Causal thinking enables humans to understand not just what is seen, but why it happens. To bring this ability to modern AI systems, we introduce visual causal discovery, the task of inferring visually verifiable direct causal dependencies among entities in a single static image. We construct the Visual Causal Graph dataset (VCG-32K), a large-scale collection of over 32,000 images annotated with entity-level causal graphs, and develop Causight, a causally aware vision-language model for this task. Our training recipe combines data curation from VCG-32K, Tree-of-Causal-Thought (ToCT) reasoning synthesis, and reinforcement learning with a graph-structured causal reward. Experiments show that Causight outperforms GPT-4.1, achieving over a threefold performance boost (21% absolute gain).

Recent advances in large language models have highlighted the critical need for precise control over model outputs through predefined constraints. While existing methods attempt to achieve this through either direct instruction-response synthesis or preferential response optimization, they often struggle with constraint understanding and adaptation. This limitation becomes particularly evident when handling fine-grained constraints, leading to either hallucination or brittle performance. We introduce Generative Adversarial Policy Optimization (GAPO), a novel framework that combines GAN-based training dynamics with an encoder-only reward model to progressively learn and adapt to increasingly complex constraints. GAPO leverages adversarial training to automatically generate training samples of varying difficulty while utilizing the encoder-only architecture to better capture prompt-response relationships. Extensive experiments demonstrate GAPO’s superior performance across multiple benchmarks, particularly in scenarios requiring fine-grained constraint handling, where it significantly outperforms existing methods like PPO, DPO, and KTO. Our results suggest that GAPO’s unique approach to preferential prompt learning offers a more robust and effective solution for controlling LLM outputs.