is a delight

  23. [31]

    a star-making project

    hayek is stunning as frida and ... a star-making project

  24. [32]

    SST2 Neuron 34: Lack of humor or wit

    tim allen is great in his role but never hogs the scenes from his fellow cast , as there are plenty of laughs and good lines for everyone in this comedy . SST2 Neuron 34: Lack of humor or wit

  25. [33]

    frenetic but not really funny

  26. [34]

    there just is n’t much to laugh at

    beyond a handful of mildly amusing lines ... there just is n’t much to laugh at

  27. [35]

    but here ’s the real damn : it is n’t funny , either

  28. [36]

    do not , under any circumstances , consider taking a child younger than middle school age to this wallow in crude humor

  29. [37]

    19 Published as a conference paper at ICLR 2025 YelpP Neuron 184: Good breakfast options

    it ’s frustrating to see these guys – who are obviously pretty clever – waste their talent on parodies of things they probably thought were funniest when they were high . 19 Published as a conference paper at ICLR 2025 YelpP Neuron 184: Good breakfast options

    2025

  30. [38]

    Loved the breakfast! Protein Berry Pancakes and eggs!

  31. [39]

    I’m obsessed with the breakfast here. There’s a huge smorgasbord of options to choose from on the brekkie menu, and the hardest part is actually pick- ing something to order because they all sound so good! I couldn’t resist ordering the eggs benedicto. What a cute twist on you...

  32. [40]

    Half order of Mashed Potatoes Omelet and an ice tea is how everyone should start their day!

  33. [41]

    which is an egg white omelette with diced turkey, spinach, feta cheese, diced onions and tomatoes

    My last two breakfasts here I have ordered the ’Healthy Turkey’ .... which is an egg white omelette with diced turkey, spinach, feta cheese, diced onions and tomatoes. It is served with an english muffin and is very tasty! ... My husband continues to order his standard raisin ...

    2025

  34. [42]

    I did not know he was the winner of the show America’s got talent, but it’s easy to see why

    I absolutely loved the show. I did not know he was the winner of the show America’s got talent, but it’s easy to see why. He’s clever, funny, has a great voice and it’s astounding to see him perform and not move his mouth. However, and though I appreciated the sentiment, I cou...

  35. [43]

    If you’re a huge Beatles fan, you will love this show. If you’re a huge Cirque du Soleil fan, you might feel a lil’ bit disappointed? But I guarantee this, you will definitely appreciate the artistic value of the show and what it’s goal was..and that was to pay homage to one o...

  36. [44]

    So, when my husband wanted to go to Vegas for a couple of days, I bought tickets

    I love the Beatles and I loved Love! (...and all you really need is love...) I wanted to see Love for awhile. So, when my husband wanted to go to Vegas for a couple of days, I bought tickets. We were in the second section, which seemed perfect. But, as oth- ers have said, ther...

  37. [45]

    this was the 3rd cirque du soleii show they never dissapoint me

    this show was great!! if you love fire and acrobatic stuff you will love this show!! its good for fami- lies as well. this was the 3rd cirque du soleii show they never dissapoint me. the set was awesome and costumes!

  38. [46]

    The most impres- sive part though is the inanimate star of the show, the incredible stage

    This show was awesome! Complete with cool stunts, music, emotion and a great story. The most impres- sive part though is the inanimate star of the show, the incredible stage. It raises, lowers and pivots eleventy billion different directions and is quite the engineer- ing feat...

    2025

  39. [47]

    The best stores are Target and Kohls which says a lot

    Not at all impressed! The place is a maze - a con- densed outdoor mall with lots of cheap stores. The best stores are Target and Kohls which says a lot. Desert Ridge seems to be for teenagers or young moms. Difficult to find your way around the nar- row streets - no large dire...

  40. [48]

    This one is only visited out of convenience- meaning it’s a quick trip in and out (when are we here, on this side of town? when we go to my MIL’s house for dinner), but I don’t really like this one. I could probably blame the area as a whole- the Wal-Mart (really ghetto) and 9...

  41. [49]

    upscale" shops, but good luck finding anything with the pacs of shoppers looking to score

    This mall- eh It’s not horrible, but it’s a waste of time. I visited from out of town and it was not worth my while. The stores were your typical "upscale" shops, but good luck finding anything with the pacs of shoppers looking to score "deals". The only stores worth going to ...

  42. [50]

    very chic

    BORING...It’s one of those "very chic" shopping venues that is sterile and dull with all the same shops you can see at any high end mall. I’d rather walk around the TL in San Francisco. It’s more interest- ing

  43. [51]

    Having worked in super- markets before I’ve noticed that products you think would be in the same aisle are in a completely irrele- vant spot

    I gave this location such a low rating because the store is usually a mess. Having worked in super- markets before I’ve noticed that products you think would be in the same aisle are in a completely irrele- vant spot. Their shelves need to be reset in a better manner. 22 Publi...

    2025

  44. [52]

    He was 41

    Ken Caminiti, 1996 NL MVP, Dies at Age 41 NEW YORK - Ken Caminiti, the 1996 National League MVP who later admitted using steroids during his major league career, died Sunday. He was 41

    1996

  45. [53]

    302, Baker Wins No

    Maddux Wins No. 302, Baker Wins No. 1,000 Greg Maddux pitched the Chicago Cubs into the lead in the NL wild-card race and gave Dusty Baker a win to remember. Maddux threw seven shutout innings for his 302nd career win, Baker got his 1,000th victory as a manager and Chicago bea...

  46. [54]

    They ended that ignominious streak Sunday in their first opportunity of the season, beating the Chicago Bears 20-16 at Soldier Field

    At Last, Success on the Road for Lions The Detroit Lions went three full seasons without winning an away game, setting an NFL record for road futility. They ended that ignominious streak Sunday in their first opportunity of the season, beating the Chicago Bears 20-16 at Soldier Field

  47. [55]

    Open NEW YORK - Lindsay Davenport’s summer of success stayed on course Thursday when the fifth-seeded former U.S

    Davenport Advances at U.S. Open NEW YORK - Lindsay Davenport’s summer of success stayed on course Thursday when the fifth-seeded former U.S. Open champion defeated Arantxa Parra Santonja 6-4, 6-2 and advanced to the third round of the season’s final Grand Slam event

  48. [56]

    Five men broke 10 seconds in qualifying Saturday, led by Crawford’s time of 9.89

    Men Set for Sizzling Duel in 100 Meters ATHENS, Greece - The preliminaries in the 100 meters were perhaps just a sample of what’s to come Sunday, when a talented group of qualifiers - including Amer- icans Shawn Crawford, Justin Gatlin and defend- ing champion Maurice Greene -...

    2025

  49. [57]

    Lynndie R

    England’s Lawyers Try to Get Photos Thrown Out Lawyers for Pfc. Lynndie R. England sought Wednesday to throw out evidence at the heart of the Abu Ghraib prison scandal – the now-infamous photos showing her smiling and pointing at naked Iraqi detainees

  50. [58]

    Mr Anwar was freed from jail on Thursday, after a conviction for sodomy was quashed by a Malaysian court

    Anwar launches bid to clear name Lawyers for An- war Ibrahim, the former deputy prime minister of Malaysia, have launched a bid to clear his name. Mr Anwar was freed from jail on Thursday, after a conviction for sodomy was quashed by a Malaysian court

  51. [59]

    Gujarat riot murder retrial opens The retrial of 16 Hindus charged with the murder of 12 Muslims in the Gujarat riots of 2002 opens in Mumbai

    2002

  52. [60]

    26, 2004 - In a dramatic turn that silenced defense lawyers, a Yemeni poet accused of crafting terrorist propaganda argued on Thursday to represent himself before a US

    Yemeni Poet Says He Is al-Qaida Member GUAN- TANAMO BAY NA V AL BASE, Cuba Aug. 26, 2004 - In a dramatic turn that silenced defense lawyers, a Yemeni poet accused of crafting terrorist propaganda argued on Thursday to represent himself before a US

    2004

  53. [61]

    AGnews Neuron 10: terrorism and se- curity threats

    Terreblanche challenges SA arrest White supremacist Eugene Terreblanche is detained after allegedly breaking the terms of his parole. AGnews Neuron 10: terrorism and se- curity threats

  54. [62]

    Thaksin in the Firing Line After Massacre BANGKOK/JEDDAH, 29 October 2004 - A bomb ripped through two bars in southern Thailand yester- day, killing two people and wounding about 20, in what could be the first reaction to the deaths of 78 Muslims in police custody this week

    2004

  55. [63]

    Seven suspected terrorists arrested in Spain Spain’s Interior Minister says police have broken up a radical Muslim cell, plotting to bomb the country’s National Court

  56. [64]

    Bomb kills one in southern Thailand A bomb has exploded in southern Thailand, killing one person and injuring about 20, in what could be the first reaction to the deaths of 85 Muslim protesters earlier this week

  57. [65]

    Rebel Attacks Hit Baghdad as Rumsfeld Visits Iraq A rocket attack and suicide car bombing killed at least four people in Baghdad Sunday as Defense Secretary Donald Rumsfeld began an unannounced visit to Iraq to gauge efforts to calm violence before January elections

  58. [66]

    24 Published as a conference paper at ICLR 2025 DBpedia Neuron 174: words related to ship, car, train

    Suicide Car Bomber Hits Baghdad Checkpoint Again (Reuters) Reuters - A suicide car bomber struck an entrance to Baghdad’s Green Zone govern- ment compound Tuesday, 24 hours after an almost identical attack at the same checkpoint on the first anniversary of Saddam Hussein’s arr...

    2025

  59. [67]

    Chase (1869-1937).Chase was launched 24 April 1943 by Norfolk Navy Yard; sponsored by Mrs

    USS Chase - Navy ArchivesUSS Chase (DE- 158/APD-54) a Buckley-class destroyer escort of the United States Navy was named in honor of Admi- ral Jehu V . Chase (1869-1937).Chase was launched 24 April 1943 by Norfolk Navy Yard; sponsored by Mrs. J. V . Chase ; and commissioned 18...

    1937

  60. [68]

    The third USS Warren was a sloop-of-war that served in the United States Navy from 1799 to 1801

  61. [69]

    USS Reuben James (DE-153) was a Buckley-class destroyer escort in the United States Navy. She was the second ship named for Reuben James a Boatswain’s Mate who distinguished himself fight- ing the Barbary pirates.Reuben James was laid down on 7 September 1942 at the Norfolk Na...

    1942

  62. [70]

    She fought at Trafalgar.The French 74-gun ship Swiftsure also took part in the battle

    HMS Swiftsure was a 74-gun third rate ship of the line of the Royal Navy launched from Bucklers Hard on 23 July 1804. She fought at Trafalgar.The French 74-gun ship Swiftsure also took part in the battle. She had originally been a British ship but was cap- tured by the French ...

  63. [71]

    The loss of the Bredenhof on her third voyage to the East Indies was meticulously recorded in the Dutch archives

    Bredenhof VOC Bredenhof was a Dutch East In- diaman transport ship that foundered on a reef 120 miles south of Mozambique and only 13 miles off the African coast near the Cape of Good Hope on 6 June 1753. The loss of the Bredenhof on her third voyage to the East Indies was met...

    2025

  64. [72]

    Rochelle Perts (born 20 March 1992) is a Dutch singer who rose to prominence after winning the fourth season of talent show X Factor on 10 June 2011

    1992

  65. [73]

    Theophilus Musa London (born February 23 1987) is a Trinidadian-born American rapper from Brooklyn New York City

    1987

  66. [74]

    Her parents are both Caribbean

    Miss Dominique [as she is generally known as] born Dominique Michalon September 7 1978 in Sarcelles France is a French singer and second-place finalist of the fourth edition of Nouvelle Star [based version of Pop Idol]. Her parents are both Caribbean

    1978

  67. [75]

    Patrick Nuo (born August 31 1982 in Canton of Lucerne) is a Swiss-Albanian recording artist and actor

    1982

  68. [76]

    April is an award-winning Australian pop/rock pioneer

    April Byron (real name April Elizabeth Dove Potts) was born March 22 1947 in Warburton Victoria Australia. April is an award-winning Australian pop/rock pioneer. DBpedia Neuron 469: the publisher and imprint of the work

    1947

  69. [77]

    The Sale & Altrincham Advertiser is a weekly free newspaper delivered to homes in Sale Altrin- cham Timperley Bowdon Partington and Hale in the Metropolitan Borough of Trafford in Greater Manch- ester England. Published every Thursday it is one of two sister MEN Media publicat...

    2010

  70. [78]

    The Enterprise is an afternoon daily newspaper pub- lished in Brockton Mass. It is considered a news- paper of record for Brockton and nearby towns in northern Bristol and Plymouth counties and southern Norfolk County.The Fuller-Thompson family owned The Enterprise for 115 yea...

    1996

  71. [79]

    state of New Jersey and is based in Newark

    The Star-Ledger is the largest circulated newspa- per in the U.S. state of New Jersey and is based in Newark

  72. [80]

    The Mercury is an upmarket English-language news- paper owned by Independent News & Media and published in Durban South Africa

  73. [81]

    Average Sunday circulation in September 2004 was 26747

    The Anniston Star is the daily newspaper serving Anniston Alabama and the surrounding six-county region. Average Sunday circulation in September 2004 was 26747. The newspaper is locally-owned by Consolidated Publishing Company which is con- trolled by the descendants of Col. H...

    2004

  74. [85]

    Unforgettable and heartwarm- ing moments

  75. [86]

    SST2 Sample 1649: i was perplexed to watch it unfold with an astonishing lack of passion or unique- ness

    Engaging character relation- ships. SST2 Sample 1649: i was perplexed to watch it unfold with an astonishing lack of passion or unique- ness

  76. [87]

    Poorly executed social com- mentary

  77. [88]

    Lack of believable conse- quences for character actions

  78. [89]

    Poorly executed voice-over narration

  79. [90]

    Unimpressive set design

  80. [91]

    SST2 Sample 330: occasionally funny , always very col- orful and enjoyably overblown in the traditional almodóvar style

    Excessive runtime. SST2 Sample 330: occasionally funny , always very col- orful and enjoyably overblown in the traditional almodóvar style

  81. [92]

    Stylish and unique costumes

  82. [93]

    Stellar and diverse ensemble cast

  83. [94]

    Charming and lovable side characters

  84. [95]

    Touching and heartfelt mo- ments

  85. [96]

    YelpP Sample 21864: These guys are money grubbing

    Stunning locations. YelpP Sample 21864: These guys are money grubbing. What WAS a $25 haircut just jumped up to a $32 haircut. It’s just a haircut for God’s sake! I’m going elsewhere

  86. [97]

    Inefficient payment systems

  87. [98]

    Low-quality materials used

  88. [99]

    YelpP Sample 34857: This place has something for everyone

    No valet service. YelpP Sample 34857: This place has something for everyone. My wife and I started going there out of convenience before attending a movie at the South Pointe. But then we continued going back because we liked the food and the staff is very helpful. This most r...

  89. [100]

    Responsive concierge service

  90. [101]

    Quiet and relaxing atmo- sphere

  91. [103]

    Seen it¨

    Clear signage for directions. 27 Published as a conference paper at ICLR 2025 YelpP Sample 10736: One of the few Cirque du Soleil that follow a story line, so if you are look- ing for a Cirque du Soleil show and a story this is the one to see. Although it strays a bit from the...

    2025

  92. [104]

    Engaging storytelling

  93. [105]

    Quick and easy setup

  94. [106]

    Thorough examinations

  95. [107]

    Interactive features. AGnews Sample 3058: Mobile phone network reaches last of China’s ethnic minorities (AFP) AFP - China has brought its mobile phone net- work to the last of its ethnic minority regions previously cut off from commu- nication with the outside world, state me...

  96. [108]

    telecommunications and 5G technology

  97. [109]

    tech giants and major industry players

  98. [110]

    consumer electronics and gad- gets

  99. [111]

    words related to technical de- vices

  100. [112]

    3D printing and additive man- ufacturing. AGnews Sample 6125: Icahn Takes The High River NEW YORK - Why has Carl Icahn set his sights on the relatively insignificant My- lan Laboratories, a generic drug com- pany with just $1.5 billion in sales and a $4.3 billion market cap?

  101. [113]

    company earnings and finan- cial results

  102. [114]

    initial public offerings (IPOs)

  103. [115]

    investment portfolio diversifi- cation

  104. [116]

    financial literacy and educa- tion programs

  105. [117]

    interest rates and central bank policies. AGnews Sample 1035: Orioles 8, Devil Rays 0 Javy Lopez drove in four runs, Daniel Cabrera be- came the first rookie to win 10 games this season, and the Baltimore Orioles held the Tampa Bay Devil Rays to two hits in an 8-0 victory

  106. [118]

    record-breaking performances

  107. [119]

    fan reactions and opinions

  108. [120]

    team rankings and standings

  109. [121]

    sports analytics and data- driven insights

  110. [122]

    28 Published as a conference paper at ICLR 2025 DBpedia Sample 52170: Narthecium is a genus of flowering plants

    sports science breakthroughs. 28 Published as a conference paper at ICLR 2025 DBpedia Sample 52170: Narthecium is a genus of flowering plants. This genus was traditionally treated as belonging to the family Liliaceae but the APG II system of 2003 placed it in the family Narthe...

    2025

  111. [123]

    The plant’s historical or cul- tural symbolism

  112. [124]

    The methods of cultivation and care for the plant

  113. [125]

    The plant’s method of repro- duction (e.g., seeds, spores, cuttings)

  114. [126]

    the genus or family of plant

  115. [127]

    DBpedia Sample 32678: Pemberton’s Headquarters also known as Willis-Cowan House is a two-story brick house that served as the head- quarters for Confederate General John C

    The plant’s contribution to bio- diversity. DBpedia Sample 32678: Pemberton’s Headquarters also known as Willis-Cowan House is a two-story brick house that served as the head- quarters for Confederate General John C. Pemberton during most of the 47 day siege of Vicksburg and t...

  116. [128]

    The architectural style of the building (e.g., Gothic, Modern, Colonial)

  117. [129]

    the location of the building

  118. [130]

    The building’s role in local or national history

  119. [131]

    The cultural or artistic signifi- cance of the building

  120. [132]

    DBpedia Sample 12750: Disma Fumagalli (born Inzago Septem- ber 8 1826 - died Milan March 9 1893) was an Italian composer and teacher of music

    The building’s awards or recognitions for design or preservation. DBpedia Sample 12750: Disma Fumagalli (born Inzago Septem- ber 8 1826 - died Milan March 9 1893) was an Italian composer and teacher of music. He was a graduate of the Milan Conservatory where he began teaching ...

  121. [133]

    the artist’s born date

  122. [134]

    The artist’s cultural signifi- cance

  123. [135]

    The artist’s enduring legacy

  124. [136]

    The artist’s unique artistic voice

  125. [137]

    29 Published as a conference paper at ICLR 2025 A.7 MTURK SURVEY DESIGN AND INTERFACE We perform the human evaluation through Amazon Mechanical Turk (MTurk)

    The artist’s famous collabora- tions. 29 Published as a conference paper at ICLR 2025 A.7 MTURK SURVEY DESIGN AND INTERFACE We perform the human evaluation through Amazon Mechanical Turk (MTurk). Each worker is paid 0.05$ per question and must sign a consent form to take the s...

    2025

  126. [138]

    Task 1 — Activation Faithfulness:In this task, workers will be presented with a neuron concept alongside the corresponding top 5 highly activated text samples. Workers need to provide a rating ranging from 1 (strongly disagree) to 5 (strongly agree) based on the agreement obse...

  127. [139]

    model 1 is clearly better

    Task 2 — Contribution Faithfulness.In this task, workers will be presented with explana- tions from two models for a text sample. The explanations are generated by showing the top 5 neuron concepts with the highest contribution to the prediction. Workers need to compare which ...

    2025

Pith tools

Reviewed August 11, 2026 · model on record in the stance chip above.