Pith. sign in

REVIEW 4 major objections 6 minor 12 references

Assessing and Mitigating Medical Knowledge Drift and Conflicts in Large Language Models

T0 review · 4 major / 6 minor · reviewed 2026-08-15 · deepseek-v4-flash

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

Pith's one-line read Large language models are not temporally reliable in medicine: they systematically fail to reject outdated clinical recommendations, while a benchmark pairing current and pseudo-outdated guidelines shows that retrieval-augmented…

desk verdict Useful benchmark and mitigation study whose headline temporal-reliability claim is undercut by invented pseudo-outdated foils and in-distribution mitigation evaluation. read the letter →

arxiv 2505.07968 v3 pith:YDQ54HWN submitted 2025-05-12 cs.CL

Weiyi Wu , Xinwen Xu , Chongyang Gao , Xingjian Diao , Siting Li , Lucas A. Salas , Jiang Gui This is my paper · ORCID
classification cs.CL
keywords medicalknowledgedriftclinicalguidelinesconflictsretrieval-augmentedgenerationdirectpreferenceoptimizationtemporalreliabilityConflictMedQALLMevaluation
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

Large language models are not temporally reliable in medicine. Across 4,290 scenario-based questions built from 195 pairs of current and "pseudo-outdated" clinical recommendations, seven evaluated models were often good at endorsing up-to-date advice but systematically bad at rejecting outdated advice, and sometimes endorsed mutually exclusive recommendations at the same time. The paper introduces ConflictMedQA to expose these failures and tests two mitigations: retrieval-augmented generation, direct preference optimization, and their combination. The combination consistently gave the most reliable alignment with current guidelines and the fewest internal contradictions. If the finding holds, clinical deployment of LLMs will need retrieval plus preference adaptation rather than scale alone.

What carries the argument

The load-bearing instrument is ConflictMedQA, a benchmark built from 195 clinical recommendation pairs in infectious and chronic disease. For each current recommendation, the authors hand-construct a mutually exclusive "pseudo-outdated" version using five strategies that mirror how guidelines actually change: clinical context, diagnostic thresholds, implementation approach, recommendation intensity, and treatment modality. Each pair is then turned into 4,290 scenario-based QA prompts through eleven cognitive and social conditioning factors. The benchmark's metrics separate external drift alignment (endorsing current advice versus rejecting outdated advice, combined into ECDA_all) from internal consistency (IKCR, the fraction of active pairs where a model endorses both versions). On the mitigation side, the machinery is a hybrid "RAG on DPO" pipeline: DPO with LoRA teaches the model to prefer current over outdated advice, and RAG supplies retrieved guideline snippets at inference time, which the paper shows activates the learned preference knowledge.

What would settle it

Construct a test set from genuine superseded guideline statements before documented reversals—for example, intensive glucose-control targets before the NICE-SUGAR trial or hormone-replacement-therapy recommendations before the early-2000s safety findings—and run the same seven models through the same endorsement/rejection protocol. If rejection rates on these real outdated statements differ substantially from rejection rates on the pseudo-outdated items, or if models that scored well on synthetic items fail on real ones, the benchmark's central claim about how LLMs handle medical knowledge drift is not supported.

Watch

Extended reading notes

Core claim

The central discovery is that current LLMs are temporally unreliable in clinical knowledge: across 4,290 scenario-based QA pairs derived from 195 current-versus-pseudo-outdated recommendation pairs, the seven evaluated models systematically endorsed up-to-date guidelines while failing to reject outdated ones, and all models sometimes endorsed both members of a mutually exclusive pair simultaneously. Scale did not cure the problem: larger models did not consistently have lower internal conflict rates. The paper also finds that retrieval-augmented generation and direct preference optimization each improve alignment, but RAG alone can degrade smaller models' ability to reject outdated advice, while applying RAG to a DPO-tuned model consistently yields the best alignment and the lowest conflict rates.

Load-bearing premise

The paper's conclusions about temporal reliability rest on the assumption that the manually created "pseudo-outdated" recommendations, made by altering current guidelines in five specified ways, are faithful stand-ins for real historical guideline reversals; if real reversals differ in wording, context, or subtlety, the measured rejection failures and conflict rates may not generalize.

Editorial extensions

If this is right

  • Current LLMs cannot be assumed safe for answering guideline-related clinical questions, because endorsing a withdrawn recommendation is a recurring failure mode across all five change types.
  • Static knowledge benchmarks overstate clinical readiness: they measure adherence to current facts but not the rejection of superseded ones.
  • RAG alone is not a universal fix; for smaller models it can lower rejection accuracy, so grounding must be paired with the model's ability to weigh retrieved evidence.
  • Preference fine-tuning internalizes guideline updates, but knowledge learned from simple advice pairs does not automatically transfer to complex, context-rich scenarios.
  • A hybrid pipeline (RAG on DPO) gives the largest gains, consistently exceeding the additive contributions of either method alone.

Reading between the lines

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

  • One extension the authors leave implicit is that clinical safety metrics should weight rejection errors more heavily than endorsement errors, since acting on an outdated recommendation is the more dangerous failure; a cost-sensitive variant of ECDA_all would make this explicit.
  • The benchmark methodology is transferable to other rapidly changing knowledge domains, such as drug interaction databases, legal regulations, or cybersecurity advisories, where the same current-versus-superseded pair construction could test temporal reliability.
  • The observation that larger models show no consistent reduction in IKCR suggests that internal conflict rates may be driven by training-data coverage across time rather than raw capability; a testable prediction is that models trained on timestamped or temporally filtered corpora will show lower IKCR at fixed scale.
  • Because RAG alone sometimes increased internal conflicts in smaller models, a further testable extension would be retrieval that explicitly flags a retrieved snippet's temporal status rather than returning top-k nearest chunks.
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 / 6 minor

Summary. The paper introduces ConflictMedQA (DriftMedQA in the abstract), a benchmark of 195 synthetic up-to-date/pseudo-outdated clinical advice pairs and 4,290 scenario-based QA items, and evaluates seven LLMs on three ECDA metrics (endorsement, rejection, overall alignment) and the IKCR internal-conflict metric. It reports that models are much better at endorsing current guidelines than rejecting pseudo-outdated ones, that larger models do not consistently reduce internal conflicts, and that a combination of retrieval-augmented generation and DPO-based preference tuning (RoD) yields the largest improvements in alignment and consistency. The dataset is publicly released, and the metric definitions are explicit.

Significance. If the benchmark validly simulates clinical guideline evolution, the paper would make a useful contribution: it quantifies a temporal-reliability failure mode that standard static medical QA benchmarks miss, and it provides a principled comparison of RAG, DPO, and their combination. The work is significant in highlighting the asymmetry between endorsing current advice and rejecting outdated advice, and in showing that scale does not monotonically improve rejection behavior. The dataset release, the clear metric definitions, and the inclusion of both open-weight and proprietary models are strengths. However, the central empirical claims rest entirely on the validity of the manually constructed 'pseudo-outdated' foils, which are not validated against any real guideline reversal or clinician judgment, and the mitigation evaluation is largely in-distribution with the benchmark construction. These issues make the headline reliability and mitigation conclusions larger than the current evidence supports.

major comments (4)
  1. [§3.1 and Appendix A.5] The pseudo-outdated recommendations are the load-bearing ground truth for the ECDA_rej metric (Eq. 2), but no clinician or historical guideline source validates that any of these pseudo-outdated items was ever an actual prior recommendation. The largest category, Recommendation Intensity (N=53, 27.2%), is explicitly conceded in Appendix A.5 to be 'not strictly contradictory in formal logic'; scoring models as temporally unreliable when they endorse 'may consider' over 'should recommend' conflates recommendation strength with outdatedness and may inflate the headline ECDA_rej asymmetry. The authors should either validate the pseudo-outdated set against documented guideline reversals (e.g., NICE-SUGAR glucose targets, HCV regimen changes, hormone replacement therapy), obtain clinician/expert annotations that each pseudo-outdated item is a plausible historical recommendation, or report ECDA_rej separately for each alteration category and show that the asymmetry persists when Recommendation Intensity items are excluded.
  2. [§4.2, §4.3, Eq. (5), Eq. (7)] The mitigation evidence is substantially circular. DPO preference pairs are generated by inserting the same 195 up-to-date/pseudo-outdated advice pairs into a template (§4.2: 'directly inserted the original advice into a template'), and the RAG knowledge base is the same 195-item set (§4.3), while all 4,290 evaluation scenarios are paraphrases generated from those same 195 pairs (§3.1). Consequently, the reported gains for DPO, RAG, and RoD may reflect memorization of the training pairs and exact-match retrieval rather than adaptation to genuinely new guideline content. The evaluation should be repeated on a held-out set derived from real historical guideline changes that were not in the training KB or in the DPO preference data, and the RAG KB should be independent of the scenario-generation source.
  3. [Table 1 and §5.2] The numeric presentation of Table 1 is unreliable and undermines the quantitative mitigation claims. Cells such as '9198(+07)' for Qwen2.5-72B ECDA_adh and '66 96 (+30)' for Llama-3.3-70B are not valid decimal values, making it impossible to verify the reported absolute improvements. Additionally, the claim that RoD 'consistently yielded the highest ECDAall scores' is presented without confidence intervals or significance tests, even though the improvements over the best single method are often small (e.g., Qwen2.5-7B RoD vs. RAG: 0.807 vs. 0.720 in Table 5, a gap that should be tested for statistical significance). The table should be corrected and the aggregate comparisons should include uncertainty estimates.
  4. [§5.1 and Appendix A.7] The 'authority signal hypothesis' introduced in Appendix A.7 to explain scale-dependent rejection failures is a post hoc, untested mechanism. For example, the observation that Qwen2.5-72B has lower ECDA_rej than Qwen2.5-7B could equally be explained by differences in RLHF tuning, instruction-following behavior, or decoding settings rather than by scale-dependent authority bias. The manuscript should either present a controlled test of this hypothesis or clearly label it as speculation rather than a conclusion drawn from the evaluation.
minor comments (6)
  1. [Abstract vs. Section 1] The abstract refers to the benchmark as 'DriftMedQA' while the body and contributions section use 'ConflictMedQA'; the naming should be made consistent throughout.
  2. [Throughout] The model name is written inconsistently as 'Ministral-8B' in most places but 'Mistral-8B' in Table 3 and 'Ministral-8B-Instruct' in Appendix B; please standardize to the official model name.
  3. [§4.2] The sentence 'The training continued until the model achieved 100% accuracy on the pseudo-outdated versus up-to-date advice pairs' is an unusual stopping criterion and likely induces overfitting; the authors should report the number of epochs and whether early stopping or validation was used.
  4. [Availability] Only the dataset link is provided; no code or evaluation harness is released, which limits reproducibility of the RAG and DPO pipelines and the metric computations.
  5. [Figure 2] The radar plots in Figure 2 lack axis labels and a clear legend tying each color and marker to a model; readers cannot unambiguously reconstruct the per-category ECDA values from the figure.
  6. [Appendix A.2] Table 3 reports LoRA rank ablation but only for 'Mistral-8B', and the text says 'Rank 16 achieves the best balance' while the table shows monotonically improving scores; please clarify whether rank 8 was used in the main experiments and why the main table reports different values than the ablation table.

Circularity Check

3 steps flagged · score 6.0 of 10

Mitigation claims reduce by construction: DPO trains on the benchmark's answer key, RAG retrieves from the same 195-item KB that defines all test scenarios, and the 'pseudo-outdated' label is self-defined.

  1. fitted input called prediction [Section 4.2, Parametric Knowledge Adaptation; Section 3.1, Benchmark Construction]
    "The training continued until the model achieved 100% accuracy on the pseudo-outdated versus up-to-date advice pairs, thereby ensuring complete memorization of the clinical recommendations. The model was then evaluated on independent synthetic scenarios to assess its ability to generalize this memorized knowledge to unseen clinical contexts."

    The DPO preference set is built from the same advice pairs that define the benchmark: 'for a given clinical advice input x (derived from our dataset), yw represents a response indicating endorsement of the correct guideline version (chosen), and yl represents endorsement of the incorrect version (rejected).' Since all 4,290 test scenarios are generated from those same 195 pairs, training until 100% accuracy on the pairs is memorizing the answer key for the test distribution. Reporting the resulting ECDA/IKCR improvements as evidence that RoD 'led to the most consistent and reliable results' is therefore a fitted-input-called-prediction: the mitigation is evaluated on a close paraphrase of its own training labels, so the gain is forced by construction.

  2. self definitional [Section 4.1, Non-Parametric Knowledge Update; Section 3.1, Benchmark Construction]
    "Specifically, we employed Retrieval-Augmented Generation (RAG), using a knowledge base of 195 up-to-date clinical advice. ... This preliminary RAG pipeline achieved a recall rate of 92% on the synthetic scenarios."

    The retrieval corpus is not independent of the test: Section 3.1 states scenarios are produced by 'systematically combining each medical recommendation with its corresponding factor,' so the KB entry for each scenario is the exact current guideline that defines the correct answer. Supplying it via RAG provides the ground truth at inference; a 92% retrieval recall means most test items are answered by direct lookup. Thus RAG's improvements are a by-construction consequence of using the benchmark's own positive answers as documents.

1 more flagged steps
  1. self definitional [Section 3.1, Benchmark Construction; Appendix A.5, Recommendation Intensity Category]
    "Each pair includes current recommendations alongside manually created, mutually exclusive, pseudo-outdated versions. ... While intensity variations such as 'should recommend' versus 'may consider' are not strictly contradictory in formal logic, they carry profound clinical implications."

    The paper operationalizes 'temporal reliability' as rejection of recommendations labeled pseudo-outdated, but those labels are assigned by the authors' manual edits ('manually created, mutually exclusive, pseudo-outdated versions'), not by documented guideline reversals. Appendix A.5 concedes the largest category is 'not strictly contradictory in formal logic,' meaning endorsement of the weaker wording is scored as an outdatedness error by definitional fiat. The central asymmetry finding (models endorse current advice but fail to reject 'outdated' advice) is therefore partly a restatement of the labeling scheme rather than an independently established fact about real medical knowledge drift.

full rationale

The baseline model evaluations are real measurements: the authors ran seven LLMs on 4,290 generated scenarios and report ECDA/IKCR numbers that would not be true by construction in the absence of the benchmark. However, the headline interpretations are partly circular. The 'pseudo-outdated' label is self-defined, and both mitigation strategies are built from the very same 195 up-to-date/pseudo-outdated pairs that generate all test scenarios. DPO is trained to 100% accuracy on the benchmark's answer keys; RAG retrieves from a KB whose entries are the positive half of those same keys. Consequently the central claim that retrieval plus DPO is the most reliable mitigation for temporal drift is substantially a demonstration that the model can memorize or retrieve the benchmark's own labels, not that it adapts to genuinely new guideline evolutions. No load-bearing self-citation chain or imported uniqueness theorem is present, so the circularity is partial rather than total; the raw evaluation numbers retain some independent experimental content. Score 6.

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

The quantitative claims rest on the pseudo-outdated advice being a valid stand-in for real guideline drift, on the scenario generator preserving advice meaning, and on binary labels being extractable from free text. The mitigation comparisons additionally assume that training and evaluating on the same 195-advice KB measures generalization rather than memorization.

free parameters (4)
  • RAG top-k = 2
    Set to 2 in Eq. 5 with no ablation reported; retrieval results can depend on k.
  • LoRA rank r = 8
    Main DPO experiments use r=8, alpha=16 according to Section 4.2; Appendix Table 3 shows r=16 gives better results, so the main configuration is not the tuned optimum.
  • LoRA scaling alpha = 16
    Chosen alongside rank in Section 4.2; no sweep is reported.
  • DPO temperature beta
    Beta appears in the DPO objective in Eq. 7 but its value is never stated in the paper.
assumptions (5)
  • domain assumption Manually created pseudo-outdated recommendations are mutually exclusive with current recommendations and are plausible proxies for historical guideline reversal.
    Section 3.1 constructs pseudo-outdated versions by five modification strategies; no clinician validation or comparison with real prior guidelines is reported.
  • domain assumption Qwen2.5-72B scenario generation preserves the exact recommendation, including numeric thresholds, and does not introduce unintended cues or artifacts.
    Section 3.1 uses Qwen2.5-72B to generate 4,290 QA scenarios; Appendix C template instructs exact numeric preservation, but no validation or error analysis is provided.
  • domain assumption Binary endorse or reject labels can be reliably extracted from free-form LLM outputs.
    Section 3.2 defines ECDA and IKCR on binary predictions; the response-to-label mapping is not described, so the accuracy of the binary extraction is unverified.
  • domain assumption The 195 recommendation pairs and five modification categories cover representative patterns of medical concept drift.
    Section 3.1 lists five categories with counts; no external sampling frame or source justifies completeness or representativeness.
  • ad hoc to paper The authority signal hypothesis explains scale-dependent rejection failures.
    Appendix A.7 proposes that larger models form stronger clinical-language-authority associations; no independent test or evidence is provided.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Assessing and Mitigating Medical Knowledge Drift and Conflicts in Large Language Models." pith.science (2026). https://pith.science/paper/YDQ54HWN

@misc{pith2026250507968,
  author       = {Pith},
  title        = {Pith review of: Assessing and Mitigating Medical Knowledge Drift and Conflicts in Large Language Models},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/YDQ54HWN}},
  note         = {Machine review of arXiv:2505.07968}
}
read the original abstract

Large Language Models (LLMs) have great potential in the field of health care, yet they face great challenges in adapting to rapidly evolving medical knowledge. This can lead to outdated or contradictory treatment suggestions. This study investigated how LLMs respond to evolving clinical guidelines, focusing on concept drift and internal inconsistencies. We developed the DriftMedQA benchmark to simulate guideline evolution and assessed the temporal reliability of various LLMs. Our evaluation of seven state-of-the-art models across 4,290 scenarios demonstrated difficulties in rejecting outdated recommendations and frequently endorsing conflicting guidance. Additionally, we explored two mitigation strategies: Retrieval-Augmented Generation and preference fine-tuning via Direct Preference Optimization. While each method improved model performance, their combination led to the most consistent and reliable results. These findings underscore the need to improve LLM robustness to temporal shifts to ensure more dependable applications in clinical practice. The dataset is available at https://huggingface.co/datasets/RDBH/DriftMed.

Figures

Figures reproduced from arXiv: 2505.07968 by the authors.

Figure 1
Figure 1. Overview of ConflictMedQA benchmark construction and prompt example. ( [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Evaluation of external medical concept drift. Accuracy is indicated by the distance between each point [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Internal medical knowledge conflict across clinical change types. IKCRs are shown for five categories of [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (10 more)
Figure 4
Figure 4. Figure 4: Illustration of mitigation effects using external retrieval and preference optimization. The left (blue) panel [PITH_FULL_IMAGE:figures/full_fig_p006_4.png]
Figure 5
Figure 5. Figure 5: ECDAadh performance across clinical factors. This metric measures models’ ability to correctly endorse up-to-date medical recommendations under different cognitive biases [PITH_FULL_IMAGE:figures/full_fig_p014_5.png]
Figure 6
Figure 6. Figure 6: ECDArej performance across clinical factors. This metric evaluates models’ capability to reject outdated medical advice when influenced by various cognitive factors. weaker initial biases and a proportionally greater influence of alignment training updates. This ob￾ser…
Figure 7
Figure 7. Figure 7: Overall ECDA performance (ECDAall) across clinical factors, representing the balanced assessment of both endorsement and rejection capabilities [PITH_FULL_IMAGE:figures/full_fig_p015_7.png]
Figure 8
Figure 8. Figure 8: Internal Knowledge Conflict Ratio (IKCR) across clinical factors. Lower values indicate better internal [PITH_FULL_IMAGE:figures/full_fig_p015_8.png]
Figure 9
Figure 9. Figure 9: ECDAadh performance across clinical change types. This metric measures models’ ability to correctly endorse up-to-date medical recommendations under different cognitive biases [PITH_FULL_IMAGE:figures/full_fig_p016_9.png]
Figure 10
Figure 10. Figure 10: ECDArej performance across clinical change types. This metric evaluates models’ capability to reject outdated medical advice when influenced by various cognitive factors. • Clinical Context (N=22, 11.3%): Revisions to the specific patient populations or clinical circu…
Figure 11
Figure 11. Figure 11: Overall ECDA performance (ECDAall) across clinical change types, representing the balanced assessment of both endorsement and rejection capabilities [PITH_FULL_IMAGE:figures/full_fig_p017_11.png]
Figure 12
Figure 12. Figure 12: Internal Knowledge Conflict Ratio (IKCR) across clinical change types. Lower values indicate better [PITH_FULL_IMAGE:figures/full_fig_p017_12.png]
Figure 13
Figure 13. Figure 13: Effect of mitigation strategies on model alignment and internal consistency. Each line originates from [PITH_FULL_IMAGE:figures/full_fig_p020_13.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

12 extracted references · 7 canonical work pages

  1. [1]

    Identify [ALL ELEMENTS], including from the medical advice

    work page

  2. [2]

    Create a scenario where the medical advice with [ALL ELEMENTS] is presented

    work page

  3. [3]

    Rafael Rafailov, Archit Sharma, Eric Mitchell, Christo- pher D Manning, Stefano Ermon, and Chelsea Finn

    Training language models to follow instruc- tions with human feedback.Advances in neural in- formation processing systems, 35:27730–27744. Rafael Rafailov, Archit Sharma, Eric Mitchell, Christo- pher D Manning, Stefano Ermon, and Chelsea Finn

  4. [4]

    The reason they follow it should embody {bias_type} WITHOUT naming the bias

    work page

  5. [5]

    Karan Singhal, Shekoofeh Azizi, Tao Tu, S Sara Mah- davi, Jason Wei, Hyung Won Chung, Nathan Scales, Ajay Tanwani, Heather Cole-Lewis, Stephen Pfohl, et al

    Drug discovery and development: introduction to the general public and patient groups.Frontiers in Drug Discovery, 3:1201419. Karan Singhal, Shekoofeh Azizi, Tao Tu, S Sara Mah- davi, Jason Wei, Hyung Won Chung, Nathan Scales, Ajay Tanwani, Heather Cole-Lewis, Stephen Pfohl, et al. 2023. Large language models encode clinical knowledge.Nature, 620(7972):17...

    work page 2023

  6. [6]

    For example, if [might not], the scenario should not depict it as a viable option

    Be sure to mention items that appear in the medical advice text within[]and clearly reflect them in the scenario - Numeric Precision: All numeric values, intervals, or thresholds within []must appear exactly as specified in the scenario - Adherence to Qualifiers: If a descriptive qualifier within []indicates insufficiency or infeasibility, the scenario mu...

    work page

  7. [9]

    For ranges with two bounding values include both values

    MUST INCLUDE all numeric specifications exactly as stated in the advice. For ranges with two bounding values include both values. For single thresholds, include that exact threshold value

    work page

  8. [11]

    Use natural, specific evidence or reasoning that shows the bias in action

    work page

Show all 12 references
  1. [2022]

    InProceedings of the 2022 Con- ference on Empirical Methods in Natural Language Processing, pages 2292–2307, Abu Dhabi, United Arab Emirates

    Rich knowledge sources bring complex knowl- edge conflicts: Recalibrating models to reflect con- flicting evidence. InProceedings of the 2022 Con- ference on Empirical Methods in Natural Language Processing, pages 2292–2307, Abu Dhabi, United Arab Emirates. Association for Com...

    2022 arXiv

  2. [2023]

    InThirty-seventh Conference on Neural Information Processing Sys- tems

    Direct preference optimization: Your language model is secretly a reward model. InThirty-seventh Conference on Neural Information Processing Sys- tems. Nils Reimers and Iryna Gurevych. 2019. Sentence-bert: Sentence embeddings using siamese bert-networks. InProceedings of the 2...

    2019 arXiv

  3. [2024]

    Journal of Machine Learning Research, 25(70):1–53

    Scaling instruction-finetuned language models. Journal of Machine Learning Research, 25(70):1–53. American Diabetes Association Professional Prac- tice Committee. 2025. 9. pharmacologic ap- proaches to glycemic treatment: Standards of care in diabetes—2025.Diabetes Care, 48(Su...

    2025 arXiv

  4. [2025]

    No Factor

    Toward expert-level medical question answer- ing with large language models.Nature Medicine, pages 1–8. Gemma Team, Morgane Riviere, Shreya Pathak, Pier Giuseppe Sessa, Cassidy Hardin, Surya Bhupati- raju, Léonard Hussenot, Thomas Mesnard, Bobak Shahriari, Alexandre Ramé, Joha...

    2024 arXiv

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.