AI & ML
impact 16
Instruction Complexity Induces Positional Collapse in Adversarial LLM Evaluation
Instruction Complexity Induces Positional Collapse in Adversarial LLM Evaluation arXiv:2604.27249v1 Announce Type: cross Abstract: When instructed to underperform on multiple-choice evaluations, do language models engag…
Why it matters
The instruction angle matters most here. If confirmed, expect ripple effects across complexity and related sectors.