Pith. sign in

REVIEW 2 major objections 6 minor 62 references

Sympathetic Framing: Evaluating AI Alignment across Sociodemographic Groups

T0 review · 2 major / 6 minor · reviewed 2026-08-01 · deepseek-v4-flash

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

Pith's one-line read Large language models vary widely in how well they perceive the sympathy a news headline elicits, with the best model (GPT-5.2, ρ=0.789) closely tracking a representative UK reader panel and the weakest (Mistral, ρ=0.41) failing almost comp

desk verdict Solid, large-scale measurement of LLM sympathetic-framing alignment worth refereeing, but the GPT-4-selected stimulus half and missing human-human ceiling make model rankings provisional. read the letter →

arxiv 2607.27232 v1 pith:YVIWZZFM submitted 2026-07-22 cs.CL cs.AIcs.CYcs.LG

Haran Shani-Narkiss , Michael Fire , Oren Tsur This is my paper
classification cs.CLcs.AIcs.CYcs.LG
keywords LLMalignmentsympathyframingnewsdemographicsubgroupssyntheticpollingdifferentialemotionalperceptionpermutationtesting
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 asks whether AI systems grasp the emotional subtext of news framing, not just the facts. Using 216 news headlines about three conflicts, it compared how 3,011 representative UK readers and seven LLMs answered mirrored yes/no questions about whether each headline evoked sympathy for one side or the other. The central claim is that alignment with human sympathy judgments varies strongly by model, from very high (GPT-5.2, ρ=0.789) to moderate (Mistral, ρ=0.41), and that even the best models align differently with different demographic groups. A sympathetic reader would care because if LLMs are used to mediate news, a model that misreads sympathetic framing for certain audiences could silently distort how events are perceived. The paper also shows that aggregate scores hide topic-specific failures, such as Mistral's near-random performance on Israel-Palestine headlines while doing moderately well on the other two topics.

What carries the argument

The mirrored sympathy question: each headline is paired with two yes/no questions asking whether it creates sympathy toward each conflicting side. Human responses are aggregated into a positive-response ratio per question (a graded signal of how explicit or one-sided the sympathetic framing is), and each model is prompted the same question 100 times so its binary answers also become a ratio. Alignment is then the Spearman rank correlation between the human ratio vector and the model ratio vector, computed separately for demographic subgroups. Significance is assessed by permutation tests that relabel entire participants (preserving within-respondent dependencies), stratify by survey split, a

What would settle it

Take a fresh sample of headlines from the same three conflicts without any LLM-based screening, run the identical survey and the same 100-prompt model protocol, and recompute the Spearman correlations: if GPT-5.2's lead over Mistral shrinks or the age/education gaps vanish, the reported alignment values are an artifact of the GPT-4-selected stimulus set.

Watch

Extended reading notes

Core claim

The paper establishes that the alignment between LLMs and human emotional perception of news framing can be measured by correlating the proportion of 'yes' answers to mirrored sympathy questions ('Does this create sympathy toward side A?' / '...toward side B?'), with each model prompted 100 times per headline to obtain a graded score. Across the full set of 432 question-headline pairs, Spearman correlations with human response ratios ranged from 0.789 (GPT-5.2) to 0.41 (Mistral Large 2512), with Grok, GPT-4, Gemini, DeepSeek, and Claude in between. The leading models showed broadly stable alignment across demographic subgroups, yet statistically significant differences emerged: GPT-5.2 align

Load-bearing premise

The set of 'nuanced' headlines was selected by screening with GPT-4 and manually keeping headlines that elicited sympathy for both sides or varied across runs; if this curated half is unrepresentative of naturally occurring news framing, the measured alignment hierarchy and the demographic gap magnitudes may not generalize.

Editorial extensions

If this is right

  • If correct, LLM-based synthetic polling can complement human surveys for tracking emotional framing in the news, at least for questions of sympathy toward conflict sides.
  • Aggregate alignment scores are insufficient: a model can be broadly aligned yet systematically less aligned with older, less educated, less knowledgeable, or politically neutral readers, so evaluation should report subgroup-level correlations.
  • Model alignment is topic-dependent: a model's overall score can mask a complete collapse on one conflict (Mistral on Israel-Palestine, ρ=0.09) while performing adequately on others.
  • Repeated prompting converts binary LLM answers into a graded sympathy signal that tracks human proportions, making black-box LLMs usable as readers without fine-tuning.
  • The released dataset of 216 headlines with thousands of human responses and model scores can serve as a benchmark for future evaluations of framing comprehension.

Reading between the lines

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

  • If news recommendations or summaries are generated by a model with a demographic-specific sympathy blind spot, the effect could be a quiet distortion of which stories feel sympathetic to which audiences, potentially amplifying polarization without any explicit political bias.
  • The mirrored-question design could be extended to other emotions (fear, anger, hope) and to non-English or non-Western panels; the UK-specific gaps observed here might be larger in cultures farther from the models' training distribution.
  • The finding that alignment is lowest among readers who know nothing about a conflict may reflect a training-distribution bias toward informed, educated prose; if so, prompting the model to adopt a low-knowledge reader's perspective could be a direct test of that mechanism.
  • The subgroup differences are computed on question-level proportions; reporting per-headline residuals would reveal whether the age/education gaps come from a few items where groups sharply diverge or from a diffuse shift across all headlines.
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.

Referee Report

2 major / 6 minor

Summary. The paper evaluates whether seven LLMs align with human readers' perception of sympathetic framing in news headlines. Using 3,011 representative UK respondents (YouGov) and 216 headlines (72 each on the Russia–Ukraine war, the Israel–Gaza war, and the 2024 US presidential campaign), the authors compute Spearman correlations between question-level human positive-response rates and model positive-response rates (obtained via 100 repeated prompts). They report a model hierarchy from GPT-5.2 (ρ=0.79) to Mistral Large 2512 (ρ=0.41), with intermediate values for Grok, GPT-4, Gemini, DeepSeek, and Claude. They further examine alignment with demographic subgroups (gender, age, education, social grade, language, political awareness, topic knowledge, and prior stance), finding generally high alignment for the leading model but statistically significant differences for age, education, political awareness, knowledge, and view intensity. The permutation-testing framework in SM4 is carefully specified, with stratified, per-participant relabeling and family-wise corrections.

Significance. If the results hold, this is a valuable large-scale benchmark for a relatively understudied aspect of LLM evaluation: emotional framing perception rather than factual accuracy or explicit bias. The use of a representative national sample, the mirrored-question design, the large number of responses per item, and the permutation-based significance testing for subgroup differences are genuine strengths. The finding that aggregate alignment can mask subgroup- and topic-specific variation is important for pluralistic-alignment discussions. However, the main empirical claim rests on a headline set half of which was selected using GPT-4's sympathy judgments, and the paper does not yet demonstrate that the headline-level results are robust to that construction choice. If the authors can provide the requested split-half and robustness analyses, the contribution would be solid.

major comments (2)
  1. [SM1, headline selection; main results, Fig. 2a] SM1, headline selection; main results, Fig. 2a
  2. [Results, 'Models' Alignment with the General Population'; Fig. 2a] Results, Models' Alignment with the General Population; Fig. 2a
minor comments (6)
  1. [SM4, §4.1] Typo: 'absulute' should be 'absolute.'
  2. [SM5, 'Alignment across Topics'] The sentence 'coercion rates were negligible' is unclear; presumably 'refusal rates' or 'coherence rates' was intended.
  3. [General] The exact prompt template and sampling parameters (temperature, top-p, API versions) used for the seven models are not given. Please provide the full prompt and inference settings in the supplementary material for reproducibility.
  4. [Data availability] The paper states that the survey data can serve as a benchmark, but I did not find a persistent data/code availability statement. The Google Sheet link for headlines is mentioned, but the raw human responses, model outputs, and analysis code should be deposited in a permanent repository.
  5. [Abstract and Results] Minor typos: 'predesposition' in Results; '𝝆=0.4' in the abstract should be '0.41' for consistency; 'inline with studies' in the Discussion should be 'in line with studies.'
  6. [Figure 3 caption] The caption says 'Subplots show (A) Gender, (B) Education, and (C) Age group classifications,' but the figure contains many more subplots. Please make the caption consistent with the figure.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the alignment metric is computed directly from independent human and model response rates; the GPT-4-based stimulus selection is a generalizability concern, not a derivation step.

full rationale

The paper's central quantity is Spearman's rho between human yes-proportions R^H and model yes-proportions R^M, computed from the same questions. There is no fitted parameter renamed as a prediction, no claimed derivation of one quantity from another, and no load-bearing self-citation. The only potentially self-referential element is that GPT-4 was used in Supplementary Material 1 to identify 'nuanced' headlines (retained if they produced sympathetic responses for both sides or run-to-run variability), and GPT-4/GPT-5.2 are among the evaluated models. However, the selection criterion is not the alignment outcome: it does not use human responses, and it does not enter the Spearman computation. R^H and R^M are independently measured for the selected headlines, so the reported correlations are not forced by construction. This is a stimulus-construction choice that could affect generalizability across headline populations, but it does not make the measurement circular. The permutation tests hold R^M fixed and relabel human respondents, which is an appropriate null. No self-citation is load-bearing (the one author self-citation, ref. 28, is background framing methodology). Therefore the paper does not exhibit the specified forms of circularity.

Assumptions & free parameters 3 free parameters · 6 assumptions · 0 invented entities

The paper fits no parameters; its 'free parameters' are hand-set design constants (number of repeated prompts, headline set composition, permutation count) that affect the precision and scope of the measured correlations. The main assumptions are that aggregate Spearman correlation operationalizes alignment, that 100 repeated LLM prompts give a stable model judgment, that the GPT-4-selected nuanced headlines do not bias the evaluation, and that the YouGov panel remains representative after exclusions. No invented entities are introduced.

free parameters (3)
  • n_repeated_prompts_per_question = 100
    Each LLM question-level positive ratio r_i^M is estimated from 100 repeated prompts; this hand-set sample size controls Monte Carlo noise in the model-side ratios but is not fitted to the human data.
  • headline_set_composition_per_topic = 36 random + 36 GPT-4-curated
    The final 72 headlines per topic mix random and curated items; the curated half was chosen by GPT-4's sympathy responses, which can bias measured alignment.
  • permutation_count = 10^5
    Number of permutations used for p-values; standard choice, not fitted.
assumptions (6)
  • domain assumption The Spearman correlation between aggregate human positive-response ratios and model positive-response ratios is a valid operationalization of 'alignment'.
    Invoked in 'Measuring Alignment'; the entire paper interprets rho as alignment. This assumes a monotonic correspondence between group-level ratios captures perception similarity.
  • domain assumption A model's perception can be represented by the distribution of 100 repeated yes/no answers.
    Footnote 6 says repeated prompting is needed due to stochasticity; this treats the empirical binomial proportion as the model's sympathy score.
  • ad hoc to paper The GPT-4-curated 'nuanced' headline subset does not systematically distort the measured alignment of the evaluated models.
    SM1 describes selecting 108 headlines via GPT-4's sympathy screening. If this selection is unrepresentative or favors OpenAI models, the central alignment hierarchy is biased.
  • domain assumption The YouGov panel is representative of UK adults after excluding 1,308 incomplete or inattentive responders.
    SM2 reports n=3,011 valid responses; exclusion may make the sample less representative, affecting subgroup alignment estimates.
  • standard math Permutation testing with fixed R_M and split-stratified relabeling yields valid p-values for the exchangeability null.
    SM4 defines the permutation procedure; standard but relies on the assumption that within-split respondents are exchangeable under the null.
  • domain assumption GDELT keyword-filtered headlines adequately represent news coverage of the three conflicts.
    SM1 describes retrieval with selected keywords; the choice of keywords and temporal windows constrains the stimulus set.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Sympathetic Framing: Evaluating AI Alignment across Sociodemographic Groups." pith.science (2026). https://pith.science/paper/YVIWZZFM

@misc{pith2026260727232,
  author       = {Pith},
  title        = {Pith review of: Sympathetic Framing: Evaluating AI Alignment across Sociodemographic Groups},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/YVIWZZFM}},
  note         = {Machine review of arXiv:2607.27232}
}
read the original abstract

Large Language Models (LLMs) are increasingly shaping how we consume information and form our worldview. This raises concerns beyond bias in AI: do LLMs grasp the emotional nuances conveyed via textual framing? In this work, we empirically evaluate how well an array of LLMs aligns with human emotional perception. Considering news headlines covering political and geopolitical conflicts, both human participants (n = 3011, a representative sample of the U.K. adult population, via a YouGov survey) and seven LLMs answered whether headlines evoked sympathy for a specified side in a conflict. We find that the correlation between AI and human evaluations varies across models, ranging from very high (0.789, GPT-5.2) to medium (0.4 ,Mistral Large 2512). Crucially, the leading models are broadly aligned with human judgments across all demographic subgroups, including age, gender, level of education, prior geopolitical knowledge, and participants' predispositions regarding the conflict, although there are statistically significant differences between groups. This research, with its robust design and large, demographically diverse dataset, offers the most comprehensive evaluation of LLMs' comprehension of news framing to date. Findings highlight an important, often-ignored aspect of differential alignment: even when aggregate performance is high, AI alignment is not universal -- it may correspond differently with demographic features and cultural norms. Considering or ignoring the need for differential alignment may therefore have significant implications for the development of ethical and useful AI systems.

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

62 extracted references · 2 canonical work pages

  1. [1]

    T. Zhang,et al., Benchmarking Large Language Models for News Summarization.Transactions of the Association for Computational Linguistics12, 39–57 (2024), doi:10.1162/tacl a 00632, https://aclanthology.org/2024.tacl-1.3/

    doi:10.1162/tacl 2024

  2. [2]

    Joren,et al., Sufficient context: A new lens on retrieval augmented generation systems, in International Conference on Learning Representations(2025), pp

    H. Joren,et al., Sufficient context: A new lens on retrieval augmented generation systems, in International Conference on Learning Representations(2025), pp. 20310–20334

    2025

  3. [3]

    Liang,et al., The widespread adoption of large language model-assisted writing across society.Patterns6(12) (2025)

    W. Liang,et al., The widespread adoption of large language model-assisted writing across society.Patterns6(12) (2025)

    2025

  4. [4]

    S. Lake, OpenAI CEO Sam Altman says Gen Z, millennials use Chat- GPT ’like a life adviser’ (2025),https://fortune.com/2025/05/13/ openai-ceo-sam-altman-says-gen-z-millennials-use-chatgpt-like-life-adviser/

    2025

  5. [5]

    A. D. Kramer, J. E. Guillory, J. T. Hancock, Experimental evidence of massive-scale emotional contagion through social networks.Proceedings of the National Academy of Sciences111(24), 8788–8790 (2014). 10

    2014

  6. [6]

    Bakshy, S

    E. Bakshy, S. Messing, L. A. Adamic, Exposure to ideologically diverse news and opinion on Facebook.Science348(6239), 1130–1132 (2015), doi:10.1126/science.aaa1160,https: //www.science.org/doi/abs/10.1126/science.aaa1160

    doi:10.1126/science.aaa1160 2015

  7. [7]

    A. M. Guess, P. Barber ´a, S. Munzert, J. Yang, The consequences of online partisan media. Proceedings of the National Academy of Sciences118(14), e2013464118 (2021)

    2021

  8. [8]

    S. Gonz ´alez-Bail´on,et al., Asymmetric ideological segregation in exposure to political news on Facebook.Science381(6656), 392–398 (2023), doi:10.1126/science.ade7138,https: //www.science.org/doi/abs/10.1126/science.ade7138

    work page doi:10.1126/science.ade7138 2023

Show all 62 references
  1. [9]

    Repke, F

    T. Repke, F. M¨ uller-Hansen, E. Cox, J. C. Minx, Attention and positive sentiments towards carbon dioxide removal have grown on social media over the past decade.Communications Earth & Environment5(1), 1–11 (2024)

    2024

  2. [10]

    Gehman, S

    S. Gehman, S. Gururangan, M. Sap, Y. Choi, N. A. Smith, RealToxicityPrompts: Evaluating Neural Toxic Degeneration in Language Models, inFindings of the Association for Computa- tional Linguistics: EMNLP 2020(2020), pp. 3356–3369

    2020

  3. [11]

    Y. Li, M. Du, R. Song, X. Wang, Y. Wang, A Survey on Fairness in Large Language Models. CoRR(2023)

    2023

  4. [12]

    Koo,et al., Benchmarking Cognitive Biases in Large Language Models as Evaluators, in Findings of the Association for Computational Linguistics ACL 2024(2024), pp

    R. Koo,et al., Benchmarking Cognitive Biases in Large Language Models as Evaluators, in Findings of the Association for Computational Linguistics ACL 2024(2024), pp. 517–545

    2024

  5. [13]

    I. O. Gallegos,et al., Bias and fairness in large language models: A survey.Computational Linguisticspp. 1–79 (2024)

    2024

  6. [14]

    Biever, ChatGPT broke the Turing test-the race is on for new ways to assess AI.Nature 619(7971), 686–689 (2023)

    C. Biever, ChatGPT broke the Turing test-the race is on for new ways to assess AI.Nature 619(7971), 686–689 (2023)

    2023

  7. [15]

    Suzgun,et al., Language models cannot reliably distinguish belief from knowledge and fact.Nature Machine Intelligence(2025)

    M. Suzgun,et al., Language models cannot reliably distinguish belief from knowledge and fact.Nature Machine Intelligence(2025)

    2025

  8. [16]

    Ouyang,et al., Training language models to follow instructions with human feedback

    L. Ouyang,et al., Training language models to follow instructions with human feedback. Advances in neural information processing systems35, 27730–27744 (2022). 11

    2022

  9. [17]

    A. Wei, N. Haghtalab, J. Steinhardt, Jailbroken: How does llm safety training fail?Advances in Neural Information Processing Systems36, 80079–80110 (2023)

    2023

  10. [18]

    Gabriel,et al., The Ethics of Advanced AI Assistants.CoRR(2024)

    I. Gabriel,et al., The Ethics of Advanced AI Assistants.CoRR(2024)

    2024

  11. [19]

    Rastogi,et al., Whose view of safety? A deep DIVE dataset for pluralistic alignment of text-to-image models.Advances in Neural Information Processing Systems38(2026)

    C. Rastogi,et al., Whose view of safety? A deep DIVE dataset for pluralistic alignment of text-to-image models.Advances in Neural Information Processing Systems38(2026)

    2026

  12. [20]

    J. Ji,et al., Language Models Resist Alignment: Evidence From Data Compression, in Proceedings of the 63rd Annual Meeting of the Association for Computational Linguis- tics (Volume 1: Long Papers), W. Che, J. Nabende, E. Shutova, M. T. Pilehvar, Eds. (As- sociation for Computa...

    2025 doi

  13. [21]

    L. Sun, C. Mao, V. Hofmann, X. Bai, Aligned but Blind: Alignment Increases Implicit Bias by Reducing Awareness of Race, inProceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), W. Che, J. Nabende, E. Shutova, M. T. Pile...

    2025 doi

  14. [22]

    Goffman,Frame analysis: An essay on the organization of experience(Harvard University Press) (1974)

    E. Goffman,Frame analysis: An essay on the organization of experience(Harvard University Press) (1974)

    1974

  15. [23]

    R. M. Entman, Framing: Toward clarification of a fractured paradigm.Journal of communica- tion43(4), 51–58 (1993)

    1993

  16. [24]

    T. A. van Dijk,Racism and the Press, Critical Studies in Racism and Migration (Routledge, London) (1991)

    1991

  17. [25]

    H. H. Clark, M. F. Schober, Asking questions and influencing answers. (1992)

    1992

  18. [26]

    D. A. Scheufele, D. Tewksbury, Framing, agenda setting, and priming: The evolution of three media effects models.Journal of communication57(1), 9–20 (2007)

    2007

  19. [27]

    W. A. Gamson, NEWS AS FRAMING.American Behavioral Scientist33(2), 157–161 (1989). 12

    1989

  20. [28]

    O. Tsur, D. Calacci, D. Lazer, A frame of mind: Using statistical models for detection of framing and agenda setting campaigns, inProceedings of the 53rd Annual Meeting of the Association for Computational Linguistics and the 7th International Joint Conference on Natural Langu...

    2015

  21. [29]

    Banks, E

    A. Banks, E. Calvo, D. Karol, S. Telhami, # polarizedfeeds: Three experiments on polarization, framing, and social media.The International Journal of Press/Politics26(3), 609–634 (2021)

    2021

  22. [30]

    Lorenz-Spreen, L

    P. Lorenz-Spreen, L. Oswald, S. Lewandowsky, R. Hertwig, A systematic review of worldwide causal and correlational evidence on digital media and democracy.Nature human behaviour 7(1), 74–101 (2023)

    2023

  23. [31]

    J. C. de Winter, Can ChatGPT pass high school exams on English language comprehension? International Journal of Artificial Intelligence in Educationpp. 1–16 (2023)

    2023

  24. [32]

    M. T. R. Laskar,et al., A Systematic Study and Comprehensive Evaluation of ChatGPT on Benchmark Datasets, inFindings of the Association for Computational Linguistics: ACL 2023 (2023), pp. 431–469

    2023

  25. [33]

    Otmakhova, S

    Y. Otmakhova, S. Khanehzar, L. Frermann, Media Framing: A typology and Survey of Computational Approaches Across Disciplines, inProceedings of the 62nd Annual Meet- ing of the Association for Computational Linguistics (Volume 1: Long Papers), L.-W. Ku, A. Martins, V. Srikumar,...

    2024 doi

  26. [34]

    G. Lior, L. Nacchace, G. Stanovsky, WildFrame: Comparing Framing in Humans and LLMs on Naturally Occurring Texts.arXiv preprint arXiv:2502.17091(2025)

    2025

  27. [35]

    S. J. Westwood, J. Grimmer, A. B. Hall, Measuring perceived slant in large language models through user evaluations.Standford Business School(2025)

    2025

  28. [36]

    Leetaru, P

    K. Leetaru, P. A. Schrodt, Gdelt: Global data on events, location, and tone, 1979–2012, inISA annual convention(Citeseer), vol. 2 (2013), pp. 1–49. 13

    1979

  29. [37]

    Wang,et al., Self-Consistency Improves Chain of Thought Reasoning in Language Models, inThe Eleventh International Conference on Learning Representations(2023)

    X. Wang,et al., Self-Consistency Improves Chain of Thought Reasoning in Language Models, inThe Eleventh International Conference on Learning Representations(2023)

    2023

  30. [38]

    J. D. Evans,Straightforward statistics for the behavioral sciences.(Thomson Brooks/Cole Publishing Co) (1996)

    1996

  31. [39]

    Schober, C

    P. Schober, C. Boer, L. A. Schwarte, Correlation coefficients: appropriate use and interpretation. Anesthesia & analgesia126(5), 1763–1768 (2018)

    2018

  32. [40]

    Akoglu, User’s guide to correlation coefficients.Turkish journal of emergency medicine 18(3), 91–93 (2018)

    H. Akoglu, User’s guide to correlation coefficients.Turkish journal of emergency medicine 18(3), 91–93 (2018)

    2018

  33. [41]

    R. J. Tibshirani, B. Efron, An introduction to the bootstrap.Monographs on statistics and applied probability57(1), 1–436 (1993)

    1993

  34. [42]

    Good,Permutation tests: a practical guide to resampling methods for testing hypotheses (Springer Science & Business Media) (2013)

    P. Good,Permutation tests: a practical guide to resampling methods for testing hypotheses (Springer Science & Business Media) (2013)

    2013

  35. [43]

    C. A. Holt, S. P. Sullivan, Permutation tests for experimental data.Experimental economics 26(4), 775–812 (2023)

    2023

  36. [44]

    YouGov,Global Code of Conduct & Ethics, Tech. rep., YouGov PLC, 50 Featherstone Street, London, EC1Y 8RT, United Kingdom (2023),https://corporate.yougov.com/ documents/277/Global_Code_of_Conduct__Ethics_June_2022_Update_19.07.22. pdf

    2023

  37. [45]

    M. T. Boykoff, The cultural politics of climate change discourse in UK tabloids.Political geography27(5), 549–569 (2008)

    2008

  38. [46]

    D. Kotz, R. West, Explaining the social gradient in smoking cessation: it’s not in the trying, but in the succeeding.Tobacco control18(1), 43–46 (2009)

    2009

  39. [47]

    Yates, E

    S. Yates, E. Lockley, Social media and social class.American Behavioral Scientist62(9), 1291–1316 (2018). 14

    2018

  40. [48]

    Potter, S

    Y. Potter, S. Lai, J. Kim, J. Evans, D. Song, Hidden persuaders: Llms’ political leaning and their influence on voters, inProceedings of the 2024 Conference on Empirical Methods in Natural Language Processing(2024), pp. 4244–4275

    2024

  41. [49]

    Taubenfeld, Y

    A. Taubenfeld, Y. Dover, R. Reichart, A. Goldstein, Systematic biases in LLM simulations of debates, inProceedings of the 2024 conference on empirical methods in natural language processing(2024), pp. 251–267

    2024

  42. [50]

    Pastorino, N

    V. Pastorino, N. S. Moosavi, Frame In, Frame Out: Do LLMs Generate More Biased News Headlines than Humans? (2025),https://arxiv.org/abs/2505.05406

    2025 arXiv

  43. [51]

    Ziems,et al., Can Large Language Models Transform Computational Social Science? Computational Linguistics50(1), 237–291 (2024), doi:10.1162/coli a 00502

    C. Ziems,et al., Can Large Language Models Transform Computational Social Science? Computational Linguistics50(1), 237–291 (2024), doi:10.1162/coli a 00502

    2024 doi

  44. [52]

    Fanous,et al., SycEval: Evaluating LLM Sycophancy (2025),https://arxiv.org/abs/ 2502.08177

    A. Fanous,et al., SycEval: Evaluating LLM Sycophancy (2025),https://arxiv.org/abs/ 2502.08177

    2025

  45. [54]

    M. Sharma,et al., Towards Understanding Sycophancy in Language Models, inThe Twelfth International Conference on Learning Representations(2024),https://openreview.net/ forum?id=tvhaxkMKAn

    2024

  46. [55]

    Cheng,et al., Social sycophancy: A broader understanding of llm sycophancy.arXiv preprint arXiv:2505.13995(2025)

    M. Cheng,et al., Social sycophancy: A broader understanding of llm sycophancy.arXiv preprint arXiv:2505.13995(2025)

    2025 arXiv

  47. [56]

    M. V. Carro, Flattering to Deceive: The Impact of Sycophantic Behavior on User Trust in Large Language Model (2024),https://arxiv.org/abs/2412.02802

    2024 arXiv

  48. [57]

    M. D. Ward,et al., Comparing GDELT and ICEWS event data.Analysis21(1), 267–297 (2013)

    2013

  49. [58]

    H. Kwak, J. An, A first look at global news coverage of disasters by using the gdelt dataset, in International conference on social informatics(Springer) (2014), pp. 300–308. 15

    2014

  50. [59]

    Imran, C

    M. Imran, C. Castillo, F. Diaz, S. Vieweg, Processing social media messages in mass emergency: A survey.ACM computing surveys (CSUR)47(4), 1–38 (2015)

    2015

  51. [60]

    Raleigh, R

    C. Raleigh, R. Kishi, A. Linke, Political instability patterns are obscured by conflict dataset scope conditions, sources, and coding choices.Humanities and Social Sciences Communica- tions10(1), 74 (2023)

    2023

  52. [61]

    Mudassar, T

    M. Mudassar, T. Lin, X. Wang, H. Geng, Global trends and influential factors of climate change adaptation recognition in GDELT.Scientific Reports15(1), 26887 (2025)

    2025

  53. [62]

    Kamala”, “Harris

    Y. Hochberg, A. C. Tamhane,Multiple comparison procedures(John Wiley & Sons, Inc.) (1987). 16 1 Survey Questions Headline Retrieval from GDELT.The Global Database of Events, Language, and Tone (GDELT7) is a comprehensive open-access project that continuously monitors global ne...

    1987

  54. [63]

    Adjacent, extreme, and range statistics, and the corresponding𝑝-values𝑝 raw,𝑝 FWER,𝑝 range, are the same as defined above

    The independence across topics reflects the fact that each respondent’s three topic-specific levels are separate observations and may be relabeled independently under the null. Adjacent, extreme, and range statistics, and the corresponding𝑝-values𝑝 raw,𝑝 FWER,𝑝 range, are the ...

Pith tools

Reviewed August 1, 2026 · model on record in the stance chip above.