The authors formalize the problem of Online Agentic Test-Time Training (OaTTT), where an LLM agent encounters a single-pass stream of tasks, executes each task once, and receives a verification signal only at the end of the episode (success or failure).
Agent improves by distilling its own verified experiences during deployment
ASCENT enables online test-time training of LLM agents by self-distilling successful trajectories into persistent weight updates without external supervision.
Academic
Haodong Lu · Dong Gong
University of New South Wales (UNSW Sydney)
Research Digest··3 min read
The authors study Online Agentic Test-Time Training (OaTTT), where a deployed LLM agent executes each task once and uses the single verified trajectory as the only learning signal.
Why this paper
From University of New South Wales (UNSW Sydney)
In one line
Deployed LLM agents can improve by self-distilling their own verified execution trajectories into persistent weight updates without a separate training phase.
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 (3 noted)
- ·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.
§