Sometimes our anthropocentric assumptions about how intelligence "should" work (like using language for reasoning) may be holding AI back. Letting AI reason in its own native "language" in latent space could unlock new capabilities, improving reasoning over Chain of Thought. https://arxiv.org/pdf/2412.06769
Comments
It’s worth noting, though, that their results were worse than CoT for grade school math, but better than CoT for random logic/planning puzzles.
Are you open to creating a post on that? And potentially ask others to share their favorite people?
Thank you for considering this request!