The authors argue that LLMs are not neuro-symbolic by themselves, but they become so when coupled with a symbolic verifier or a reasoning engine (e.g., Toolformer, Program of Thoughts).
Current research categorizes NeSy systems based on how "neural" and "symbolic" components interact: The authors argue that LLMs are not neuro-symbolic
: Integrating Large Language Models (LLMs) with Knowledge Graphs to ground statistical predictions in factual, structured data. on the other hand
Traditional neural networks excel at pattern recognition and prediction tasks but often lack interpretability and common sense. Symbolic AI, on the other hand, provides a framework for representing knowledge and reasoning but can be brittle and inflexible. The authors argue that LLMs are not neuro-symbolic