The authors built long-horizon coding tasks through a weak-to-strong synthesis pipeline.
Coding assistants falter when long projects involve less technical users
SWE-Journey tests agents on evolving repository work and finds a large performance gap between simulated software architects and non-coders.
Chinese Tech
Hexuan Deng · Yue Wang · Wenyu Jiang · Cheng Yang · Haolin Yang · Zhaohua Zhang · +10 more
Tencent Hy AI Data · Beijing Zhongguancun Academy
Research Digest··2 min read
Deng et al.
Why this paper
From Tencent Hy AI Data and Beijing Zhongguancun Academy
In one line
Coding assistants pass over 75% of functionality tests with architect users but under 25% with non-coder users on long-horizon, multi-turn SWE-Journey tasks.
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
Observed from the paper text and links we have. Absence here means we did not find it, not that it does not exist.
§