mirror of
https://github.com/hanzoai/papers.git
synced 2026-11-02 13:32:30 +00:00
Resolves high-severity review findings: anachronistic dates aligned to citation floors; arithmetic corrected to match formulas; internal contradictions collapsed to the body-supported value; LP self-references aligned to filenames; liquidity-surface repo paths scrubbed; AI-slop trimmed. PDFs rebuilt separately.
137 lines
5.1 KiB
TeX
137 lines
5.1 KiB
TeX
\section{Conclusion}
|
|
|
|
\label{sec:conclusion}
|
|
|
|
Operative provides a principled, model-agnostic framework for computer use that advances the state of the art through structured perception, formally specified actions, and hierarchical planning with safety guarantees. By decoupling the perception, planning, and execution layers, Operative enables rapid experimentation with different models and strategies while maintaining consistent safety properties. The consistent improvement over baseline computer use approaches---for example, from 22.0\% to 38.1\% on OSWorld---validates the importance of systematic framework design in this emerging capability domain.
|
|
|
|
\bibliographystyle{plain}
|
|
\begin{thebibliography}{30}
|
|
|
|
\bibitem{shi2017world}
|
|
T.~Shi, A.~Karpathy, L.~Fan, J.~Hernandez, and P.~Liang.
|
|
\newblock World of bits: An open-domain platform for web-based agents.
|
|
\newblock In \emph{ICML}, 2017.
|
|
|
|
\bibitem{schick2024toolformer}
|
|
T.~Schick, J.~Dwivedi-Yu, R.~Dess{\`i}, et~al.
|
|
\newblock Toolformer: Language models can teach themselves to use tools.
|
|
\newblock In \emph{NeurIPS}, 2023.
|
|
|
|
\bibitem{anthropic2024computer}
|
|
Anthropic.
|
|
\newblock Introducing computer use.
|
|
\newblock \url{https://www.anthropic.com/news/3-5-models-and-computer-use}, 2024.
|
|
|
|
\bibitem{openai2025operator}
|
|
OpenAI.
|
|
\newblock Operator: A new AI agent that can use a web browser.
|
|
\newblock \url{https://openai.com/index/introducing-operator/}, 2025.
|
|
|
|
\bibitem{google2024mariner}
|
|
Google DeepMind.
|
|
\newblock Project Mariner: Building AI agents for the web.
|
|
\newblock Technical report, Google, 2024.
|
|
|
|
\bibitem{yeh2009sikuli}
|
|
T.~Yeh, T.-H. Chang, and R.~Miller.
|
|
\newblock Sikuli: Using GUI screenshots for search and automation.
|
|
\newblock In \emph{UIST}, 2009.
|
|
|
|
\bibitem{hong2024cogagent}
|
|
W.~Hong, W.~Wang, Q.~Lv, et~al.
|
|
\newblock CogAgent: A visual language model for GUI agents.
|
|
\newblock In \emph{CVPR}, 2024.
|
|
|
|
\bibitem{cheng2024seeclick}
|
|
K.~Cheng, Q.~Sun, Y.~Chu, et~al.
|
|
\newblock SeeClick: Harnessing GUI grounding for advanced visual GUI agents.
|
|
\newblock \emph{arXiv preprint arXiv:2401.10935}, 2024.
|
|
|
|
\bibitem{he2024webvoyager}
|
|
H.~He, W.~Yao, K.~Ma, et~al.
|
|
\newblock WebVoyager: Building an end-to-end web agent with large multimodal models.
|
|
\newblock \emph{arXiv preprint arXiv:2401.13919}, 2024.
|
|
|
|
\bibitem{zheng2024agentstudio}
|
|
L.~Zheng, Z.~Wang, K.~Ma, et~al.
|
|
\newblock AgentStudio: A toolkit for building general virtual agents.
|
|
\newblock \emph{arXiv preprint arXiv:2403.17918}, 2024.
|
|
|
|
\bibitem{dosovitskiy2020image}
|
|
A.~Dosovitskiy, L.~Beyer, A.~Kolesnikov, et~al.
|
|
\newblock An image is worth 16x16 words: Transformers for image recognition at scale.
|
|
\newblock In \emph{ICLR}, 2021.
|
|
|
|
\bibitem{baek2019craft}
|
|
Y.~Baek, B.~Lee, D.~Han, S.~Yun, and H.~Lee.
|
|
\newblock Character region awareness for text detection.
|
|
\newblock In \emph{CVPR}, 2019.
|
|
|
|
\bibitem{playwright2024}
|
|
Microsoft.
|
|
\newblock Playwright: Reliable end-to-end testing for modern web apps.
|
|
\newblock \url{https://playwright.dev}, 2024.
|
|
|
|
\bibitem{xie2024osworld}
|
|
T.~Xie, D.~Zhang, J.~Chen, et~al.
|
|
\newblock OSWorld: Benchmarking multimodal agents for open-ended tasks in real computer environments.
|
|
\newblock \emph{arXiv preprint arXiv:2404.07972}, 2024.
|
|
|
|
\bibitem{zhou2024webarena}
|
|
S.~Zhou, F.~Xu, H.~Zhu, et~al.
|
|
\newblock WebArena: A realistic web environment for building autonomous agents.
|
|
\newblock In \emph{ICLR}, 2024.
|
|
|
|
\bibitem{liu2024visualwebbench}
|
|
J.~Liu, H.~Li, X.~Liu, et~al.
|
|
\newblock VisualWebBench: How far have multimodal LLMs evolved in web page understanding and grounding?
|
|
\newblock \emph{arXiv preprint arXiv:2404.05955}, 2024.
|
|
|
|
\bibitem{wang2024mobile}
|
|
J.~Wang, H.~Xu, J.~Ye, et~al.
|
|
\newblock Mobile-Agent: Autonomous multi-modal mobile device agent with visual perception.
|
|
\newblock \emph{arXiv preprint arXiv:2401.16158}, 2024.
|
|
|
|
\bibitem{hanzo2026gateway}
|
|
Hanzo AI Research.
|
|
\newblock Hanzo LLM Gateway: Unified proxy for 100+ AI providers.
|
|
\newblock Technical report, Hanzo AI, 2026.
|
|
|
|
\bibitem{koh2024visualwebarena}
|
|
J.~Koh, R.~Lo, L.~Jang, et~al.
|
|
\newblock VisualWebArena: Evaluating multimodal agents on realistic visual web tasks.
|
|
\newblock In \emph{ACL}, 2024.
|
|
|
|
\bibitem{deng2024mind2web}
|
|
X.~Deng, Y.~Gu, B.~Zheng, et~al.
|
|
\newblock Mind2Web: Towards a generalist agent for the web.
|
|
\newblock In \emph{NeurIPS}, 2023.
|
|
|
|
\bibitem{kim2024language}
|
|
G.~Kim, P.~Baldi, and S.~McAleer.
|
|
\newblock Language agents with reinforcement learning for strategic play in the Werewolf game.
|
|
\newblock \emph{arXiv preprint arXiv:2310.18602}, 2024.
|
|
|
|
\bibitem{zheng2024gpt4vision}
|
|
B.~Zheng, B.~Gou, J.~Kil, et~al.
|
|
\newblock GPT-4V(ision) is a generalist web agent, if grounded.
|
|
\newblock \emph{arXiv preprint arXiv:2401.01614}, 2024.
|
|
|
|
\bibitem{rawles2024androidworld}
|
|
C.~Rawles, S.~Li, D.~Wiegand, et~al.
|
|
\newblock AndroidWorld: A dynamic benchmarking environment for autonomous agents.
|
|
\newblock \emph{arXiv preprint arXiv:2405.14573}, 2024.
|
|
|
|
\bibitem{lu2024weblinx}
|
|
X.~Lu, S.~Gao, Z.~Chen, et~al.
|
|
\newblock WebLINX: Real-world website navigation with multi-turn dialogue.
|
|
\newblock In \emph{ICML}, 2024.
|
|
|
|
\bibitem{putta2024agent}
|
|
S.~Putta, E.~Mills, N.~Garg, et~al.
|
|
\newblock Agent Q: Advanced reasoning and learning for autonomous AI agents.
|
|
\newblock \emph{arXiv preprint arXiv:2408.07199}, 2024.
|
|
|
|
\end{thebibliography}
|
|
|