Scaling Agentic Capabilities via Grounded Interaction Synthesis CCS: Computing methodologies Natural language generation CCS: Computing methodologies Intelligent agents
Wenhang Shi, Jinhao Dong, Yiren Chen, Zhe Zhao, Shuqing Bian, Wei Lu, Xiaoyong Du · 2026
Read Scaling Agentic Capabilities via Grounded Interaction Synthesis CCS: Computing methodologies Natural language generation CCS: Computing methodologies Intelligent agents by Wenhang Shi, Jinhao Dong, Yiren Chen, Zhe Zhao, Shuqing Bian, Wei Lu, Xiaoyong Du (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2606-02001.txt.
Description
Shi and co-authors build training data for agents from grounded real interaction. Agent skill hinges on using diverse real-world tools. Quality interaction data drives that capability directly. Human annotation is too costly to scale for it. Current methods let chat models synthesize everything. Unconstrained generation drifts into biased random sampling instead.
Internal model priors then dominate the synthetic data. Real domain diversity and difficulty are missed. Long-horizon tasks of high fidelity are rare too. Their GAIS framework grounds synthesis in two phases. Protocol-anchored environments derive from real Model Context Protocol sources. Tasks are built to match those environments closely. The result is more diverse, harder agent training material.
Catalog record
| CREATOR | Wenhang Shi, Jinhao Dong, Yiren Chen, Zhe Zhao, Shuqing Bian, Wei Lu, Xiaoyong Du |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 67 KB |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2606.02001