Offline evaluation of LLMs is crucial in understanding their capacities, though current methods remain underexplored in existing research. In this work, we focus on the offline evaluation of the chain-of-thought capabilities and show how to optimize LLMs based on the proposed evaluation method. To enable offline feedback with rich knowledge and reasoning paths, we use knowledge graphs (KGs) (e.g., Wikidata5M) to provide feedback on the generated chain of thoughts. Due to the heterogeneity between LLM reasoning and KG structures, direct interaction and feedback from knowledge graphs on LLM behavior are challenging, as they require accurate entity linking and grounding of LLM-generated chains of thought in the KG. To address the above challenge, we propose an offline chain-of-thought evaluation framework, OCEAN, which models chain-of-thought reasoning in LLMs as a Markov Decision Process (MDP), and evaluate the policy’s alignment with KG preference modeling. To overcome the reasoning heterogeneity and grounding problems, we leverage on-policy KG exploration and reinforcement learning to model a KG policy that generates token-level likelihood distributions for LLM-generated chain-of-thought reasoning paths, simulating KG reasoning preference. Then we incorporate the knowledge-graph feedback on the validity and alignment of the generated reasoning paths into inverse propensity scores and propose KG-IPS estimator. Theoretically, we prove the unbiasedness of the proposed KG-IPS estimator and provide a lower bound on its variance. With the off-policy evaluated value function, we can directly enable off-policy optimization to further enhance chain-of-thought alignment. Our empirical study shows that OCEAN can be efficiently optimized for generating chain-of-thought reasoning paths with higher estimated values without affecting LLMs’ general abilities in downstream tasks or their internal knowledge.

Aligning large language models (LLMs) with human values is imperative to mitigate potential adverse effects resulting from their misuse. Drawing from the sociological insight that acknowledging all parties' concerns is a key factor in shaping human values, this paper proposes a novel direction to align LLMs by themselves: social scene simulation. To achieve this, we present MATRIX, a novel social scene simulator that emulates realistic scenes around a user's input query, enabling the LLM to take social consequences into account before responding. MATRIX serves as a virtual rehearsal space, akin to a Monopolylogue, where the LLM performs diverse roles related to the query and practice by itself. To inject this alignment, we fine-tune the LLM with MATRIX-simulated data, ensuring adherence to human values without compromising inference speed. We theoretically show that the LLM with MATRIX outperforms existing methods under mild assumptions. Finally, extensive experiments validate that our method outperforms over 10 baselines across 4 benchmarks. As evidenced by 875 user ratings, our tuned 13B-size LLM exceeds GPT-4 in aligning with human values. See our project page at https://shuotang123.github.io/MATRIX.

Vision-based emergent communication (EC) aims to learn to communicate through sketches and demystify the evolution of human communication. Ironically, previous works neglect multi-round interaction, which is indispensable in human communication. To fill this gap, we first introduce a novel Interactive Sketch Question Answering (ISQA) task, where two collaborative players are interacting through sketches to answer a question about an image. To accomplish this task, we design a new and efficient interactive EC system, which can achieve an effective balance among three evaluation factors, including the question answering accuracy, drawing complexity and human interpretability. Our experimental results demonstrate that multi-round interactive mechanism facilitates tar- geted and efficient communication between intelligent agents. The code will be released.