Debate: Symbolic Recursion in RLMs is the Key to Compositional Generalization

lateinteraction · x · 2026-08-06

Amidst debates over recent ARC-AGI results, a researcher argues that the claim of achieving results "without symbols" is misleading. They clarify that Recursive Language Models (RLMs) are definitionally built on "symbolic recursion" over symbolic references, making them neurosymbolic. This architecture is precisely why RLMs excel at compositional generalization, something a vanilla Transformer cannot easily match.

Related event: LLM Architecture Debate: Equipped Transformers No Longer Pure DNNs(5 posts)→

Original post →

More from Research

Research channel →