Pith. sign in

REVIEW 3 major objections 4 minor 29 references

Quantifying the Relationship Between Clinical Safety and Environmental Impact in Therapeutic LLMs

T0 review · 3 major / 4 minor · reviewed 2026-08-16 · deepseek-v4-flash

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

Pith's one-line read For therapeutic LLMs, near-maximal clinical safety scores come with disproportionately large estimated environmental footprints: roughly a 60-fold energy increase for a 2.61-point safety gain, while added test-time reasoning does not…

desk verdict A small, honest paper that quantifies the safety-environment trade-off in therapeutic LLMs with public data; the qualitative pattern is real, but the headline 60x figure rests on modeled, configuration-invariant estimates and needs a sensitivity bound before it can be quoted as a measurement. read the letter →

arxiv 2608.11830 v1 pith:2R3E5LFI submitted 2026-08-12 cs.CY cs.CL

Alireza A. Safaei , Laura M. Vowels , Matthew J. Vowels , Apoorv Jha , Shekoufeh Rahimi This is my paper · ORCID
classification cs.CYcs.CL
keywords clinicalAIsafetytherapeuticLLMsK-BenchEcoLogitslife-cycleassessmentsustainabletest-timecomputeParetofrontier
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 tries to establish that, among large language models proposed for therapeutic use, the last few points of clinical safety come at a disproportionate environmental price. By pairing K-Bench clinical safety scores with EcoLogits life-cycle estimates for 47 model configurations, the authors find that the highest-scoring model (gpt-5.5, 96.02) is estimated to use about 60 times more energy per million output tokens than a model scoring 2.61 points lower (claude-haiku-4.5, 93.41). They also find that adding test-time reasoning does not consistently improve safety scores across configurations. If true, this matters because it gives deployers a concrete reason to treat model choice as a multi-objective decision, reserving the largest models for high-risk cases rather than defaulting to them everywhere.

What carries the argument

The argument runs on two instruments and one analytic device. K-Bench is a transcript-based benchmark that evaluates therapist-style AI in simulated multi-turn mental health conversations: a factorial vignette generator creates risk scenarios across suicide, self-harm, domestic violence, and substance misuse, and a clinician-calibrated automated judge (Cohen's kappa of 0.850 on the calibration set) scores risk recognition and exploration on a 0-100 scale. EcoLogits is a Python package that converts model metadata, assumed hardware, Power Usage Effectiveness, and regional electricity mix into life-cycle impact estimates per million output tokens for energy, global warming potential, water consumption, and abiotic depletion. The Pareto frontier then sorts the 47 supported configurations so that each point on a frontier is not outperformed by another model on both safety and impact; this device turns raw pairs into the claim that the highest safety scores sit on a steep-impact tail.

What would settle it

Provider-side metered energy use per million output tokens for gpt-5.5 and claude-haiku-4.5 on comparable hardware, made public, would settle it: if the measured ratio is far below the modeled 60-fold, the central trade-off is an artifact of EcoLogits assumptions; likewise, a K-Bench evaluation showing high-reasoning configurations consistently outperform no-reasoning configurations across all matched comparisons would falsify the secondary claim.

Watch

Extended reading notes

Core claim

The central claim is that clinical safety and environmental footprint are in a non-linear trade-off for therapeutic LLMs, with sharply rising impact at the top of the safety distribution. On K-Bench's Best Risk score, gpt-5.5 reaches 96.02 at an estimated 6.4390 kWh per million output tokens, while claude-haiku-4.5 reaches 93.41 at 0.1095 kWh, a 98.3 percent reduction in energy for a 2.61-point safety gap, equivalently a roughly 60-fold energy increase for the higher score. The pattern repeats across global warming potential, water consumption, and abiotic depletion. A secondary claim is that additional test-time reasoning does not reliably buy safety: across 12 high-reasoning comparisons, only 4 improved over no reasoning, and across 9 low-reasoning comparisons, only 5 improved. The authors conclude that selecting solely by safety score, or assuming larger models and more compute are the route to safety, is inefficient, and they recommend dynamic model selection and cascading.

Load-bearing premise

The load-bearing premise is that EcoLogits' modeled life-cycle estimates, with their assumed hardware, data-center efficiency, and electricity mix, faithfully represent the relative environmental cost of serving each configuration; if providers serve flagship models on different hardware or with caching that the model does not capture, the reported 60-fold energy ratio could change materially.

Editorial extensions

If this is right

  • A deployer who wants the very highest K-Bench risk score should expect an estimated environmental footprint roughly 60 times larger than a model only 2.61 points behind, across energy, carbon, water, and resource depletion.
  • Efficient models such as claude-haiku-4.5 can sit on the Pareto frontier, meaning no other evaluated model achieves both a higher safety score and a lower estimated impact.
  • Test-time reasoning should not be treated as a reliable safety lever: in the evaluated configurations, added reasoning often lowered K-Bench scores, so compute budgets may be spent with no safety return.
  • Model selection in therapeutic AI should be framed as multi-objective optimization, with dynamic routing or cascading used to send high-risk cases to larger models and routine interactions to smaller ones.
  • Because EcoLogits returns identical environmental estimates for every configuration of a given base model, prompt- and reasoning-level safety differences are evaluated against a fixed per-model footprint rather than a per-configuration one.

Reading between the lines

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

  • A natural but implicit extension: the Pareto 'knee' where added safety per unit of environmental cost drops sharply could be formalized as a selection rule, such as choosing the model with the best safety score per kilowatt-hour above a clinically acceptable threshold; this is my inference, not stated in the paper.
  • If the EcoLogits hardware assumptions are right, provider-published energy meters should show the same ordering across models, though not necessarily the same ratios; a testable prediction is that the gpt-5.5 versus claude-haiku-4.5 energy ratio lies well above 10 in real deployments.
  • The reasoning findings suggest a mechanism worth testing: K-Bench's rubric may reward direct, well-bounded responses, while longer reasoning chains introduce more opportunities for unsafe or off-topic statements, a hypothesis the authors did not explore.
  • The same paired-analysis approach could be applied to other clinical tasks or to fine-tuned smaller models; if domain-specific training closes the safety gap, the case for frontier models in therapy weakens further.
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

3 major / 4 minor

Summary. This paper combines K-Bench clinical safety scores with EcoLogits life-cycle assessment estimates for 47 supported model configurations across 13 base models, and reports a non-linear safety--sustainability trade-off. The headline result is that the highest-scoring configuration, openai/gpt-5.5 (Best Risk 96.02), is estimated to use 6.4390 kWh per million output tokens, whereas anthropic/claude-haiku-4.5 achieves 93.41 at 0.1095 kWh, an approximately 60-fold energy increase for a 2.61-point safety gain. The paper also reports row-level analyses suggesting that additional test-time reasoning does not consistently improve safety, and recommends dynamic model selection and model cascading as a way to reduce environmental impact while preserving clinical performance in high-risk cases.

Significance. The paper addresses a genuine gap: clinical safety and environmental impact are rarely evaluated together for therapeutic LLMs, and the qualitative direction of the trade-off is visible in the data. Strengths include reliance on public benchmark data, use of a reproducible estimation package, absence of fitted parameters, and candid discussion of limitations. The secondary finding on test-time reasoning is appropriately hedged and does not depend on the environmental estimates. However, the central quantitative claim is a single-pair comparison built on modeled, configuration-invariant EcoLogits estimates that carry no uncertainty bounds; without a sensitivity analysis, the specific 60-fold disproportionality claim cannot be fully evaluated. This is a load-bearing issue for the paper's main conclusion, so the manuscript needs revision before the claim is accepted.

major comments (3)
  1. [Table 1 and Section 4] The headline ratio of approximately 60x energy per million output tokens (gpt-5.5 at 6.4390 kWh vs. claude-haiku-4.5 at 0.1095 kWh) rests entirely on EcoLogits model-level estimates. The paper states that these estimates depend on assumed hardware, PUE, and electricity mix, and Section 4 notes that EcoLogits returns identical impacts for every configuration of a base model. For closed models, provider-side serving details are not observable, so the true ratio could be materially different if gpt-5.5 is served on newer accelerators, different caching, or different routing than EcoLogits assumes. Please add a sensitivity analysis that varies hardware profile, PUE, electricity-mix zone, and token-normalization assumptions, and report how the headline ratio and the Pareto frontier change. Without such bounds, the quantitative disproportionality claim cannot be evaluated.
  2. [Section 4, Configuration-Level Variation] Because EcoLogits returns identical impact estimates within each base model, the comparison pairs configuration-specific safety scores with base-model-level environmental cost. The '2.61 points for ~60x' comparison uses the Best Risk score for each model but does not know the energy use of that particular prompt--reasoning configuration; the 6.4390 kWh figure is a base-model-level estimate. The paper should explicitly acknowledge this unit mismatch and, if possible, bound it by estimating configuration-specific inference cost, for example by accounting for the actual number of output tokens generated by each configuration in the K-Bench evaluation.
  3. [Section 3, Data Synthesis and Filtering] Of the 28 base architectures in the initial K-Bench dataset, 15 were excluded because EcoLogits lacked the required metadata, leaving 13 base models. The paper reports this transparently, but it does not discuss how this selection could bias the Pareto frontier. Excluded models such as llama-4-maverick, deepseek-v4-flash, claude-fable-5, and grok-4.20 may be systematically newer, less documented, or at different points on the safety--efficiency curve. Please characterize the K-Bench safety scores of the excluded models and state whether their inclusion or exclusion could change the location of the frontier or the qualitative conclusion.
minor comments (4)
  1. [Abstract] The abstract says '47 supported model configurations' while Table 1 and Figure 1 present 13 base models; the relationship between configurations and base models could be stated more explicitly in the methodology.
  2. [Section 5, Test-Time Compute] The example comparing gpt-5.5 with 'none' reasoning (96.02) and 'low' reasoning (95.50) would be more reproducible if the full configuration details, including the prompt variant, were given.
  3. [Figure 1] Some model labels in Figure 1 appear crowded or overlapping (e.g., the gemini labels in panels a and b); consider using a legend with distinct markers to improve readability.
  4. [References] There is a minor typo in the ACM reference line: 'InACM/IEEE' should read 'In ACM/IEEE'.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the safety–sustainability comparison uses two independent external data sources and fits no parameters.

full rationale

The paper's central claim is an observed correlation between K-Bench Best Risk scores and EcoLogits life-cycle impact estimates across 47 model configurations. K-Bench scores are produced by an external benchmark whose automated judge was calibrated against clinician consensus (Cohen's kappa = 0.850), and the safety values are not generated or fitted by the authors. EcoLogits estimates are produced by an external, cited software package with stated assumptions about hardware (e.g., H100/A100), PUE, and electricity mix, and the paper explicitly notes that these are modeled estimates rather than direct measurements. The analysis fits no parameters, derives no equation, and constructs neither the safety scores nor the environmental figures from its own assumptions; the only operation performed is merging the two external datasets by model endpoint. The acknowledged limitation that EcoLogits returns identical impact estimates within each base model affects the granularity of the environmental variable, but it is not circular because the environmental estimates are not defined in terms of the safety scores or vice versa. Similarly, the approximate 60-fold energy ratio is a ratio of two independent external estimates and is subject to sensitivity concerns, but sensitivity to modeling assumptions is an accuracy/robustness issue, not a circularity issue. The model-cascading recommendation is an application of the observed Pareto trade-off, not an input used to produce the result. There are no self-citations that carry the argument, and no uniqueness theorem or prior-work assertion is invoked to force the conclusion. The derivation chain is therefore self-contained with respect to circularity, and the appropriate finding is no significant circularity.

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

The paper contributes no new fitted parameters and no invented entities. Its central claim rests on two external artifacts, the K-Bench leaderboard and EcoLogits v0.11.0, plus a normalization choice. The epistemic debt is that these modeled estimates and benchmark scores are treated as ground truth, an assumption the paper itself flags as unverified for closed systems.

assumptions (3)
  • domain assumption EcoLogits v0.11.0 modeled life-cycle estimates (energy, GWP, water, ADPe) accurately represent relative inference footprints across the 47 supported configurations.
    The central trade-off is computed from these estimates; the paper states they are modeled rather than measured and are constant within each base model, so the comparison inherits EcoLogits' assumptions about hardware, PUE, and electricity mix (§3 Standardized Sustainability Estimation, §4 Configuration-Level Variation).
  • domain assumption K-Bench Best Risk scores, produced by an AI judge calibrated against clinicians (Cohen's kappa 0.85), are a valid comparative measure of clinical safety.
    Safety scores are taken from the K-Bench public leaderboard as the outcome metric; the paper relies on K-Bench's rubric and calibration without independent verification (§3 Dataset Curation and K-Bench).
  • domain assumption Normalizing environmental estimates to 1,000,000 output tokens makes cross-model comparisons meaningful despite differences in conversation length and token counts.
    The paper compares per-1M-token impacts across models; if actual token counts differ systematically, for example longer responses from frontier models, the normalized ratios may misstate true per-conversation costs (§3 Standardized Sustainability Estimation).

how reviews work

0 comments
Cite this review

Pith. "Pith review of Quantifying the Relationship Between Clinical Safety and Environmental Impact in Therapeutic LLMs." pith.science (2026). https://pith.science/paper/2R3E5LFI

@misc{pith2026260811830,
  author       = {Pith},
  title        = {Pith review of: Quantifying the Relationship Between Clinical Safety and Environmental Impact in Therapeutic LLMs},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/2R3E5LFI}},
  note         = {Machine review of arXiv:2608.11830}
}
read the original abstract

The deployment of large language models (LLMs) in mental health contexts raises questions about the relationship between clinical safety and environmental cost. In this paper, we examine this relationship by combining K-Bench clinical safety scores with EcoLogits life-cycle assessment estimates across 47 supported model configurations. We evaluate model performance and environmental impact across four dimensions: energy use, carbon emissions, water consumption, and abiotic depletion. The results indicate a non-linear trade-off at the upper end of the safety distribution: a 2.61 percentage-point increase in clinical safety score corresponded to an approximately 60-fold increase in estimated energy use per million output tokens. Row-level analyses further suggest that additional test-time compute did not consistently improve clinical safety and, in some configurations, was associated with lower clinical safety scores. These findings suggest that relying solely on larger models or additional inference-time computation may be an inefficient strategy for improving safety in therapeutic AI systems. We discuss the implications for sustainable deployment and highlight dynamic model selection, including model cascading, as a potential approach for reducing environmental impact while preserving clinical performance in higher-risk cases.

Figures

Figures reproduced from arXiv: 2608.11830 by the authors.

Figure 1
Figure 1. Clinical safety scores and estimated environmental impacts across base models. Red points identify nondominated [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

29 extracted references · 17 canonical work pages

  1. [1]

    Adrian Arnaiz-Rodriguez, Miguel Baidal, Erik Derner, Jenn Layton Annable, Mark Ball, Mark Ince, Elvira Perez Vallejos, and Nuria Oliver. 2026. Between Help and Harm: An Evaluation Study of Mental Health Crisis Handling by Large Language Models.JMIR Mental Health13 (2026). doi:10.2196/88435

    doi:10.2196/88435 2026

  2. [2]

    Bentley, Luca Belli, Adam M

    Kate H. Bentley, Luca Belli, Adam M. Chekroud, Emily J. Ward, Emily R. Dworkin, Emily Van Ark, Kelly M. Johnston, Will Alexander, Millard Brown, and Matt Hawrilenko. 2026. AI Chatbot Suicide Risk Detection and Response: Human Validation Study of the Open-Source VERA-MH Safety Evaluation.JMIR AI5 (2026). doi:10.2196/92817

    work page doi:10.2196/92817 2026

  3. [3]

    Tiago da Silva Barros, Frédéric Giroire, Ramón Aparicio-Pardo, and Joanna Moulierac. 2025. Small is Sufficient: Reducing the World AI Energy Consumption Through Model Selection. Preprint. arXiv:2510.01889 doi:10.48550/arXiv.2510. 01889

    work page doi:10.48550/arxiv.2510.01889 2025

  4. [4]

    Srimonti Dutta and Ratna Kandala. 2026. Mental Health AI Safety Claims Must Preserve Temporal Evidence. Preprint. arXiv:2605.08827 doi:10.48550/arXiv.2605. 08827

    work page Pith review arXiv doi:10.48550/arxiv.2605.08827 2026

  5. [5]

    Gazi, Carla Gorban, Carolyn I

    Bridget Dwyer, Matthew Flathers, Akane Sano, Allison Dempsey, Andrea Cipri- ani, Asim H. Gazi, Carla Gorban, Carolyn I. Rodriguez, Charles IV Stromeyer, Darlene King, Eden Rozenblit, Gillian Strudwick, Jake Linardon, Jiaee Cheong, Joseph Firth, Julian Herpertz, Julian Schwarz, Margaret Emerson, Martin P. Paulus, Michelle Patriquin, Yining Hua, Soumya Chou...

    work page arXiv 2025

  6. [6]

    Cooper Elsworth, Keguo Huang, David Patterson, Ian Schneider, Robert Sedivy, Savannah Goodman, Ben Townsend, Parthasarathy Ranganathan, Jeff Dean, Amin Vahdat, Ben Gómes, and James Manyika. 2025. Measuring the environ- mental impact of delivering AI at Google Scale. Preprint. arXiv:2508.15734 doi:10.48550/arXiv.2508.15734

    arXiv doi:10.48550/arxiv.2508.15734 2025

  7. [7]

    Ahmad Faiz, Sotaro Kaneda, Ruhan Wang, Rita Osi, Prateek Sharma, Fan Chen, and Lei Jiang. 2024. LLMCarbon: Modeling the End-to-End Carbon Footprint of Large Language Models. InInternational Conference on Learning Representations. doi:10.48550/arXiv.2309.14393

    arXiv doi:10.48550/arxiv.2309.14393 2024

  8. [8]

    Zhenxiao Fu, Fan Chen, Shan Zhou, Haitong Li, and Lei Jiang. 2025. LLMCO2: Ad- vancing Accurate Carbon Footprint Prediction for LLM Inferences.SIGENERGY Energy Inform. Rev.5, 2 (2025), 63–68. doi:10.1145/3757892.3757901

    work page arXiv 2025

Show all 29 references
  1. [9]

    Lecheng Gong, Weimin Fang, Ting Yang, Dongjie Tao, Chunxiao Guo, Peng Wei, Bo Xie, Jinqun Guan, Zixiao Chen, Fang Shi, Jinjie Gu, and Junwei Liu

  2. [10]

    Y. He, L. Yang, C. Qian, T. Li, Z. Su, Q. Zhang, and X. Hou. 2023. Conversational Agent Interventions for Mental Health Problems: Systematic Review and Meta- analysis of Randomized Controlled Trials.J Med Internet Res25 (2023). doi:10. 2196/43862

    2023

  3. [11]

    Rae, Oriol Vinyals, and Laurent Sifre

    Jordan Hoffmann, Sebastian Borgeaud, Arthur Mensch, Elena Buchatskaya, Trevor Cai, Eliza Rutherford, Diego de Las Casas, Lisa Anne Hendricks, Jo- hannes Welbl, Aidan Clark, Tom Hennigan, Eric Noland, Katie Millican, George van den Driessche, Bogdan Damoc, Aurelia Guy, Simon Os...

    2022 doi arXiv

  4. [12]

    Nidhal Jegham, Marwan Abdelatti, Chan Young Koh, Lassad Elmoubarki, and Abdeltawab Hendawi. 2025. How Hungry is AI? Benchmarking Energy, Water, and Carbon Footprint of LLM Inference. Preprint. arXiv:2505.09598 doi:10.48550/ arXiv.2505.09598

    2025 doi

  5. [13]

    Yu Jin, Jiayi Liu, Pan Li, Baosen Wang, Yangxinyu Yan, Huilin Zhang, Chenhao Ni, Jing Wang, Yi Li, Yajun Bu, and Yuanyuan Wang. 2025. The Applications of Large Language Models in Mental Health: Scoping Review.J Med Internet Res27 (2025). doi:10.2196/69284

    2025 doi

  6. [14]

    Yunho Jin, Gu-Yeon Wei, and David Brooks. 2025. The Energy Cost of Rea- soning: Analyzing Energy Usage in LLMs with Test-time Compute. Preprint. arXiv:2505.14733 doi:10.48550/arXiv.2505.14733

    2025 doi

  7. [15]

    Junyu Liu, Zirui Li, Qian Niu, Zequn Zhang, Yue Xun, Wenlong Hou, Shu- jun Wang, Yusuke Iwasawa, Yutaka Matsuo, and Kan Hatakeyama-Sato. 2026. JMedEthicBench: A Multi-Turn Conversational Benchmark for Evaluating Med- ical Safety in Japanese Large Language Models. Preprint. arX...

    2026 doi arXiv

  8. [16]

    Alexandra Sasha Luccioni, Sylvain Viguier, and Anne-Laure Ligozat. 2023. Es- timating the Carbon Footprint of BLOOM, a 176B Parameter Language Model. Journal of Machine Learning Research24, 253 (2023), 1–15. http://jmlr.org/papers/ v24/23-0069.html

    2023

  9. [17]

    Sasha Luccioni, Yacine Jernite, and Emma Strubell. 2024. Power Hungry Process- ing: Watts Driving the Cost of AI Deployment?. InACM Conference on Fairness, Accountability, and Transparency. 85–99. doi:10.1145/3630106.3658542

    2024

  10. [18]

    Maity and M

    S. Maity and M. J. Saikia. 2025. Large Language Models in Healthcare and Medical Applications: A Review.Bioengineering12, 6 (2025). doi:10.3390/ bioengineering12060631

    2025

  11. [19]

    Matteo Malgaroli, Katharina Schultebraucks, Keris Jan Myrick, Alexandre An- drade Loch, Laura Ospina-Pinillos, Tanzeem Choudhury, Roman Kotov, Munmun De Choudhury, and John Torous. 2025. Large language models for the mental health community: framework for translating code to c...

    2025 doi

  12. [20]

    Richardson, Angel Y

    Lotenna Olisaeloka, Chris G. Richardson, Angel Y. Wang, Richard J. Munthali, and Daniel V. Vigo. 2026. Safety Mechanisms and Risk Mitigation in Generative AI Mental Health Chatbots: A Systematic Scoping Review.Healthcare14, 10 (2026). doi:10.3390/healthcare14101395

    2026 doi

  13. [21]

    Lavista Ferres

    Felipe Oviedo, Fiodar Kazhamiaka, Esha Choukse, Allen Kim, Amy Luers, Melanie Nakagawa, Ricardo Bianchini, and Juan M. Lavista Ferres. 2026. Energy use of AI inference, efficiency pathways, and test-time scaling.Joule(2026). doi:10.1016/j. joule.2026.102430

    2026

  14. [22]

    Miray Özcan, Philipp Wiesner, Philipp Weiß, and Odej Kao. 2025. Quantifying the Energy Consumption and Carbon Emissions of LLM Inference via Simulations. Preprint. arXiv:2507.11417 doi:10.48550/arXiv.2507.11417

    2025 doi arXiv

  15. [23]

    Samuel Rincé and Adrien Banse. 2025. EcoLogits: Evaluating the Environmental Impacts of Generative AI.Journal of Open Source Software10, 111 (2025). doi:10. 21105/joss.07471

    2025

  16. [24]

    Smith, and Oren Etzioni

    Roy Schwartz, Jesse Dodge, Noah A. Smith, and Oren Etzioni. 2020. Green AI. Commun. ACM63, 12 (2020), 54–63. doi:10.1145/3381831

    2020 doi

  17. [25]

    Stamatis, Jonah Meyerhoff, Richard Zhang, Olivier Tieleman, Matteo Malgaroli, and Thomas D

    Caitlin A. Stamatis, Jonah Meyerhoff, Richard Zhang, Olivier Tieleman, Matteo Malgaroli, and Thomas D. Hull. 2026. Beyond simulations: what 20,000 real conversations reveal about mental health AI safety. Preprint. doi:10.21203/rs.3.rs- 8642399/v1

    2026 doi

  18. [26]

    Samantha Weber, Dustin Klebe, Lukas Wolf, Christopher Aeberli, Stephanie Homan, Nikolas Psathakis, Andrea Casanova, José Garcia Macias, Jack Sykstus, Jie-Ming Li, Barbora Provaznikova, Judith Rohde, Laura Frühschütz, Charlotta Rühlmann, Tobias Welt, Tobias Kowatsch, Birgit Kle...

    2026 doi

  19. [27]

    Elaine Chen, David Sontag, and Adler Perotte

    Alexander Wind, Alexander D’Amour, Vinu Sivaraman, I. Elaine Chen, David Sontag, and Adler Perotte. 2026. Safety and Accuracy Follow Different Scaling Laws in Clinical Large Language Models. Preprint. arXiv:2606.00063 https: //arxiv.org/abs/2606.00063

    2026 arXiv

  20. [28]

    Lin Yang, Yuancheng Yang, Xu Wang, Changkun Liu, and Haihua Yang. 2026. MedMT-Bench: Can LLMs Memorize and Understand Long Multi-Turn Conver- sations in Medical Scenarios? Preprint. arXiv:2603.23519 doi:10.48550/arXiv.2603. 23519

    2026 doi

  21. [2026]

    Preprint

    MedDialogRubrics: A Comprehensive Benchmark and Evaluation Frame- work for Multi-turn Medical Consultations in Large Language Models. Preprint. arXiv:2601.03023 doi:10.48550/arXiv.2601.03023

    doi

Pith tools

Reviewed August 16, 2026 · model on record in the stance chip above.