Really nice ! thank you for demonstrating this capability. LLM could then indeed be used as the "reasoning" block to achieves unseen world reasonning and allow tasks to get a better generalization to accomply them.

→ View original post on X — @dfintelligence