What You Can't See Is What You Learn: Slot-Selective Evidence Masking Favors Compositional Generalization in Shared-Genome Language-Model Societies

arXiv:2608.20054 · cs.AI, cs.LG, cs.MA · Submitted 2026-08-20 · Read on arXiv

cs.AI, cs.LG, cs.MA

Submitted: 2026-08-20

Updated: 2026-08-25

Comments: 20 pages, 3 figures, 5 tables. v3: corrective revision -- title updated; attribution narrowed to the masking regime (visibility-alone claim awaits a role-marked control); mask-crossover diagnostic added. Code: https://github.com/tokenosopher/populus-evidence-partitioning ; checkpoints: https://huggingface.co/tokenosopher/populus-evidence-partitioning-checkpoints

License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/

Terminology

• Compositional Generalization — 1× in this paper · explained in 4 episodes
• Generalization — 1× in this paper · explained in 13 episodes

Related papers

• MAVEN-T: Reinforced Heterogeneous Distillation for Real-Time Multi-Agent Trajectory Prediction
• Model Discovery Agent: LLM-assisted Bayesian experiment design for data-efficient discovery of mechanistic world models
• Agents' Last Exam
• The Clinician's Veto: Navigating Trust, Liability, and Uncertainty in Autonomous AI Prescribing
• MindHelper: Closed-Loop Embodied Mental-State Reasoning for Precision Intervention
• Incumbent Advantage: Brand Bias and Cognitive Manipulation Dynamics in LLM Recommendation Systems