The authors treat each existing software workflow as a hidden program mapping structured inputs to outputs.
Existing scientific software can generate verified training tasks for terminal agents
The authors reconstruct 500 executable workflows from observed behavior, then use successful reconstructions to improve a terminal-agent model.
Independent
Zhongzhi Li · Yucheng Shi · Zongxia Li · Junyao Yang · Ruhan Wang · Yu Wang · +5 more
Research Digest··2 min read
Li and colleagues introduce software-in-the-loop reconstruction, a self-supervised method that turns existing scientific programs into terminal-agent tasks without requiring a separately written reference solution for every task.
Why this paper
Independent
In one line
Existing scientific software workflows provide self-supervised training data for terminal agents via software-in-the-loop reconstruction.
What we could check
- ·No code link found
- ·No weights link found
- ·No dataset link found
- ·No compute details found
- ✓Limitations stated by the authors (2 noted)
- ✓Reports numbers on named benchmarks (2 benchmarks)
Observed from the paper text and links we have. Absence here means we did not find it, not that it does not exist.
§