A new paper refutes the "Ingenia Theorem," which claimed to prove the impossibility of AGI via ML using complexity theory. The refutation argues the original proof is flawed due to an undefined "human-level classifier" and a swap in formal definitions.
This paper provides a theoretical explanation for the efficiency of diffusion models in learning the score function for high-dimensional data supported on low-dimensional manifolds. It identifies a "collapse-and-refine" mechanism driven by the geometry of the score function, where the denoising map projects onto the data manifold and refines the intrinsic density.
This research introduces the Tool-Augmented Markov Decision Process (TA-MDP) to formally model multimodal agentic decision-making, addressing theoretical gaps in reinforcement fine-tuning for Large Vision-Language Models (LVLMs). It specifically investigates how composite verifiable rewards affect GRPO convergence and why training on small datasets generalizes to out-of-distribution domains for agentic LVLMs.