Despite significant progress, several challenges remain. The field is actively working on:

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).

Despite massive progress, the state of the art faces foundational hurdles:

Incorporating structured knowledge bases directly into transformer-based architectures, enhancing reasoning in NLP and CV tasks.

Towards Cognitive AI (Henry Kautz) — Explores the foundational taxonomies of NeSy.

Hak Cipta © 2017 CARIDOKUMEN Inc.