Announcement
‘The Alien Space of Science’ AI research-idea sampler accepted to NeurIPS 2026
The authors say their system ranks plausible ideas by how unlikely existing research communities are to propose them.
TLDR
The paper’s authors describe a method that combines concepts from research papers, then favors ideas scored as both coherent and unlikely to come from existing author communities. They report that, using 16,068 peer-reviewed papers on large language models, it produced more diverse sets of research ideas than frontier AI model baselines while matching or exceeding their per-idea quality in the authors’ evaluations.
Combined views
2.4K
3 Sources, first seen ago
