Yang Liu
Titles as the sources gave them. In 1 entry, with 0 checked quotations.
In the Base
Every entry they appear in.
Instructions Are Evidence, Not Objectives: Measuring Outcome Alignment of Agents
Tsinghua University researchers build TiGanEval, a 64-task benchmark where a simulated user states only part of what a software or research task needs, the rest being findable in the project. They report that eight frontier models lose 53 to 67 points of success versus a full specification, and four rounds of feedback do not close the gap.