Shiwei Xu

Shuang Wu, Liwen Zhu, Tao Yang et al.

This paper presents an innovative framework that integrates Large Language Models (LLMs) with an external Thinker module to enhance the reasoning capabilities of LLM-based agents. Unlike augmenting LLMs with prompt engineering, Thinker directly harnesses knowledge from databases and employs various optimization techniques. The framework forms a reasoning hierarchy where LLMs handle intuitive System-1 tasks such as natural language processing, while the Thinker focuses on cognitive System-2 tasks that require complex logical analysis and domain-specific knowledge. Our framework is presented using a 9-player Werewolf game that demands dual-system reasoning. We introduce a communication protocol between LLMs and the Thinker, and train the Thinker using data from 18800 human sessions and reinforcement learning. Experiments demonstrate the framework's effectiveness in deductive reasoning, speech generation, and online game evaluation. Additionally, we fine-tune a 6B LLM to surpass GPT4 when integrated with the Thinker. This paper also contributes the largest dataset for social deduction games to date.

Sergiu Bursuc, Christian Johansen, Shiwei Xu

Automated verification of security protocols based on dynamic root of trust, typically relying on protected hardware such as TPM, involves several challenges that we address in this paper. We model the semantics of trusted computing platforms (including CPU, TPM, OS, and other essential components) and of associated protocols in a classical process calculus accepted by ProVerif. As part of the formalization effort, we introduce new equational theories for representing TPM specific platform states and dynamically loaded programs. Formal models for such an extensive set of features cannot be readily handled by ProVerif, due especially to the search space generated by unbounded extensions of TPM registers. In this context we introduce a transformation of the TPM process, that simplifies the structure of the search space for automated verification, while preserving the security properties of interest. This allows to run ProVerif on our proposed models, so we can derive automatically security guarantees for protocols running in a dynamic root of trust context.