The authors treated base language models as continuation models rather than incomplete chatbots.
Base models can reveal agent skills before instruction tuning
ASAP adapts interactive evaluation scaffolds to completion models, exposing how agent-like behavior changes across pretraining checkpoints.
Chinese Tech
Jiarui Qin · Junjie Huang · Yunjia Xi · Di Yin · Yong Yu · Weinan Zhang · +2 more
Tencent Youtu Lab · Shanghai Jiao Tong University
Research Digest··2 min read
Qin et al.
Why this paper
From Tencent Youtu Lab and Shanghai Jiao Tong University
In one line
ASAP enables end-to-end agentic evaluation of pretrained base models without weight updates, distinguishing checkpoints while exposing changes in exploration, editing, testing, and recovery.
What we could check
- ·No code link found
- ·No weights link found
- ·No dataset link found
- ·No compute details found
- ·No stated limitations found
- ·No benchmark numbers found
Observed from the paper text and links we have. Absence here means we did not find it, not that it does not exist.
§