Markus Dücker
Titles as the sources gave them. In 1 entry, with 0 checked quotations.
In the Base
Every entry they appear in.
Verifying Agents in Rubric-Graded Environments
According to the abstract, Markus Dücker and colleagues study how to verify AI agents' work against written rubrics. They build a dataset of 2,464 human-judged criteria from investment-banking tasks and release an open-source verifier, Gandalf, which they report matches or beats three baselines at lower cost.