04:00
2026-07-13
arxiv.org
artificial-intelligence
WILDTRACE: Benchmarking Natural Evidence Trails in Long-Context Reasoning
Researchers introduced WILDTRACE, a benchmark of 481 tasks over 214 long-form sources like technical reports and literary narratives, designed to test AI models' ability to integrate naturally dispers…