pith. machine review for the scientific record. sign in

arxiv: 2604.18031 · v1 · submitted 2026-04-20 · 💻 cs.CL · cs.LG· q-bio.BM

Recognition: unknown

How Creative Are Large Language Models in Generating Molecules?

Wen Tao , Yiwei Wang , Peng Zhou , Bryan Hooi , Wanlong Fang , Tianle Zhang , Xiao Luo , Yuansheng Liu
show 1 more author
Alvin Chan
Authors on Pith no claims yet

Pith reviewed 2026-05-10 04:18 UTC · model grok-4.3

classification 💻 cs.CL cs.LGq-bio.BM
keywords large language modelsmolecule generationcreativityconstraint satisfactionmolecular discoveryconvergent creativitydivergent creativitydrug design
0
0 comments X

The pith

LLMs exhibit distinct creative patterns in molecule generation by satisfying more constraints when additional ones are imposed.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper treats molecule generation as a search problem that must meet many simultaneous chemical, biological, and safety constraints inside a vast structured space. It therefore defines two functional forms of creativity: convergent creativity, which finds non-obvious solutions that obey the limits, and divergent creativity, which keeps exploring to escape poor local results. Large language models are tested directly on physicochemical, ADMET, and activity tasks by varying the number and type of constraints given in natural-language prompts. The central observation is that these models improve at meeting constraints precisely when more constraints are added, producing recognizable patterns rather than uniform or random behavior. A reader cares because the finding supplies a practical way to decide when LLMs belong in a molecular-discovery pipeline and when other methods are still needed.

Core claim

Molecule generation requires satisfying multiple chemical and biological constraints while searching a large and structured chemical space, turning the task into a non-binary problem that demands both convergent creativity to locate solutions under limits and divergent creativity to maintain exploration and avoid local optima. Large language models are shown to generate molecular representations from natural-language prompts and, when evaluated across physicochemical, ADMET, and biological-activity tasks, display distinct creative behaviors, including a clear rise in constraint satisfaction as additional constraints are introduced in the prompts.

What carries the argument

Convergent creativity (meeting imposed constraints) and divergent creativity (sustained exploration to escape local optima), used as the two complementary dimensions to characterize and measure LLM performance in molecule generation.

If this is right

  • Imposing additional constraints on LLMs can raise their rate of producing valid molecules that meet chemical and biological requirements.
  • LLMs appear better suited to constrained, goal-directed generation than to unconstrained exploration in molecular tasks.
  • The two-dimensional creativity framing supplies a repeatable way to benchmark LLMs before placing them inside discovery pipelines.
  • The observed patterns help delineate the specific roles LLMs can usefully play versus other generative methods.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same constraint-sensitivity pattern could be tested in related generative domains such as protein or material design.
  • Hybrid systems might deliberately increase prompt constraints to exploit LLMs' convergent strengths while pairing them with separate exploration engines.
  • Prompt-engineering techniques that systematically vary constraint load could become a practical control knob for LLM-based molecular work.

Load-bearing premise

That the chosen tasks and evaluation criteria for convergent and divergent creativity accurately capture what is required for effective molecule generation.

What would settle it

An experiment in which adding constraints produces no measurable rise in LLM constraint-satisfaction rates, or in which LLM outputs are statistically indistinguishable from random or non-exploratory baselines on the same tasks, would undermine the claim of distinct creative patterns.

Figures

Figures reproduced from arXiv: 2604.18031 by Alvin Chan, Bryan Hooi, Peng Zhou, Tianle Zhang, Wanlong Fang, Wen Tao, Xiao Luo, Yiwei Wang, Yuansheng Liu.

Figure 1
Figure 1. Figure 1: Overview of how LLM creativity in molecule generation is evaluated. Top: LLMs generate molecules conditioned on con￾straints (physicochemical, ADMET, and biological activity) spec￾ified in the prompt. Bottom: Creativity is operationalized along two complementary dimensions: convergent creativity (constraint satisfaction) and divergent creativity (chemical space exploration). additional constraints related … view at source ↗
Figure 2
Figure 2. Figure 2: Creativity profiles vary systematically by task type. Physicochemical and ADMET tasks show high convergent creativity but lower exploration, while activity tasks exhibit the opposite pattern. constraints [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Generated LogP distributions shift systematically toward target values. improvements in SR and validity. For example, LLaMA3.1- 70B consistently outperforms LLaMA3.1-8B, and a simi￾lar trend is observed when comparing LLaMA3-70B with LLaMA3-8B. At the same time, uniqueness and diversity decrease sub￾stantially as model size increases. This indicates that larger models generate molecules that more reliably … view at source ↗
Figure 5
Figure 5. Figure 5: Larger models maintain stable, high validity across diverse constraint settings, while smaller models show substantial variation. The red line represents the median, and the red dot represents the mean. 4.5. How do more constraints influence creativity? Adding compatible constraints can increase success rate rather than reduce it. Following (Zhou et al., 2025; Liu et al., 2025), we evaluate three two-const… view at source ↗
Figure 4
Figure 4. Figure 4: Larger models achieve higher constraint satisfaction but reduced exploration. Larger models produce more consistent molecular valid￾ity. As shown in [PITH_FULL_IMAGE:figures/full_fig_p006_4.png] view at source ↗
Figure 6
Figure 6. Figure 6: Sampling temperature enables inference-time control of the creativity trade-off. Moderate increases (0.25→1.0) improve diversity without sacrificing validity, but excessive temperature degrades constraint satisfaction. 4.8. How to find the most creative LLM? Our framework defines Overall Creativity as a unified met￾ric that combines convergent and divergent creativity, en￾abling a direct comparison of how … view at source ↗
Figure 7
Figure 7. Figure 7: Scatter plots of molecular LogP distributions across different target conditions. 19 [PITH_FULL_IMAGE:figures/full_fig_p019_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Conditional LogP distributions under discrete numerical constraints. GPT-3.5 GPT-4o-mini LLaMA3-8b LLaMA3-70b LLaMA3.1-8b LLaMA3.1-70b Deepseek-V3 0.0 0.2 0.4 0.6 0.8 1.0 Creativity 0.839 0.737 0.787 0.548 0.74 0.498 0.605 (a) Physicochemical tasks GPT-3.5 GPT-4o-mini LLaMA3-8b LLaMA3-70b LLaMA3.1-8b LLaMA3.1-70b Deepseek-V3 0.0 0.2 0.4 0.6 0.8 1.0 Creativity 0.853 0.802 0.843 0.689 0.79 0.528 0.623 (b) AD… view at source ↗
Figure 9
Figure 9. Figure 9: Average overall creativity of all models on physicochemical, ADMET, activity, and numerical constraint tasks. 20 [PITH_FULL_IMAGE:figures/full_fig_p020_9.png] view at source ↗
read the original abstract

Molecule generation requires satisfying multiple chemical and biological constraints while searching a large and structured chemical space. This makes it a non-binary problem, where effective models must identify non-obvious solutions under constraints while maintaining exploration to improve success by escaping local optima. From this perspective, creativity is a functional requirement in molecular generation rather than an aesthetic notion. Large language models (LLMs) can generate molecular representations directly from natural language prompts, but it remains unclear what type of creativity they exhibit in this setting and how it should be evaluated. In this work, we study the creative behavior of LLMs in molecular generation through a systematic empirical evaluation across physicochemical, ADMET, and biological activity tasks. We characterize creativity along two complementary dimensions, convergent creativity and divergent creativity, and analyze how different factors shape these behaviors. Our results indicate that LLMs exhibit distinct patterns of creative behavior in molecule generation, such as an increase in constraint satisfaction when additional constraints are imposed. Overall, our work is the first to reframe the abilities required for molecule generation as creativity, providing a systematic understanding of creativity in LLM-based molecular generation and clarifying the appropriate use of LLMs in molecular discovery pipelines.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

3 major / 2 minor

Summary. The manuscript empirically evaluates LLMs on molecule generation tasks (physicochemical, ADMET, biological activity) by reframing the required abilities as two dimensions of creativity—convergent (constraint satisfaction) and divergent (exploration to escape local optima). It reports distinct behavioral patterns, notably an increase in constraint satisfaction when additional constraints are imposed, and positions the work as the first systematic reframing of molecule generation as creativity to guide LLM use in discovery pipelines.

Significance. If the creativity dimensions are shown to be more than relabelings of existing metrics and to have predictive value for downstream discovery success, the work could help clarify when and how LLMs should be deployed in molecular pipelines. The empirical scope across multiple constraint types is a strength, but the contribution hinges on demonstrating that the framework adds explanatory or prescriptive power beyond standard validity/uniqueness/diversity benchmarks.

major comments (3)
  1. [Abstract and §4] Abstract and §4 (results): the central claim that LLMs exhibit 'distinct patterns of creative behavior' (e.g., increased constraint satisfaction with added constraints) is presented without any numerical metrics, tables, error bars, or statistical tests in the abstract and is only summarized at high level in the results. This prevents assessment of effect sizes or whether the pattern exceeds what would be expected from improved prompt adherence alone.
  2. [§3 and §5] §3 (creativity dimensions) and §5 (discussion): the operationalization of convergent vs. divergent creativity is not shown to capture functional requirements beyond standard chemical generation metrics. The reported increase in constraint satisfaction could be explained by better instruction following rather than creativity; no ablation or comparison demonstrates that the two-dimensional framing has unique predictive power for discovery success over existing benchmarks.
  3. [§4] §4 (experimental setup): the tasks and evaluation criteria used to measure the two creativity dimensions are not shown to be falsifiable or to escape circularity with conventional metrics (validity, uniqueness, diversity, constraint satisfaction). Without a control showing that the creativity lens predicts outcomes that standard metrics miss, the reframing risks being primarily terminological.
minor comments (2)
  1. [§4] Clarify the exact prompting templates and decoding strategies used for each task, as these are likely confounders for any observed 'creativity' patterns.
  2. [Introduction and §5] Add explicit comparison to prior work on LLM-based molecule generation that already reports constraint satisfaction and diversity; the 'first to reframe' claim requires a more precise novelty statement.

Simulated Author's Rebuttal

3 responses · 1 unresolved

We thank the referee for their constructive and detailed feedback, which has helped us better articulate the contributions and limitations of our work. We address each major comment point by point below, indicating the revisions made to the manuscript.

read point-by-point responses

  1. Referee: [Abstract and §4] Abstract and §4 (results): the central claim that LLMs exhibit 'distinct patterns of creative behavior' (e.g., increased constraint satisfaction with added constraints) is presented without any numerical metrics, tables, error bars, or statistical tests in the abstract and is only summarized at high level in the results. This prevents assessment of effect sizes or whether the pattern exceeds what would be expected from improved prompt adherence alone.

    Authors: We agree that the abstract would benefit from greater specificity to allow readers to assess effect sizes directly. In the revised manuscript, we have updated the abstract to reference key quantitative patterns from our experiments (e.g., the direction and consistency of constraint satisfaction improvements across tasks). We have also expanded the high-level summary in §4 to include additional tables with error bars and statistical tests (previously detailed only in the appendix) and have added a brief comparison to baseline prompting strategies to address whether the patterns exceed simple prompt adherence. These changes make the empirical support more transparent while preserving the abstract's conciseness. revision: yes

  2. Referee: [§3 and §5] §3 (creativity dimensions) and §5 (discussion): the operationalization of convergent vs. divergent creativity is not shown to capture functional requirements beyond standard chemical generation metrics. The reported increase in constraint satisfaction could be explained by better instruction following rather than creativity; no ablation or comparison demonstrates that the two-dimensional framing has unique predictive power for discovery success over existing benchmarks.

    Authors: We maintain that the two-dimensional framing organizes standard metrics to surface functional patterns—such as the counterintuitive rise in convergent creativity under higher constraint loads—that are not the primary focus of conventional benchmarks. We acknowledge that instruction following may contribute to the observed increase and that the work does not yet demonstrate unique predictive power for downstream discovery success. In the revision, we have added an ablation comparing creativity-framed prompts against standard multi-constraint prompts and expanded §5 to discuss how the framing can inform pipeline decisions, while noting that full predictive validation remains future work. revision: partial

  3. Referee: [§4] §4 (experimental setup): the tasks and evaluation criteria used to measure the two creativity dimensions are not shown to be falsifiable or to escape circularity with conventional metrics (validity, uniqueness, diversity, constraint satisfaction). Without a control showing that the creativity lens predicts outcomes that standard metrics miss, the reframing risks being primarily terminological.

    Authors: We have revised §3 to explicitly state falsifiability criteria (e.g., the framework would be falsified if LLMs showed no increase in convergent creativity with added constraints or no benefit from divergent exploration in escaping local optima). The dimensions are operationalized via existing metrics but interpreted through behavioral patterns under varying constraint regimes, which we argue provides interpretive value beyond relabeling. We have added discussion in §4 and §5 addressing potential circularity. A direct control experiment demonstrating superior prediction of downstream discovery outcomes is, however, outside the scope of this characterization-focused study. revision: partial

standing simulated objections not resolved
  • Providing an empirical control that demonstrates the creativity framework has predictive power for downstream discovery success beyond what standard metrics already capture, as the current work is limited to characterizing generation behaviors rather than validating pipeline outcomes.

Circularity Check

0 steps flagged

No circularity: empirical characterization with independent evaluation metrics

full rationale

The paper conducts a systematic empirical study of LLMs on molecule generation across physicochemical, ADMET, and biological tasks. Creativity is operationalized via two dimensions (convergent and divergent) measured through constraint satisfaction, exploration, and standard metrics such as validity, uniqueness, and diversity. No equations, derivations, or fitted parameters are presented as predictions; the central claims rest on observed performance patterns rather than self-referential definitions or reductions. Self-citations, if present, are not load-bearing for the reframing or results. The work is self-contained against external chemical generation benchmarks and does not reduce any result to its inputs by construction.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The central claim rests on the domain assumption that creativity concepts apply directly to constrained molecular search; no free parameters or invented entities are mentioned.

axioms (2)
  • domain assumption Molecule generation requires satisfying multiple chemical and biological constraints while searching a large and structured chemical space, making creativity a functional requirement.
    Opening perspective stated in the abstract.
  • domain assumption Creativity can be characterized along convergent and divergent dimensions for the purpose of evaluating LLM molecule generation.
    Core framing introduced in the abstract.

pith-pipeline@v0.9.0 · 5526 in / 1393 out tokens · 54120 ms · 2026-05-10T04:18:36.263331+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

48 extracted references · 14 canonical work pages · 3 internal anchors

  1. [1]

    arXiv preprint arXiv:2410.04197 , year =

    Cs4: Measuring the creativity of large language models automatically by controlling the number of story-writing constraints , author=. arXiv preprint arXiv:2410.04197 , year=

    work page arXiv

  2. [2]

    AI as humanity’s Salieri: Quantifying linguistic creativity of language models via systematic attribution of machine text against web text.arXiv preprint arXiv:2410.04265, 2024

    AI as Humanity's Salieri: Quantifying Linguistic Creativity of Language Models via Systematic Attribution of Machine Text against Web Text , author=. arXiv preprint arXiv:2410.04265 , year=

    work page arXiv

  3. [3]

    2024 , url=

    Li-Chun Lu and Shou-Jen Chen and Tsung-Min Pai and Chan-Hung Yu and Hung-yi Lee and Shao-Hua Sun , booktitle=. 2024 , url=

    2024

  4. [4]

    Benchmarking Language Model Creativity: A Case Study on Code Generation , author=. Proceedings of the 2025 Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers) , pages=

    2025

  5. [5]

    International conference on machine learning , pages=

    Multi-objective molecule generation using interpretable substructures , author=. International conference on machine learning , pages=. 2020 , organization=

    2020

  6. [6]

    Nature Machine Intelligence , volume=

    Multi-constraint molecular generation based on conditional transformer, knowledge distillation and reinforcement learning , author=. Nature Machine Intelligence , volume=. 2021 , publisher=

    2021

  7. [7]

    arXiv preprint arXiv:2002.03244 , year=

    Composing molecules with multiple property constraints , author=. arXiv preprint arXiv:2002.03244 , year=

    work page arXiv 2002

  8. [8]

    2012 , publisher=

    admetSAR: a comprehensive source and free tool for assessment of chemical ADMET properties , author=. 2012 , publisher=

    2012

  9. [9]

    BMC biology , volume=

    Instruction multi-constraint molecular generation using a teacher-student large language model , author=. BMC biology , volume=. 2025 , publisher=

    2025

  10. [10]

    Proceedings of the AAAI Conference on Artificial Intelligence , volume=

    Multi-objective molecular design through learning latent Pareto set , author=. Proceedings of the AAAI Conference on Artificial Intelligence , volume=

  11. [11]

    IEEE Transactions on Knowledge and Data Engineering , year=

    Large language models are in-context molecule learners , author=. IEEE Transactions on Knowledge and Data Engineering , year=

  12. [12]

    Journal of cheminformatics , volume=

    Molecular de-novo design through deep reinforcement learning , author=. Journal of cheminformatics , volume=. 2017 , publisher=

    2017

  13. [13]

    IEEE transactions on knowledge and data engineering , volume=

    Empowering molecule discovery for molecule-caption translation with large language models: A chatgpt perspective , author=. IEEE transactions on knowledge and data engineering , volume=. 2024 , publisher=

    2024

  14. [14]

    Advances in Neural Information Processing Systems , volume=

    Grammar prompting for domain-specific language generation with large language models , author=. Advances in Neural Information Processing Systems , volume=

  15. [15]

    Journal of chemical information and modeling , volume=

    ZINC 15--ligand discovery for everyone , author=. Journal of chemical information and modeling , volume=. 2015 , publisher=

    2015

  16. [16]

    International conference on machine learning , pages=

    Junction tree variational autoencoder for molecular graph generation , author=. International conference on machine learning , pages=. 2018 , organization=

    2018

  17. [17]

    International conference on machine learning , pages=

    Equivariant diffusion for molecule generation in 3d , author=. International conference on machine learning , pages=. 2022 , organization=

    2022

  18. [18]

    Advances in neural information processing systems , volume=

    Sample efficiency matters: a benchmark for practical molecular optimization , author=. Advances in neural information processing systems , volume=

  19. [19]

    Briefings in bioinformatics , volume=

    Molecular design in drug discovery: a comprehensive review of deep generative models , author=. Briefings in bioinformatics , volume=. 2021 , publisher=

    2021

  20. [20]

    Drug discovery today , volume=

    The rise of deep learning in drug discovery , author=. Drug discovery today , volume=. 2018 , publisher=

    2018

  21. [21]

    Release , volume=

    Rdkit documentation , author=. Release , volume=

  22. [22]

    Huang, T

    Therapeutics data commons: Machine learning datasets and tasks for drug discovery and development , author=. arXiv preprint arXiv:2102.09548 , year=

    work page arXiv

  23. [23]

    Bioinformatics , volume=

    ADMET-AI: a machine learning ADMET platform for evaluation of large-scale chemical libraries , author=. Bioinformatics , volume=. 2024 , publisher=

    2024

  24. [24]

    DeepSeek-V3 Technical Report

    Deepseek-v3 technical report , author=. arXiv preprint arXiv:2412.19437 , year=

    work page internal anchor Pith review Pith/arXiv arXiv

  25. [25]

    GPT-4o System Card

    Gpt-4o system card , author=. arXiv preprint arXiv:2410.21276 , year=

    work page internal anchor Pith review Pith/arXiv arXiv

  26. [26]

    The Llama 3 Herd of Models

    The llama 3 herd of models , author=. arXiv preprint arXiv:2407.21783 , year=

    work page internal anchor Pith review Pith/arXiv arXiv

  27. [27]

    The Journal of Creative Behavior , volume=

    Creativity: Yesterday, today and tomorrow , author=. The Journal of Creative Behavior , volume=. 1967 , publisher=

    1967

  28. [28]

    Proceedings of the 2024 CHI Conference on Human Factors in Computing Systems , pages=

    Art or artifice? large language models and the false promise of creativity , author=. Proceedings of the 2024 CHI Conference on Human Factors in Computing Systems , pages=

    2024

  29. [29]

    Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers) , pages=

    MacGyver: Are Large Language Models Creative Problem Solvers? , author=. Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers) , pages=

    2024

  30. [30]

    2024 IEEE LLM Aided Design Workshop (LAD) , pages=

    Creativeval: Evaluating creativity of llm-based hardware code generation , author=. 2024 IEEE LLM Aided Design Workshop (LAD) , pages=. 2024 , organization=

    2024

  31. [31]

    Educational and psychological measurement , year=

    Torrance tests of creative thinking , author=. Educational and psychological measurement , year=

  32. [32]

    ACS central science , volume=

    Automatic chemical design using a data-driven continuous representation of molecules , author=. ACS central science , volume=. 2018 , publisher=

    2018

  33. [33]

    International conference on artificial neural networks , pages=

    Graphvae: Towards generation of small graphs using variational autoencoders , author=. International conference on artificial neural networks , pages=. 2018 , organization=

    2018

  34. [34]

    Objective-reinforced generative adversarial networks (organ) for sequence generation models.arXiv preprint arXiv:1705.10843, 2017

    Objective-reinforced generative adversarial networks (organ) for sequence generation models , author=. arXiv preprint arXiv:1705.10843 , year=

    work page arXiv

  35. [35]

    Molgan: An implicit generative model for small molecular graphs.arXiv preprint arXiv:1805.11973, 2018

    MolGAN: An implicit generative model for small molecular graphs , author=. arXiv preprint arXiv:1805.11973 , year=

    work page arXiv

  36. [36]

    Graphaf: a flow-based autoregressive model for molecular graph generation.arXiv preprint arXiv:2001.09382, 2020

    Graphaf: a flow-based autoregressive model for molecular graph generation , author=. arXiv preprint arXiv:2001.09382 , year=

    work page arXiv 2001

  37. [37]

    Advances in Neural Information Processing Systems , volume=

    Equivariant flow matching with hybrid probability transport for 3d molecule generation , author=. Advances in Neural Information Processing Systems , volume=

  38. [38]

    arXiv preprint arXiv:1905.13372 , year=

    MolecularRNN: Generating realistic molecular graphs with optimized properties , author=. arXiv preprint arXiv:1905.13372 , year=

    work page arXiv 1905

  39. [39]

    Advances in neural information processing systems , volume=

    Symmetry-adapted generation of 3d point sets for the targeted discovery of molecules , author=. Advances in neural information processing systems , volume=

  40. [40]

    Geodiff: A geo- metric diffusion model for molecular conformation generation.arXiv preprint arXiv:2203.02923, 2022

    Geodiff: A geometric diffusion model for molecular conformation generation , author=. arXiv preprint arXiv:2203.02923 , year=

    work page arXiv

  41. [41]

    Frontiers in pharmacology , volume=

    Molecular sets (MOSES): a benchmarking platform for molecular generation models , author=. Frontiers in pharmacology , volume=. 2020 , publisher=

    2020

  42. [42]

    arXiv preprint arXiv:2412.14642 , year=

    TOMG-Bench: Evaluating LLMs on text-based open molecule generation , author=. arXiv preprint arXiv:2412.14642 , year=

    work page arXiv

  43. [43]

    Journal of chemical information and modeling , volume=

    Extended-connectivity fingerprints , author=. Journal of chemical information and modeling , volume=. 2010 , publisher=

    2010

  44. [44]

    Clinical Pharmacology & Therapeutics , volume=

    Trends in risks associated with new drug development: success rates for investigational drugs , author=. Clinical Pharmacology & Therapeutics , volume=. 2010 , publisher=

    2010

  45. [45]

    Communications Chemistry , volume=

    Molecular optimization using a conditional transformer for reaction-aware compound exploration with reinforcement learning , author=. Communications Chemistry , volume=. 2025 , publisher=

    2025

  46. [46]

    arXiv preprint arXiv:2410.04628 , year=

    Control large language models via divide and conquer , author=. arXiv preprint arXiv:2410.04628 , year=

    work page arXiv

  47. [47]

    Bioinformatics , volume=

    Optimizing blood--brain barrier permeation through deep reinforcement learning for de novo drug design , author=. Bioinformatics , volume=. 2021 , publisher=

    2021

  48. [48]

    Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing , pages=

    How to Make Large Language Models Generate 100\ author=. Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing , pages=

    2025