Hong Zhou

Invited Speaker from Industry

Talk

Title: Integrating Statistical Intelligence and Logical Intelligence to better meet the future challenges

Abstract: While LLMs have broken new ground in NLP, their limitations are notable: they rely on statistical, correlation-driven learning, reasoning, and generation. Despite generating coherent text and handling complex dialogues, LLMs lack real-world perception, true knowledge and proof—precluding in-depth comprehension of the complex world and the development of open-ended intelligence. Looking ahead, the synergy between data knowledge and formalization will take the intelligence to better meet the challenge of the future world.

This presentation takes "The Monkeys and Peaches Distribution Problem" as an entry point to analyze the hierarchical relationship between computational efficiency and logical reasoning: starting with brute-force enumeration (using existing knowledge) to solve problems, progressing to exploring efficient complexity-reduction strategies, and ultimately extending to problem-solving logics in complex environments. On this basis, three core hypotheses are proposed: Expand the Boundary of Knowledge, Generalize Efficient Representation, and Develop a Better World Model.

Looking for the future work, integration of Topos, symbols to improve the capability of mathematics competition and prover, FSL combination to improve the levels of software auto designing and testing, even the non-computability hypotheses for the future bionics and quantum… AI plus formalization bring us to the broader theories and unknown worlds.