Sparse Autoencoders Are a Shared Coordinate System for Symbolic Matching

Key points

• Large language models (LLMs) hold broad knowledge in one set of weights, but lack interpretability and trustworthy reasoning, especially in high-stakes use cases.1
• A neuro-symbolic model, with its interpretable architecture and System 2 reasoning capacity, is a promising alternative if built at the scale of LLMs; however, its symbolic backbone is usually constructed with human effort, so existing neuro-symbolic models are bottlenecked by the scope of the symbol component and usually domain-specific.1
• We target its first, foundational step: matching units across these heterogeneous systems.1
• On 12 tasks across 6 kinds of symbolic system, SAE-IDF outperforms the same LLM's dense baseline on all tasks with Gemma-2-9B and all but one with Llama-3.1-8B, by up to 45 points, consistently across SAE layers and configurations.1

Sources (2)

• [1]Sparse Autoencoders Are a Shared Coordinate System for Symbolic Matching
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 14, 06:05 AM

  Our method, SAE-IDF, encodes a symbol in its native context into a sparse code with a sparse autoencoder (SAE), re-weights each dimension by its rarity and matches across systems.

  Large language models (LLMs) hold broad knowledge in one set of weights, but lack interpretability and trustworthy reasoning, especially in high-stakes use cases.

• [2]Sparse Autoencoders Are a Shared Coordinate System for Symbolic Matching
  Hugging Face Daily Papers · Oct 14, 12:00 AM · same content

Extractive summary: sentences quoted from the sources.