Pith. sign in

REVIEW 4 major objections 6 minor 51 references

SCION turns scientific research into an executable operating system: a Science Agent Meta-Harness compiles intent into Research Execution Plans and coordinates agents, tools, and memory, beating existing research-agent baselines on multi-st

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · grok-4.5

2026-07-11 23:25 UTC pith:HPPIMOJH

load-bearing objection Useful OS packaging for agentic science work; the architecture is coherent, but the causal claim that SCION’s design drives the wins is still underdetermined. the 4 major comments →

arxiv 2607.03863 v2 pith:HPPIMOJH submitted 2026-07-04 cs.CL

Rethinking Scientific Discovery in the Agentic Era

Yining Zheng , Yuxin Wang , Jiahao Lu , Shicheng Fang , Weiyi Wang , Yongzhuo Yang , Bowen Li , Haochen Ma
show 22 more authors
Chen Hu Bowen Chen Yang Wang Huanhui Chen Yitong Chen Jiajun Chen Zhiyuan Li Yanlin Li Zhuo Yang Qifeng Wu Jiaying He Zhijie Jinluo Xiaohu Xu Yi Feng Juncheng Qian Yizhou Chen Yang Cheng Tong Zhu Tianlei Ying Hongyu Yu Hongjun Xiang Xipeng Qiu
This is my paper
classification cs.CL
keywords agentic AIscientific discoveryResearch Execution Planmulti-agent systemsAI4Scienceinverse designbatch active searchepistemic memory
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

Most AI-for-science systems still leave humans as the glue between problem formulation, literature, models, simulation, validation, and reuse. This paper introduces SCION as an agentic scientific operating system—an organizational nexus—that makes the research workflow itself a first-class computational object. A Science Agent acting as a Meta-Harness compiles high-level intent into a Research Execution Plan with stages, dependencies, verification checkpoints, tools, expected artifacts, and fallbacks, then runs hierarchical multi-agent execution with selective context, governed delegation, and layered epistemic memory. Discovery is cast as target-conditioned inverse search and, when the target is hidden, as batch active search under a finite experimental budget. On scientific reading, idea generation, multi-property molecule generation, and antibody screening, SCION is reported to outperform pipeline and agentic research baselines especially where decomposition, verification, refinement, and memory reuse matter.

Core claim

The paper claims that scientific discovery improves when AI is no longer a set of isolated tools but a coordinated operational layer: compiling scientific intent into Research Execution Plans, executing them through a hierarchical multi-agent runtime with verification and recovery, and storing trajectories as reusable epistemic memory yields more auditable, recoverable work and higher performance than existing autonomous research agents on multi-stage scientific tasks.

What carries the argument

The Research Execution Plan (REP): a structured, machine-operable plan that compiles high-level scientific intent into staged objectives, dependencies, verification checkpoints, tool requirements, expected artifacts, and fallback conditions. The Science Agent Meta-Harness uses the REP to coordinate agents, tools, and layered memory as an approximate inverse-search procedure over candidates and experimental trajectories.

Load-bearing premise

The reported gains come from SCION’s organizational design rather than from stronger or differently prompted model backbones, unequal baseline adaptations, or the preferences of the LLM judge used for idea novelty.

What would settle it

Fix one shared model backbone and identical task adapters, then ablate REP compilation, layered memory, and governed verification on the multi-property molecule-generation and one-round antibody-screening suites; if success rate and F1 drop to match strong baselines, the organizational claim holds; if performance stays high, the OS design is not what carries the gains.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • Human scientists can move from manual tool dispatch toward strategy, value constraints, and high-level judgment while the system handles coordination and recovery.
  • Materials analysis, multi-property molecule design, and antibody screening can be run as governed inverse-search or budgeted batch-active-search procedures with auditable branches.
  • Failed attempts, intermediate artifacts, and decision traces become reusable project memory instead of discarded session noise.
  • Evaluation of AI for science expands from single-model accuracy to system-level recoverability, auditability, and knowledge reuse.
  • Coupling this operating-system model with automated experimentation and wet labs can form a cyber-physical infrastructure for science.

Where Pith is reading between the lines

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

  • If REP-style plans become common, labs will need shared schemas for verification checkpoints and provenance before multi-group collaboration scales cleanly.
  • The same batch-active-search harness should transfer to other sparse-hit discovery settings (for example catalyst or assay screening) without inventing a new architecture.
  • The largest reported lifts on multi-constraint molecule tasks suggest one-shot generators remain weak where validity filters and trade-offs must be enforced online.
  • A stripped planner-plus-memory layer on a fixed backbone would be a strong control experiment to isolate how much of the gain is truly organizational.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

4 major / 6 minor

Summary. This paper presents SCION, an agentic scientific operating system that treats research workflows as first-class computational objects. A Science Agent Meta-Harness compiles high-level intent into Research Execution Plans (REPs) specifying stages, dependencies, verification checkpoints, tools, artifacts, and fallbacks, then coordinates hierarchical profile-driven agents, selective context, governed delegation, and layered L1–L3 epistemic memory. Discovery is formalized as target-conditioned inverse search (known targets) and as batch active search under finite budgets (hidden targets). Applications are sketched for materials analysis, multi-property molecule design, and antibody screening. Experiments on CMPhysBench scientific reading, idea novelty (LLM pairwise), multi-property molecule generation, and one-round antibody screening report gains over pipeline and agentic research baselines, attributed especially to decomposition, verification, refinement, and memory reuse.

Significance. If the architectural claims hold under controlled evaluation, SCION would be a meaningful systems contribution to AI4Science: it reframes the bottleneck as coordination, provenance, and recoverability rather than isolated model accuracy, and it offers a concrete OS-like substrate (REP, profile kernel, layered memory, governed delegation) that many current research agents lack. The inverse-search / batch-active-search framing is useful as an operational lens even though it is standard optimization language rather than a new theorem. Strengths include a coherent multi-agent design (Tables 1–4), explicit mapping of applications to formulations (Table 5), and multi-task empirical coverage. The work is significant primarily as systems architecture plus empirical systems evaluation; its lasting value depends on whether gains can be attributed to the OS design rather than backbone or evaluation artifacts.

major comments (4)
  1. Section 7 (Implementation details and all four benchmarks): The central causal claim—that SCION’s Meta-Harness / REP / multi-agent OS design drives outperformance in decomposition, verification, refinement, and memory reuse—is not isolated. Benchmarks use different backbones (Kimi 2.5 for reading; nex-n1.1 for idea/molecule/antibody; Claude Opus 4.6 and GPT-5.5 for ARIS), and pipeline agents are adapted “to specified tasks.” Without fixed-backbone, fixed-protocol controls and component ablations (REP off, memory off, verification off, single-agent vs hierarchical), absolute score advantages cannot be attributed to the organizational design. This is load-bearing for the abstract’s and conclusion’s OS-level claim.
  2. §7.2 and Table 6: Idea novelty is evaluated solely by GPT-5.5 pairwise judgments, with SCION win rates of 100% vs AutoResearchClaw and ScienceClaw and 98.33% vs EvoScientist. Extreme margins under a single LLM judge are consistent with format, verbosity, or stylistic preference rather than OS-driven novelty. The paper needs either human expert ratings, multi-judge agreement, or an objective novelty proxy (e.g., literature-overlap / citation-grounded distinctness), plus disclosure of prompt templates and output-length controls, before novelty can support the architecture claim.
  3. §5.2 and §7.4 / Figure 7: Batch active search is formulated as multi-round nonmyopic maximization of expected positives under budget T=Hb (Eqs. 17–18), yet the experiment evaluates only a single intermediate round with synthetic feedback, a fixed top-10% labeling rule, a 1:1 train/test split, and a single F1 (0.370) without error bars, multi-seed variance, multi-round budget curves, or comparison to standard active-search baselines (e.g., myopic top-p, nonmyopic search). This under-tests the hidden-target formulation that the paper presents as a core contribution.
  4. §7.3 / Figure 6: Multi-property molecule success rates favor SCION on average, but the paper does not report candidate-set sizes, generation budgets, verifier definitions, or whether baselines had equal access to the same property predictors, validity filters, and iterative refine loops. Without matched tool surfaces and equal query budgets, higher success rates may reflect tool orchestration privileges rather than the REP/memory architecture per se. Equalize tools and report per-objective constraint satisfaction rates.
minor comments (6)
  1. Abstract and §1: Several compound words appear without spaces (e.g., “systemsremainfragmented,” “ScientificCollaborativeInnovation”). Clean typesetting throughout.
  2. §5.1 Eqs. (12)–(13): f^{-1}_SA is defined as an informal runtime procedure then placed inside an arg min over procedure class F_inv; clarify that this is schematic notation, not a solved optimization, to avoid over-reading as a closed-form result.
  3. Figures 5–7: Report numerical values in tables as well as bar charts; add error bars or multi-run statistics where stochastic generation or sampling is involved.
  4. Related Work §2.2: Position more clearly against general LLM-agent OS work (e.g., AIOS) and against self-driving-lab literature beyond the brief future-work note in the conclusion.
  5. Reproducibility: No code, prompts, REP schemas, or dataset splits are released. At minimum, provide REP schema examples, evaluation scripts, and antibody split definitions.
  6. Table 1 vs text: “idea” and “reading” are listed as shared specialists while sci/exec are project-scoped; briefly justify why ideation and literature are not project-private when projects often need confidential literature and idea branches.

Circularity Check

1 steps flagged

No load-bearing circular derivation: empirical wins use external metrics; inverse-search math is a framing rename, not a forced prediction.

specific steps
  1. renaming known result [Section 5.1, Eqs. (12)–(13) and composition after Eq. (13)]
    "From this perspective, the role of Science Agent is to operationalize a target-conditioned approximate inverse procedure. We denote this realized procedure informally as X^*=f^{-1}_SA(Y^*). ... f^{-1}_SA ≈ MemoryUpdate◦Verify◦Execute◦Generate◦Plan. ... Concretely, these stages map onto SCION’s runtime: planning corresponds to REP compilation by the coordinator, generation primarily to the idea and sci agents, execution to the exec agent and external tools, verification to critic checkpoints and domain verifiers, and memory update to the layered memory subsystem."

    The paper presents discovery as solving an inverse problem via f^{-1}_SA, but defines f^{-1}_SA as the already-described SCION loop (plan/generate/execute/verify/memory). The math renames the architecture rather than deriving a new inverse from first principles; it does not force the later empirical success rates, so this is mild framing circularity, not a fitted prediction.

full rationale

SCION is an architecture-and-systems paper. Its central empirical claims (SEED/accuracy on CMPhysBench, pairwise idea novelty, multi-property molecule success rates, antibody F1) are scored against external task metrics and held-out or reference labels, not quantities that equal fitted inputs by construction. The Target-conditioned Inverse Search and batch active search sections (Eqs. 6–21) recast plan–generate–execute–verify–memory as an approximate inverse operator f^{-1}_SA and a history-conditioned policy; that is renaming/framing of the runtime, not a closed-form derivation that forces the reported numbers. There is no uniqueness theorem imported from the authors, no parameter fit re-labeled as prediction of the same quantity, and no self-citation chain that is the sole support for the outperformance claim. Residual mild circularity risk is limited to (i) LLM-as-judge novelty (Table 6) and (ii) adapted baselines under mixed backbones—evaluation-design concerns, not definitional circularity of a derivation. Score 2 for the inverse-search renaming and judge-mediated ideation metric; not 0 only because those steps are soft self-referential framing, not because the main results reduce to their inputs.

Axiom & Free-Parameter Ledger

5 free parameters · 5 axioms · 5 invented entities

This is a systems/architecture paper, not a derivation from axioms. Load-bearing content is mostly design postulates, evaluation protocol choices, and standard search formalisms. The free parameters are experimental and implementation choices that affect the headline comparisons; invented entities are the named runtime abstractions that carry the central claim.

free parameters (5)
  • Model backbone assignment per benchmark = Kimi 2.5; nex-n1.1; Claude Opus 4.6; GPT-5.5
    Scientific Reading uses Kimi 2.5; Idea Maker / Molecule / Antibody use nex-n1.1; ARIS uses Claude Opus 4.6 and GPT-5.5. These choices can dominate system comparisons.
  • Antibody positive threshold = top 10%
    Top 10% candidates are labeled positive; F1 is computed against top 10% selected. Threshold choice defines the metric target.
  • Train/test split for antibody screening = 1:1
    Even 1:1 split of candidate pools into history vs next-round set determines available feedback and evaluation set.
  • LLM judge and pairwise protocol for idea novelty = GPT-5.5; 30 queries × 2 orders
    GPT-5.5 pairwise Win/Tie/Lose with order swapping is the sole novelty metric; judge choice and prompt are free evaluation parameters.
  • Hard multi-property molecule thresholds and constraint set
    Success rate depends on unspecified numerical thresholds for BBBP, pLogP, QED, mutagenicity, HIA and hard chemical filters.
axioms (5)
  • ad hoc to paper Scientific discovery can be usefully compiled into an explicit Research Execution Plan with stages, dependencies, verification checkpoints, and fallbacks.
    Central design premise of Sections 1 and 5; not independently proven, but assumed as the right OS abstraction.
  • domain assumption Role-specialized hierarchical agents with selective context and layered memory reduce interference and improve long-horizon scientific work versus flat pipelines.
    Architecture justification in Section 4; common multi-agent assumption, treated as design truth rather than ablated rigorously.
  • domain assumption Target-conditioned inverse search / batch active search are appropriate formal models for known-target and hidden-target discovery under SCION.
    Section 5 maps SCION onto standard inverse design and active search objectives; useful framing, not a derived necessity.
  • domain assumption LLM pairwise novelty judgments and thresholded property success rates are adequate proxies for scientific idea quality and design utility.
    Evaluation sections 7.2–7.3 rely on these proxies without human expert validation or wet-lab confirmation.
  • standard math Standard optimization/search identities: minimize discrepancy of f(X) to Y*, maximize expected positives under batch budget.
    Equations (8)–(20) restate inverse design and nonmyopic batch active search.
invented entities (5)
  • SCION (Scientific Collaborative Innovation with Agentic Organizational Nexus) no independent evidence
    purpose: Name the proposed agentic scientific operating system / organizational nexus.
    Primary system introduced by the paper; existence is definitional to the contribution.
  • Research Execution Plan (REP) no independent evidence
    purpose: Compile high-level scientific intent into executable staged plans with verification and fallbacks.
    Core abstraction claimed to make research a first-class computational object.
  • Science Agent Meta-Harness no independent evidence
    purpose: Orchestrate tools, agents, artifacts, and memory above task-level harnesses.
    Named orchestration layer that converts goals into adaptive search procedures.
  • Profile-driven agent kernel (project-main, sci, idea, reading, exec, lark) no independent evidence
    purpose: Realize specialized scientific labor over one common runtime via profiles for tools, skills, memory, and scope.
    Section 4 taxonomy; system design entities without external independent validation.
  • Layered epistemic memory (L1/L2/L3 scopes) no independent evidence
    purpose: Preserve continuity, durable knowledge, and audit/artifact traces without prompt contamination.
    Memory architecture is a postulated runtime structure central to reuse claims.

pith-pipeline@v1.1.0-grok45 · 25372 in / 3977 out tokens · 33256 ms · 2026-07-11T23:25:08.418477+00:00 · methodology

0 comments
read the original abstract

Artificial intelligence has advanced scientific discovery, but most AI4Science systems remain fragmented tools that rely on humans to coordinate problem formulation, literature grounding, model use, simulation, validation, and knowledge reuse. This paper presents \textbf{SCION (Scientific Collaborative Innovation with Agentic Organizational Nexus)}, an agentic scientific operating system that acts as an \textbf{organizational nexus}. Through a Science Agent serving as a \textbf{Meta-Harness}, SCION connects scientific tasks, tools, agents, artifacts, and memory, transforming research into an executable, auditable, and reusable operational process. At its core is the \textbf{Research Execution Plan (REP)}, which compiles high-level scientific intent into staged objectives, dependencies, verification checkpoints, tool requirements, expected artifacts, and fallback conditions. SCION further integrates hierarchical multi-agent execution, profile-driven specialization, selective context construction, governed delegation, and layered epistemic memory to support long-horizon scientific work. We formulate discovery under SCION as \textbf{Target-conditioned Inverse Search} and extend it to hidden-target settings through batch active search under finite experimental budgets. Applications in materials analysis, molecule design, and protein or antibody screening, together with experiments on scientific reading, idea generation, molecule generation, and antibody screening, show that SCION outperforms existing autonomous research-agent baselines, especially in decomposition, verification, refinement, and memory reuse. Overall, SCION shifts AI from isolated tools toward a coordinated operational layer for traceable and reusable scientific innovation.

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

51 extracted references · 3 canonical work pages

  1. [1]

    Scientific discovery in the age of artificial intelligence.Nature, 620(7972): 47–60, 2023

    Hanchen Wang, Tianfan Fu, Yuanqi Du, Wenhao Gao, Kexin Huang, Ziming Liu, Payal Chandak, Shengchao Liu, Peter Van Katwyk, Andreea Deac, et al. Scientific discovery in the age of artificial intelligence.Nature, 620(7972): 47–60, 2023

    2023

  2. [2]

    Highly accurate protein structure prediction with alphafold.nature, 596(7873):583–589, 2021

    John Jumper, Richard Evans, Alexander Pritzel, Tim Green, Michael Figurnov, Olaf Ronneberger, Kathryn Tunyasu- vunakool, Russ Bates, Augustin Žídek, Anna Potapenko, et al. Highly accurate protein structure prediction with alphafold.nature, 596(7873):583–589, 2021

    2021

  3. [3]

    De novo design of protein structure and function with rfdiffusion.Nature, 620(7976):1089–1100, 2023

    Joseph L Watson, David Juergens, Nathaniel R Bennett, Brian L Trippe, Jason Yim, Helen E Eisenach, Woody Ahern, Andrew J Borst, Robert J Ragotte, Lukas F Milles, et al. De novo design of protein structure and function with rfdiffusion.Nature, 620(7976):1089–1100, 2023

    2023

  4. [4]

    Skarlinski, Sam Cox, Jon M

    Michael D. Skarlinski, Sam Cox, Jon M. Laurent, James D. Braza, Michaela Hinks, Michael J. Hammerling, Manvitha Ponnapati,SamuelG.Rodriques,andAndrewD.White. Languageagentsachievesuperhumansynthesisofscientific knowledge, 2024. URLhttps://arxiv.org/abs/2409.13740

    Pith/arXiv arXiv 2024

  5. [5]

    Scaling deep learning for materials discovery.Nature, 624(7990):80–85, 2023

    Amil Merchant, Simon Batzner, Samuel S Schoenholz, Muratahan Aykol, Gowoon Cheon, and Ekin Dogus Cubuk. Scaling deep learning for materials discovery.Nature, 624(7990):80–85, 2023

    2023

  6. [6]

    The (r)evolution of scientific workflows in the agentic ai era: Towards autonomous science, 2025

    Woong Shin, Renan Souza, Daniel Rosendo, Frédéric Suter, Feiyi Wang, Prasanna Balaprakash, and Rafael Ferreira da Silva. The (r)evolution of scientific workflows in the agentic ai era: Towards autonomous science, 2025. URL https://arxiv.org/abs/2509.09915

    Pith/arXiv arXiv 2025

  7. [7]

    Towards end-to-end automation of ai research.Nature, 651(8107):914–919, 2026

    Chris Lu, Cong Lu, Robert Tjarko Lange, Yutaro Yamada, Shengran Hu, Jakob Foerster, David Ha, and Jeff Clune. Towards end-to-end automation of ai research.Nature, 651(8107):914–919, 2026

    2026

  8. [8]

    Role of the human-in-the-loop in emerging self-driving laboratories for heterogeneous catalysis.Nature Catalysis, 8(1):13–19, 2025

    Christoph Scheurer and Karsten Reuter. Role of the human-in-the-loop in emerging self-driving laboratories for heterogeneous catalysis.Nature Catalysis, 8(1):13–19, 2025

    2025

  9. [9]

    A review of prominent paradigms for llm-based agents: Tool use, planning (including rag), and feedback learning

    Xinzhe Li. A review of prominent paradigms for llm-based agents: Tool use, planning (including rag), and feedback learning. InProceedings of the 31st international conference on computational linguistics, pages 9760–9779, 2025

    2025

  10. [10]

    Towards a science of scaling agent systems.arXiv preprint arXiv:2512.08296, 2025

    Yubin Kim, Ken Gu, Chanwoo Park, Chunjong Park, Samuel Schmidgall, A Ali Heydari, Yao Yan, Zhihan Zhang, Yuchen Zhuang, Mark Malhotra, et al. Towards a science of scaling agent systems.arXiv preprint arXiv:2512.08296, 2025

    Pith/arXiv arXiv 2025

  11. [11]

    Crystal graph convolutional neural networks for an accurate and interpretable prediction of material properties.Physical review letters, 120(14):145301, 2018

    Tian Xie and Jeffrey C Grossman. Crystal graph convolutional neural networks for an accurate and interpretable prediction of material properties.Physical review letters, 120(14):145301, 2018

    2018

  12. [12]

    Benchmarking materials property prediction methods: the matbench test set and automatminer reference algorithm.npj Computational Materials, 6(1): 138, 2020

    Alexander Dunn, Qi Wang, Alex Ganose, Daniel Dopp, and Anubhav Jain. Benchmarking materials property prediction methods: the matbench test set and automatminer reference algorithm.npj Computational Materials, 6(1): 138, 2020

    2020

  13. [13]

    From ai for science to agentic science: A survey on autonomous scientific discovery.arXiv preprint arXiv:2508.14111, 2025

    Jiaqi Wei, Yuejin Yang, Xiang Zhang, Yuhan Chen, Xiang Zhuang, Zhangyang Gao, Dongzhan Zhou, Guangshuai Wang, Zhiqiang Gao, Juntai Cao, et al. From ai for science to agentic science: A survey on autonomous scientific discovery.arXiv preprint arXiv:2508.14111, 2025

    arXiv 2025

  14. [14]

    The future of self-driving laboratories: from human in the loop interactive ai to gamification

    Holland Hysmith, Elham Foadian, Shakti P Padhy, Sergei V Kalinin, Rob G Moore, Olga S Ovchinnikova, and Mahshid Ahmadi. The future of self-driving laboratories: from human in the loop interactive ai to gamification. Digital Discovery, 3(4):621–636, 2024

    2024

  15. [15]

    Chemcrow: Augmenting large-language models with chemistry tools.arXiv preprint arXiv:2304.05376, 2023

    Andres M Bran, Sam Cox, Oliver Schilter, Carlo Baldassari, Andrew D White, and Philippe Schwaller. Chemcrow: Augmenting large-language models with chemistry tools.arXiv preprint arXiv:2304.05376, 2023

    Pith/arXiv arXiv 2023

  16. [16]

    Autonomouschemicalresearchwithlargelanguage models.Nature, 624(7992):570–578, 2023

    DaniilABoiko,RobertMacKnight,BenKline,andGabeGomes. Autonomouschemicalresearchwithlargelanguage models.Nature, 624(7992):570–578, 2023

    2023

  17. [17]

    Ai-researcher: Autonomous scientific innovation, 2025

    Jiabin Tang, Lianghao Xia, Zhonghang Li, and Chao Huang. Ai-researcher: Autonomous scientific innovation, 2025. URLhttps://arxiv.org/abs/2505.18705. 23

    Pith/arXiv arXiv 2025

  18. [18]

    The ai scientist-v2: Workshop-level automated scientific discovery via agentic tree search, 2025

    Yutaro Yamada, Robert Tjarko Lange, Cong Lu, Shengran Hu, Chris Lu, Jakob Foerster, Jeff Clune, and David Ha. The ai scientist-v2: Workshop-level automated scientific discovery via agentic tree search, 2025. URL https://arxiv.org/abs/2504.08066

    Pith/arXiv arXiv 2025

  19. [19]

    Internagent: When agent becomes the scientist – building closed-loop system from hypothesis to verification, 2025

    InternAgent Team, Bo Zhang, Shiyang Feng, Xiangchao Yan, Jiakang Yuan, Runmin Ma, Yusong Hu, Zhiyin Yu, Xiaohan He, Songtao Huang, Shaowei Hou, Zheng Nie, Zhilong Wang, Jinyao Liu, Tianshuo Peng, Peng Ye, Dongzhan Zhou, Shufei Zhang, Xiaosong Wang, Yilan Zhang, Meng Li, Zhongying Tu, Xiangyu Yue, Wangli Ouyang, Bowen Zhou, and Lei Bai. Internagent: When a...

    Pith/arXiv arXiv 2025

  20. [20]

    Autoresearchclaw: Fullyautonomousresearchfromideatopaper,2026

    Jiaqi Liu, Peng Xia, Siwei Han, Shi Qiu, Letian Zhang, Guiming Chen, Haoqin Tu, Xinyu Yang, Jiawei Zhou, Hongtu Zhu, Yun Li, Jiaheng Zhang, Yuyin Zhou, Zeyu Zheng, Cihang Xie, Mingyu Ding, and Huaxiu Yao. Autoresearchclaw: Fullyautonomousresearchfromideatopaper,2026. URL https://github.com/aiming-lab/ AutoResearchClaw

    2026

  21. [21]

    Evoscientist: Towards multi-agent evolving ai scientists for end-to-end scientific discovery, 2026

    Yougang Lyu, Xi Zhang, Xinhao Yi, Yuyue Zhao, Shuyu Guo, Wenxiang Hu, Jan Piotrowski, Jakub Kaliski, Jacopo Urbani, Zaiqiao Meng, Lun Zhou, and Xiaohui Yan. Evoscientist: Towards multi-agent evolving ai scientists for end-to-end scientific discovery, 2026. URLhttps://arxiv.org/abs/2603.08127

    arXiv 2026

  22. [22]

    Dingjie Song, Hanrong Zhang, Dawei Liu, Yixin Liu, Zongxia Li, Zhengqing Yuan, Siqi Zhang, and Lichao Sun. Dr. claw: An ai research workspace from idea to paper, 2026. URLhttps://github.com/OpenLAIR/dr-claw

    2026

  23. [23]

    Wang, Lee Marom, Subhadeep Pal, Rachel K

    Fiona Y. Wang, Lee Marom, Subhadeep Pal, Rachel K. Luu, Wei Lu, Jaime A. Berkovich, and Markus J. Buehler. Autonomous agents coordinating distributed discovery through emergent artifact exchange, 2026. URLhttps: //arxiv.org/abs/2603.14312

    arXiv 2026

  24. [24]

    Aris: Autonomous research via adversarial multi-agent collaboration, 2026

    Ruofeng Yang, Yongcan Li, and Shuai Li. Aris: Autonomous research via adversarial multi-agent collaboration, 2026. URLhttps://arxiv.org/abs/2605.03042

    Pith/arXiv arXiv 2026

  25. [25]

    Aios: Llm agent operating system.arXiv preprint arXiv:2403.16971, 2024

    Kai Mei, Xi Zhu, Wujiang Xu, Wenyue Hua, Mingyu Jin, Zelong Li, Shuyuan Xu, Ruosong Ye, Yingqiang Ge, and Yongfeng Zhang. Aios: Llm agent operating system.arXiv preprint arXiv:2403.16971, 2024

    Pith/arXiv arXiv 2024

  26. [26]

    A survey on the memory mechanism of large language model-based agents.ACM T ransactions on Information Systems, 43(6):1–47, 2025

    Zeyu Zhang, Quanyu Dai, Xiaohe Bo, Chen Ma, Rui Li, Xu Chen, Jieming Zhu, Zhenhua Dong, and Ji-Rong Wen. A survey on the memory mechanism of large language model-based agents.ACM T ransactions on Information Systems, 43(6):1–47, 2025

    2025

  27. [27]

    Agenthallu: Benchmarking automated hallucination attribution of llm-based agents.arXiv preprint arXiv:2601.06818, 2026

    Xuannan Liu, Xiao Yang, Zekun Li, Peipei Li, and Ran He. Agenthallu: Benchmarking automated hallucination attribution of llm-based agents.arXiv preprint arXiv:2601.06818, 2026

    arXiv 2026

  28. [28]

    Evaluation and benchmarking of llm agents: A survey

    Mahmoud Mohammadi, Yipeng Li, Jane Lo, and Wendy Yip. Evaluation and benchmarking of llm agents: A survey. InProceedings of the 31st ACM SIGKDD Conference on Knowledge Discovery and Data Mining V . 2, pages 6129–6139, 2025

    2025

  29. [29]

    Hart, Michal Jahnatek, Roman V

    Stefano Curtarolo, Wahyu Setyawan, Gus L.W. Hart, Michal Jahnatek, Roman V. Chepulskii, Richard H. Taylor, Shidong Wang, Junkai Xue, Kesong Yang, Ohad Levy, Michael J. Mehl, Harold T. Stokes, Denis O. Demchenko, and Dane Morgan. Aflow: An automatic framework for high-throughput materials discovery.Computational Materials Science, 58:218–226, 2012. ISSN 09...

    doi:10.1016/j.commatsci.2012.02.005 2012

  30. [30]

    Anubhav Jain, Shyue Ping Ong, Geoffroy Hautier, Wei Chen, William Davidson Richards, Stephen Dacek, Shreyas Cholia, Dan Gunter, David Skinner, Gerbrand Ceder, and Kristin A. Persson. Commentary: The materials project: A materialsgenomeapproachtoacceleratingmaterialsinnovation.APL Materials,1(1):011002,072013. ISSN2166-532X. doi: 10.1063/1.4812323. URLhttp...

    doi:10.1063/1.4812323

  31. [31]

    Graph networks as a universal machine learning framework for molecules and crystals.Chemistry of Materials, 31(9):3564–3572, April 2019

    Chi Chen, Weike Ye, Yunxing Zuo, Chen Zheng, and Shyue Ping Ong. Graph networks as a universal machine learning framework for molecules and crystals.Chemistry of Materials, 31(9):3564–3572, April 2019. ISSN 1520-5002. doi: 10.1021/acs.chemmater.9b01294. URLhttp://dx.doi.org/10.1021/acs.chemmater.9b01294

    doi:10.1021/acs.chemmater.9b01294 2019

  32. [32]

    A universal graph deep learning interatomic potential for the periodic table.Nature Computational Science, 2(11):718–728, November 2022

    Chi Chen and Shyue Ping Ong. A universal graph deep learning interatomic potential for the periodic table.Nature Computational Science, 2(11):718–728, November 2022. ISSN 2662-8457. doi: 10.1038/s43588-022-00349-3. URL http://dx.doi.org/10.1038/s43588-022-00349-3. 24

    doi:10.1038/s43588-022-00349-3 2022

  33. [33]

    Strubbe, and Isaac Tamblyn

    Kevin Ryczko, David A. Strubbe, and Isaac Tamblyn. Deep learning and density-functional theory.Physical Review A, 100(2), August 2019. ISSN 2469-9934. doi: 10.1103/physreva.100.022512. URLhttp://dx.doi.org/10.1103/ PhysRevA.100.022512

    work page doi:10.1103/physreva.100.022512 2019

  34. [34]

    Yi, Conrard Giresse Tetsassi Feugmo, Anna Klinkova, Kyla Sask, Agustinus Kristiadi, Boyu Wang, Elizabeth R

    Yan Sun, Lianghong Chen, Zihao Jing, Yan Yi Li, Dongkyu Kim, Jing-Yan Gao, Reza Noroozi, Grace Y. Yi, Conrard Giresse Tetsassi Feugmo, Anna Klinkova, Kyla Sask, Agustinus Kristiadi, Boyu Wang, Elizabeth R. Gillies, Kun Ping Lu, HaoTian Harvey Shi, and Pingzhao Hu. Generative ai for the design of molecules: Advances and challenges.Journal of Chemical Infor...

    work page doi:10.1021/acs.jcim.5c02234 2025

  35. [35]

    A survey of generative ai for de novo drug design: New frontiers in molecule and protein generation, 2024

    Xiangru Tang, Howard Dai, Elizabeth Knight, Fang Wu, Yunyang Li, Tianxiao Li, and Mark Gerstein. A survey of generative ai for de novo drug design: New frontiers in molecule and protein generation, 2024. URLhttps: //arxiv.org/abs/2402.08703

    Pith/arXiv arXiv 2024

  36. [36]

    Junction tree variational autoencoder for molecular graph generation, 2019

    Wengong Jin, Regina Barzilay, and Tommi Jaakkola. Junction tree variational autoencoder for molecular graph generation, 2019. URLhttps://arxiv.org/abs/1802.04364

    Pith/arXiv arXiv 2019

  37. [37]

    Molecular de novo design through deep reinforcement learning, 2017

    Marcus Olivecrona, Thomas Blaschke, Ola Engkvist, and Hongming Chen. Molecular de novo design through deep reinforcement learning, 2017. URLhttps://arxiv.org/abs/1704.07555

    Pith/arXiv arXiv 2017

  38. [38]

    Richard Bickerton, Gaia V

    G. Richard Bickerton, Gaia V. Paolini, Jeremy Besnard, Sorel Muresan, and Andrew L. Hopkins. Quantifying the chemical beauty of drugs.Nature Chemistry, 4(2):90–98, feb 2012. doi: 10.1038/nchem.1243. URLhttps: //doi.org/10.1038/nchem.1243

    doi:10.1038/nchem.1243 2012

  39. [39]

    Dauparas, I

    J. Dauparas, I. Anishchenko, N. Bennett, H. Bai, R. J. Ragotte, L. F. Milles, B. I. M. Wicky, A. Courbet, R. J. de Haas, N. Bethel, P. J. Y. Leung, T. F. Huddy, S. Pellock, D. Tischer, F. Chan, B. Koepnick, H. Nguyen, A. Kang, B. Sankaran, A. K. Bera, N. P. King, and D. Baker. Robust deep learning–based protein sequence design using proteinmpnn. Science, ...

    doi:10.1126/science.add2187 2022

  40. [40]

    Kyro, Tianyin Qiu, and Victor S

    Gregory W. Kyro, Tianyin Qiu, and Victor S. Batista. A model-centric review of deep learning for protein design,

  41. [41]

    URLhttps://arxiv.org/abs/2502.19173

    Pith/arXiv arXiv

  42. [42]

    Alphafolding: 4d diffusion for dynamic protein structure prediction with reference and motion guidance, 2024

    Kaihui Cheng, Ce Liu, Qingkun Su, Jun Wang, Liwei Zhang, Yining Tang, Yao Yao, Siyu Zhu, and Yuan Qi. Alphafolding: 4d diffusion for dynamic protein structure prediction with reference and motion guidance, 2024. URL https://arxiv.org/abs/2408.12419

    Pith/arXiv arXiv 2024

  43. [43]

    Schneider, Andrew W

    Jun Cheng, Guido Novati, Joshua Pan, Clare Bycroft, Akvil˙e Žemgulyt˙e, Taylor Applebaum, Alexander Pritzel, Lai Hong Wong, Michal Zielinski, Tobias Sargeant, Rosalia G. Schneider, Andrew W. Senior, John Jumper, Demis Hassabis, Pushmeet Kohli, and Žiga Avsec. Accurate proteome-wide missense variant effect prediction with alphamissense.Science, 381(6664):e...

    doi:10.1126/science.adg7492 2023

  44. [44]

    Kimi-audio technical report.arXiv preprint arXiv:2504.18425, 2025

    Ding Ding, Zeqian Ju, Yichong Leng, Songxiang Liu, Tong Liu, Zeyu Shang, Kai Shen, Wei Song, Xu Tan, Heyi Tang, et al. Kimi-audio technical report.arXiv preprint arXiv:2504.18425, 2025

    Pith/arXiv arXiv 2025

  45. [45]

    nex-n1.1, 2025

    Yuxuan Cai, Lu Chen, Qiaoling Chen, Yuyang Ding, Liwen Fan, Wenjie Fu, Yufei Gao, Honglin Guo, Pinxue Guo, Zhenhua Han, et al. nex-n1.1, 2025

    2025

  46. [46]

    Cmphysbench: A benchmark for evaluating large language models in condensed matter physics, 2025

    Weida Wang, Dongchen Huang, Jiatong Li, Tengchao Yang, Ziyang Zheng, Di Zhang, Dong Han, Benteng Chen, Binzhao Luo, Zhiyu Liu, Kunling Liu, Zhiyuan Gao, Shiqi Geng, Wei Ma, Jiaming Su, Xin Li, Shuchen Pu, Yuhan Shui, Qianjia Cheng, Zhihao Dou, Dongfei Cui, Changyong He, Jin Zeng, Zeke Xie, Mao Su, Dongzhan Zhou, Yuqiang Li, Wanli Ouyang, Yunqi Cai, Xi Dai...

    Pith/arXiv arXiv 2025

  47. [47]

    Gellmo: Generalizing large language models for multi-property molecule optimization, 2025

    Vishal Dey, Xiao Hu, and Xia Ning. Gellmo: Generalizing large language models for multi-property molecule optimization, 2025. URLhttps://arxiv.org/abs/2502.13398

    Pith/arXiv arXiv 2025

  48. [48]

    Deep batch active learning for drug discovery

    Michael Bailey, Saeed Moayedpour, Ruijiang Li, Alejandro Corrochano-Navarro, Alexander Kötter, Lorenzo Kogler- Anele, Saleh Riahi, Christoph Grebner, Gerhard Hessler, Hans Matter, Marc Bianciotto, Pablo Mas, Ziv Bar-Joseph, 25 and Sven Jager. Deep batch active learning for drug discovery. January 2024. doi: 10.7554/elife.89679.2. URL http://dx.doi.org/10.7...

    work page doi:10.7554/elife.89679.2 2024

  49. [49]

    Llms for bayesian optimization in scientific domains: Are we there yet?,

    Rushil Gupta, Jason Hartford, and Bang Liu. Llms for bayesian optimization in scientific domains: Are we there yet?,

  50. [50]

    URLhttps://arxiv.org/abs/2509.21403

    arXiv

  51. [51]

    Efficient nonmyopic active search

    Shali Jiang, Gustavo Malkomes, Geoff Converse, Alyssa Shofner, Benjamin Moseley, and Roman Garnett. Efficient nonmyopic active search. In Doina Precup and Yee Whye Teh, editors,Proceedings of the 34th International Conference on Machine Learning, volume 70 ofProceedings of Machine Learning Research, pages 1714–1723. PMLR, 06–11 Aug 2017. URLhttps://procee...

    2017