The authors built ACED-Bench around a hidden probabilistic graphical model (PGM) with categorical variables, a causal graph, and conditional probability tables.
New benchmark tests whether LLM agents can actively investigate causal systems
ACED-Bench uses hidden probabilistic graphical models to evaluate whether agents design experiments, interpret finite samples, and commit to causal conclusions.
Big Tech
Yiwei Chen · Amila Weerasinghe · Siddharth Biswal · Utkrisht Rajkumar · Olcay Boz · Taylor Berg-Kirkpatrick · +1 more
UC San Diego · Amazon
Research Digest··2 min read
Chen et al.
Why this paper
From Amazon and UC San Diego
In one line
ACED-Bench evaluates whether LLM agents can actively gather causal evidence and act on it; current models fail at experiment selection, evidence retention, and stopping.
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.
§