Bingguang Hao
1* · 2026
Read Bingguang Hao by 1* (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-2601-01498.txt.
Description
Hao and 13 co-authors introduce HardGen, a pipeline that makes hard training examples for agents that use outside tools. Current training sets are often too easy and too alike, missing the hidden links between steps in a real task. HardGen starts instead from cases where agents already failed.
It builds a graph of program interfaces based on those failures, samples hard action traces, and creates advanced abstract tools from them. Hard questions and checked step-by-step answers follow, with test feedback guiding more rounds. A small model trained on the resulting data beat several leading larger models on tool-use tests. Code and data are promised as open releases.
Catalog record
| CREATOR | 1* |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 62 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2601-01498