New benchmark measures structural novelty in language model outputs
Researchers introduced AlienBench, a benchmark on arXiv that evaluates whether large language models produce structurally novel outputs for creative tasks like writing, design, and idea generation.
Read the full story at arXiv cs.CL ↗
Timeline · 1 report
- 2026-11-09 05:00 · arXiv cs.CL
AlienBench: A Benchmark of Structural Novelty in Language Model Outputs