Pith. sign in

REVIEW 3 major objections 7 minor 2 cited by

Echoes in AI: Quantifying lack of plot diversity in LLM outputs

T0 review · 3 major / 7 minor · reviewed 2026-08-10 · deepseek-v4-flash

Pith's one-line read LLM-generated stories reuse the same plot elements far more often than human-written ones, and a new automatic metric measures that echo quantitatively.

desk verdict A promising new metric for plot-level diversity, but the headline human-vs-LLM gap leans on an entailment judgment that hasn't been validated for source bias. read the letter →

arxiv 2501.00273 v2 pith:XM4ME6UP submitted 2024-12-31 cs.CL

classification cs.CL
keywords SuiGenerisscoreplotdiversityLLMstorygenerationnarrativeuniquenesscollectivecreativityechosurpriseevaluationtextmetrics
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 argues that when a large language model is asked to continue the same story prompt many times, its continuations reuse the same narrative beats—plot elements that reappear across generations, often in different words and orders. To make that claim quantitative, the authors introduce the Sui Generis score, which measures how often a given story segment shows up, in spirit, in alternative continuations of the same prompt. Across 100 stories from two datasets, human-written segments score far higher (more unique) than segments produced by GPT-4 or LLaMA-3, and the effect persists when one model is used to score another model's output. A human study finds that the score's ranking of segments matches human surprise judgments moderately well (Spearman rho about 0.55), even though the score is computed automatically.

What carries the argument

The load-bearing machinery is the Sui Generis score $SG_i = -(\sum_{j=1}^{i-1} \lambda_j \log p_{i,j}) /(\sum_{j=1}^{i-1} \lambda_j)$, where $\lambda_j$ is an exponentially decaying weight controlled by a constant $\lambda<1$ (the experiments use $\lambda=0.9$). Here $p_{i,j}$ is an echo probability: over $K=20$ alternative continuations sampled after truncating the story at segment $j$, $p_{i,j}$ is the fraction in which a prompted GPT-4 entailment function $a(s_i, C_j^k)$ judges the plot of segment $i$ to be present. The earlier in the story an echo can be induced, the more heavily it is penalized, so a segment the LLM would spontaneously produce from a short prefix counts as less unique. This construction operates at the narrative level, independent of surface wording, and it also yields the drop ratio between consecutive segments that measures pacing.

What would settle it

Run the Sui Generis pipeline on a fresh set of prompts while replacing the GPT-4 entailment judge with human judgers on several hundred segments. If the average score gap between human- and LLM-written stories disappears or reverses, or if the automatic score's correlation with human surprise ratings becomes non-significant, the paper's central claim about echo-prone LLM plots would be falsified.

Watch

Extended reading notes

Core claim

The central discovery is that LLM-generated stories are largely assembled from plot elements that the same model—and even a different LLM—would readily generate again under the same prompt, while human-written stories contain turns that fall almost entirely outside the LLM's distribution of possible continuations. The paper formalizes this via the Sui Generis score, defined for a segment $s_i$ as a prefix-weighted average of negative log probabilities that $s_i$'s plot appears in alternative continuations $C_j$ sampled from the model after truncating the story at earlier positions $j < i$. The metric is fully automatic, using a prompt-based entailment judgment by GPT-4 for whether a plot segment is echoed. It reveals not only lower average scores for LLM output, but also a 'drop ratio' reflecting the rushed, unresolved pacing noted in prior qualitative studies.

Load-bearing premise

The whole score depends on GPT-4's prompt-based entailment judgment being accurate and unbiased for both human- and LLM-written segments, yet that judgment was validated on only 20 pairs with fair inter-annotator agreement (Fleiss kappa 0.33), and the same model family is used for generation and scoring.

Editorial extensions

If this is right

  • If the claim holds, the Sui Generis score gives a cheap, automatic way to measure whether an LLM is producing genuinely diverse plots, without needing human raters or external corpora.
  • The drop-ratio result links the metric to narrative pacing: LLM stories show abrupt peaks followed by plunges in uniqueness, matching expert descriptions of rushed, unresolved endings.
  • Because the echoes are cross-model, simply switching from one LLM to another will not restore narrative diversity; the tendency appears shared.
  • Prompt choice matters: stories generated under more evocative prompts score higher for both humans and LLMs, so prompt engineering can partly counteract homogenization.
  • The score can be used as a generation-time filter, selecting the most unique among many sampled continuations, at the cost of additional LLM calls.

Reading between the lines

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

  • By analogy with the paper's narrative test, the same prefix-truncation-and-echo procedure could quantify diversity in other sequential creative domains—music, animation, dialogue—where an entailment or similarity function can be defined; the authors note this only in the conclusion.
  • The finding implies that an individual user who sees just one LLM output may overestimate its novelty; the 'echo' is only visible when sampling many alternatives. This could be tested directly by measuring whether users' novelty ratings drop after they view multiple generations.
  • A practical extension would be to use the Sui Generis score as an inference-time reward in a rejection-sampling loop with a smaller entailment model, checking whether the diversity gains found with GPT-4 persist when the judge is weaker and cheaper.
  • The moderate correlation with human surprise suggests the score could serve as a proxy for 'interestingness' in co-writing tools, flagging segments that deserve human attention.
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

3 major / 7 minor

Summary. The paper introduces the Sui Generis (SG) score, a fully automatic metric that measures the uniqueness of plot segments in a story by sampling alternative continuations from an LLM and using GPT-4 to judge whether each segment's plot is echoed in those continuations. On 100 stories from WritingPrompts and Wikipedia plot summaries, the authors report that LLM-generated stories (GPT-4 and LLaMA-3) have significantly lower SG scores than human-written stories, that LLM plot elements are echoed across different models, that LLM stories exhibit higher drop ratios (suggesting abrupt pacing), and that SG scores correlate moderately with human surprisal judgments. The metric is positioned as a narrative-level alternative to lexical or embedding-based diversity measures.

Significance. If the SG score is valid, it offers a practical and automatic way to quantify one aspect of LLM output diversity, with direct implications for human-AI co-creation and the evaluation of generative models. The paper has several strengths: the metric is operationally defined without fitted parameters, the experimental setup is described in sufficient detail to reproduce, the human studies were IRB-approved, and code is made available. The reported correlations with human surprisal and the paraphrase-robustness experiment are informative. However, the headline human-vs-LLM comparison rests entirely on GPT-4's binary entailment judgments, and the validation of that judgment function is too thin to rule out a systematic source-dependent bias. Because the central claim depends on this, the current evidence does not yet fully support the paper's strongest conclusion.

major comments (3)
  1. [§3 Eq. (2) and §6 (Comparison with Other Similarity Metrics)] The GPT-4 entailment function a(s,C) is validated on only 20 pairs with a Fleiss kappa of 0.33, and the validation is not stratified by whether the source segment comes from a human-written or an LLM-generated story. Since the central comparison in Figure 3 contrasts human segments (scored against LLM continuations) with LLM segments (scored against LLM continuations), a systematic bias in GPT-4's judgments—such as higher sensitivity to echoes in LLM-style text or lower sensitivity to human-paraphrased echoes—would directly inflate the reported gap. The manuscript does not provide any test of source invariance. Please add a stratified human validation of a(s,C) separately for human and LLM source segments, or otherwise demonstrate that the entailment judgments are not biased by the origin of the segment.
  2. [§5 (Table 3) and §4B] The drop-ratio analysis uses the hyperparameters λ=0.9 and θ=0.5 without any sensitivity or robustness check. The reported 7–9 percentage-point differences between human and LLM stories in Table 3 are not accompanied by confidence intervals or a test of whether the ordering holds for a range of λ and θ. I would like to see a small sensitivity analysis (e.g., λ∈{0.8,0.9,0.95}, θ∈{0.3,0.5,0.7}) to confirm that the qualitative result is not an artifact of these particular thresholds.
  3. [§7 (Limitations)] The limitations section candidly notes that substituting a less capable model for entailment judgment could lead to bias, but it does not address the possibility that GPT-4 itself exhibits a source-dependent bias in this task. The 20-pair validation in Section 6 has low inter-annotator agreement (κ=0.33) and is too small to detect such a calibration offset. As a result, the main quantitative claim of the paper is not yet shielded from the plausible scorer-bias artifact described above.
minor comments (7)
  1. [Abstract] The phrase "the first automatic metric" is an overclaim given that other diversity metrics exist and are compared in Section 6; suggest softening to "an automatic metric".
  2. [§4C] The GPT-4 model version is not specified (e.g., GPT-4-0613 or GPT-4-1106-preview); please state the exact model and API version for reproducibility.
  3. [§5 (Cross-Model Scoring)] The cross-model comparison uses only 20 stories per dataset; the text says "Results on 20 Wiki and 20 WritingPrompts stories show..." but the heatmap in Figure 5 is for 20 Wiki stories. Please report standard errors or confidence intervals for the cross-model differences.
  4. [§5 (Sui Generis scores correlate with human judgment)] The human surprise study uses 9 stories (159 segments) from LLaMA-3 only; the claim that SG correlates with human judgment would be strengthened by testing on GPT-4-generated and human-written stories as well.
  5. [§6 (Comparison with Other Similarity Metrics)] The human judges were asked to give "yes, no, or partially" and the average was taken with "partially"=0.5; this coarse scale may compress variance. Please report agreement with a stricter criterion or use a finer scale.
  6. [Figure 4] The heatmap uses triangular matrices for the lower and upper halves, but the colorbar and axis labels are not explained in the caption; please clarify the orientation of the two triangles and what the transpose represents.
  7. [§5 (paraphrase robustness)] The paraphrase generation process and manual verification are not described in the main text; please add a short description or a reference to the appendix.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the Sui Generis score is defined operationally, and the human-vs-LLM gap is an empirical measurement, not a reduction to the metric's inputs.

full rationale

The paper defines the Sui Generis score operationally: Eq. 2 defines echo probability p_{i,j} via GPT-4's entailment judgment a(s_i, C_j^k), and Eq. 3 aggregates negative log echoes over prefixes. The central claim (human segments score higher than LLM segments, Figure 3) is a measurement outcome, not an identity: nothing in the definition forces human-written segments to be less echoed than LLM-generated ones, and the difference is shown empirically. The main potential concern, that GPT-4 serves as both generator and entailment judge so the score partly measures self-predictability, is a real validity limitation, acknowledged in Section 7, but it is not circular in the sense of fitting a parameter and then re-predicting it, nor does any load-bearing step reduce to a self-citation. The human evaluation (Section 6) provides independent, albeit small, grounding for the entailment function and the surprisal correlation. No fitted-parameter or self-citation chain is load-bearing. The small 20-pair human validation and the lack of source-invariance testing for entailment judgments are calibration concerns, not circularity.

Assumptions & free parameters 5 free parameters · 4 assumptions · 0 invented entities

The central claim relies on the validity of the GPT-4 entailment function, on a fixed segmentation scheme, and on settings of K and lambda that are chosen rather than derived. No new physical or conceptual entities are introduced; the Sui Generis score is a metric, not an entity.

free parameters (5)
  • lambda (weight decay) = 0.9
    Chosen in Section 4B to control exponential decay of prefix-length weighting in the Sui Generis score (Eq. 3). Not fitted to data; affects how strongly early echoes are penalized.
  • K (number of alternative continuations) = 20
    Section 4B; number of samples per prefix to estimate echo probabilities. Larger K would reduce variance but increase cost.
  • theta (drop ratio threshold) = 0.5
    Section 4B; threshold below which a score drop is not counted in the drop ratio (Eq. 4).
  • segment length = 50 words (WritingPrompts), 30 words (Wiki)
    Section 4A; fixed word-length segmentation is used to normalize across authors and LLMs; choice affects granularity of plot units.
  • entailment model settings = GPT-4, top_p=0.5, tau=0
    Section 4C; decoding settings for the function a(s,C). The paper states temperature 0 for deterministic judgments.
assumptions (4)
  • domain assumption The GPT-4 prompt template implements a human-equivalent plot entailment judgment.
    Section 3 states this and Section 6 validates on 20 pairs with Spearman 0.85, but Fleiss kappa among human annotators is only 0.33.
  • domain assumption Segmenting stories into fixed word counts (50 or 30 words) produces meaningful plot-level units.
    Section 4A; paragraph lengths vary between humans and LLMs, so fixed word segments are used to equalize information distribution.
  • domain assumption K=20 alternative continuations provide a stable estimate of the echo probability distribution.
    Section 4B; no sensitivity analysis is reported for K.
  • domain assumption The Sui Generis score is comparable between human and LLM stories under the same prompt/prefix distribution.
    The score uses LLM continuations conditioned on the story's prefix; for human stories, the human-written prefix is used. It is assumed this does not bias the comparison.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Echoes in AI: Quantifying lack of plot diversity in LLM outputs." pith.science (2026). https://pith.science/paper/XM4ME6UP

@misc{pith2026250100273,
  author       = {Pith},
  title        = {Pith review of: Echoes in AI: Quantifying lack of plot diversity in LLM outputs},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/XM4ME6UP}},
  note         = {Machine review of arXiv:2501.00273}
}
read the original abstract

With rapid advances in large language models (LLMs), there has been an increasing application of LLMs in creative content ideation and generation. A critical question emerges: can current LLMs provide ideas that are diverse enough to truly bolster collective creativity? We examine two state-of-the-art LLMs, GPT-4 and LLaMA-3, on story generation and discover that LLM-generated stories often consist of plot elements that are echoed across a number of generations. To quantify this phenomenon, we introduce the Sui Generis score, an automatic metric that measures the uniqueness of a plot element among alternative storylines generated using the same prompt under an LLM. Evaluating on 100 short stories, we find that LLM-generated stories often contain combinations of idiosyncratic plot elements echoed frequently across generations and across different LLMs, while plots from the original human-written stories are rarely recreated or even echoed in pieces. Moreover, our human evaluation shows that the ranking of Sui Generis scores among story segments correlates moderately with human judgment of surprise level, even though score computation is completely automatic without relying on human judgment.

Figures

Figures reproduced from arXiv: 2501.00273 by the authors.

Figure 1
Figure 1. Two continuations of the same story with (a) longer prefix, with j = 8 segments, and (b) shorter prefix, with j = 2 segments from the same original story, partially shown in (c). Segments are delineated with []. We highlight two segments in (c), i = 7 in blue and i = 9 in red, that are echoed in these alternative continuations. The red one (visiting a scientist’s lab) is echoed only in continuations conditioned on a… view at source ↗
Figure 2
Figure 2. Prompt template used for estimating function a(si, Cj ), i.e. if the plot in segment si is present in continuation Cj . The texts in black are part of the prompt while the texts in pink should be generated by the LLM. weighted more heavily. Thus, we compute the Sui Generis score of segment si by taking the weighted average of the negative log of echoes {− log pi,1, − log pi,2, ..., − log pi,i−1} where we give higher… view at source ↗
Figure 3
Figure 3. Average Sui Generis scores of story segments generated by humans versus GPT-4 (in (a) and (c)) or LLaMA-3 (in (b) and (d)) at varying segment positions (x-axis). The shaded area represents the confidence interval with confidence level γ = 0.95. the Echo rate increases as more of the previous plot is revealed (i.e., as the prefix length increases), while for human-written stories, the LLM can only occasionally predic… view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: Heatmap of the average Echo scores pi,j for the i-th story segment given story prefix S−j on 50 Wiki stories. The lower left triangle shows pi,j on LLM (GPT￾4 or LLaMA-3) generated stories, while the upper right triangle shows the transpose matrix of pi,j on human stor…
Figure 5
Figure 5. Figure 5: Heatmap of the average Echo scores pi,j for the i-th story segment given story prefix S−j on 20 Wiki stories by LLaMA-3. The lower left triangle shows the original Echo scores pi,j , while the upper right triangle shows the transpose matrix of cross-model Echo scores p…

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 2 Pith papers

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

  1. Correlated Errors in Large Language Models

    cs.CL 2025-06 conditional novelty 6.0 of 10

    Large language models from different providers and architectures often make the same errors, and more accurate models are especially likely to share mistakes.

  2. The AI Fiction Paradox

    cs.AI 2026-03 conditional novelty 5.0 of 10

    AI's inability to generate compelling long-form fiction stems from narrative causation, informational revaluation, and multi-scale emotional architecture—three constraints current transformer architectures lack.

Reference graph

Works this paper leans on

36 extracted references · 31 canonical work pages · cited by 2 Pith papers

  1. [1]

    Give it up!

    Introduction Rapid advances in large language models (LLMs) have spurred an ongoing debate on the usefulness of these models for tasks that require human-level creativity. On the one hand, there are works that highlight the strengths of LLMs in creative writing (1, 2), poetry generation (3), idea generation (4, 5) and even creative thinking (6). On the ot...

  2. [2]

    of its own kind,

    Related Work LLMs are increasingly used in creative writing ( 4, 13). However, there is a debate on whether LLMs boost creativity. Some studies have suggested that LLM-generated content is considered more creative or preferred by users. For example, ( 14) finds that ideas generated by ChatGPT were more likely to be purchased than those generated by Wharto...

  3. [3]

    alterna- tive continuations

    Method We evaluate the uniqueness of a story segment based on the “alterna- tive continuations” generated by LLMs themselves. Formally, in a story (string)S segmented inton segmentsS = (s1,s 2,...,s n) (as shown in Figure 1(c) with square brackets), at any point j ∈ [1,n− 1], we truncate the story to its prefixS−j = (s1,s 2,...,s j) and consider replacing...

  4. [4]

    Datasets

    Experimental Setup A. Datasets. We test our Sui Generis score on 100 stories from two story datasets: 1) the WritingPrompts dataset (21), which contains story prompts and the corresponding human-written stories posted on an online forum|| by the year of 2018, and 2) plot summaries of TV episodes in the year of 2023 crawled from Wikipedia. For WritingPromp...

  5. [5]

    LLMs We first compare the average Sui Generis scores of story segments produced by humans versus LLMs (including GPT-4 and LLaMA-3)

    Results Sui Generis Score: Human vs. LLMs We first compare the average Sui Generis scores of story segments produced by humans versus LLMs (including GPT-4 and LLaMA-3). As shown in Figure 3, for both datasets, human-written stories yield significantly higher Sui Generis scores than LLM-generated ones. We also show the average Echo score matrix pi,j on Wi...

  6. [6]

    We conduct a human evaluation in which we invite four human judges (who are fluent in English) to judge 20 randomly sampled pairs of plot segments

    Discussion Comparison with Other Similarity Metrics We compare our prompting-based plot entailment assessment with other similar- ity/diversity metrics by measuring how these metrics correlate with human judgments. We conduct a human evaluation in which we invite four human judges (who are fluent in English) to judge 20 randomly sampled pairs of plot segm...

  7. [7]

    Limitations The WritingPrompts corpus and part of the Wiki data used in the experiment may have been in the training corpora of the LLMs evaluated. Although our results show no effect of a memorization effect for the stories we tested, it is possible that this kind of training effect might surface for other datasets that were included in the LLMs’ trainin...

  8. [8]

    Our experiments on 100 LLM- versus human-written stories demonstrated the lack of plot-level diversity in LLM-generated stories

    Conclusion We introduced the Sui Generis score to quantify the uniqueness of LLM-generated stories at a narrative level when compared to their alternative generations. Our experiments on 100 LLM- versus human-written stories demonstrated the lack of plot-level diversity in LLM-generated stories. Regardless of the LLM used to generate them, these stories a...

Show all 36 references
  1. [9]

    A Bellemare-Pepin, et al., Divergent creativity in humans and large language models (2024)

  2. [10]

    W Orwig, ER Edenbaum, JD Greene, DL Schacter, The language of creativity: Evidence from humans and large language models. The J. creative behavior 58, 128–136 (2024)

  3. [11]

    B Porter, E Machery, Ai-generated poetry is indistinguishable from human-written poetry and is rated more favorably. Sci. Reports 14, 26133 (2024)

  4. [12]

    BC Lee, J Chung, An empirical investigation of the impact of chatgpt on creativity. Nat. Hum. Behav. 8, 1906–1914 (2024)

  5. [13]

    arXiv preprint arXiv:2409.04109 (2024)

    C Si, D Y ang, T Hashimoto, Can llms generate novel research ideas? a large-scale human study with 100+ nlp researchers. arXiv preprint arXiv:2409.04109 (2024)

  6. [14]

    EE Guzik, C Byrge, C Gilde, The originality of machines: AI takes the Torrance Test. J. Creat. 33, 100065 (2023)

  7. [15]

    (Association for Computing Machinery, New Y ork, NY , USA), (2024)

    T Chakrabarty, P Laban, D Agarwal, S Muresan, CS Wu, Art or artifice? large language models and the false promise of creativity in Proceedings of the CHI Conference on Human Factors in Computing Systems , CHI ’24. (Association for Computing Machinery, New Y ork, NY , USA), (2024)

  8. [16]

    The Verge (2023)

    M Sato, Ai-generated fiction is flooding literary magazines — but not fooling anyone. The Verge (2023)

  9. [17]

    The New Y ork Times(2023)

    M Levenson, Science fiction magazines battle a flood of chatbot-generated stories. The New Y ork Times(2023)

  10. [18]

    AR Doshi, OP Hauser, Generative ai enhances individual creativity but reduces the collective diversity of novel content. Sci. Adv. 10, eadn5290 (2024)

  11. [19]

    V Padmakumar, H He, Does writing with language models reduce content diversity? in The T welfth International Conference on Learning Representations. (2024)

  12. [20]

    B Mohammadi, Creativity has left the chat: The price of debiasing language models (2024)

  13. [21]

    D Kobak, R González-Márquez, E Ágnes Horvát, J Lause, Delving into chatgpt usage in academic writing through excess vocabulary (2024)

  14. [22]

    M Kefford, Wharton study pits chatgpt against mba students in creativity test. Bus. Because (2023)

  15. [23]

    N Begus, Experimental narratives: A comparison of human crowdsourced storytelling and ai storytelling (2023)

  16. [24]

    (Association for Computing Machinery, New Y ork, NY , USA), p

    BR Anderson, JH Shah, M Kreminski, Homogenization effects of large language models on human creative ideation in Proceedings of the 16th Conference on Creativity and Cognition . (Association for Computing Machinery, New Y ork, NY , USA), p. 413–425 (2024)

  17. [25]

    T ransactions Assoc

    RT McCoy, P Smolensky, T Linzen, J Gao, A Celikyilmaz, How much do language models copy from their training data? evaluating linguistic novelty in text generation using RAVEN. T ransactions Assoc. for Comput. Linguist. 11, 652–670 (2023)

  18. [26]

    C Shaib, et al., Standardizing the measurement of text diversity: A tool and a comparative analysis of scores (2024)

  19. [27]

    T Ghosal, T Saikh, T Biswas, A Ekbal, P Bhattacharyya, Novelty detection: A perspective from natural language processing. Comput. Linguist. 48, 77–117 (2022)

  20. [28]

    TF Jaeger, Redundancy and reduction: Speakers manage syntactic information density. Cogn. psychology 61, 23–62 (2010)

  21. [29]

    I Gurevych, Y Miyao

    A Fan, M Lewis, Y Dauphin, Hierarchical neural story generation in ACL:2018:1, eds. I Gurevych, Y Miyao. (Association for Computational Linguistics, Melbourne, Australia), pp. 889–898 (2018)

  22. [30]

    OpenAI, et al., Gpt-4 technical report (2024)

  23. [31]

    A Dubey, et al., The llama 3 herd of models (2024)

  24. [32]

    M Peeperkorn, T Kouwenhoven, D Brown, A Jordanous, Is temperature the creativity parameter of large language models? (2024)

  25. [33]

    The 41st Int

    Y Zhu, et al., Texygen: A benchmarking platform for text generation models. The 41st Int. ACM SIGIR Conf. on Res. & Dev. Inf. Retr . (2018)

  26. [34]

    PJ Wang, M Kreminski, Guiding and diversifying llm-based story generation via answer set programming (2024)

  27. [35]

    Futures 160, 103383 (2024)

    L Bojic, Ai alignment: Assessing the global impact of recommender systems. Futures 160, 103383 (2024)

  28. [36]

    Github (2025)

    W Xu, N Jojic, S Rao, C Brockett, B Dolan, Sui generis. Github (2025). 8 | www.pnas.org/cgi/doi/10.1073/pnas.2504966122 Xu et al

Pith tools

Reviewed August 10, 2026 · model on record in the stance chip above.