Filtered to reasoning · clear filter
This paper develops a benchmark to measure an agent's ability to make good decisions along the way, or "taste", in long-horizon tasks. Practitioners might care because better taste can lead to improved overall performance and success in complex tasks.
This paper benchmarks a conversational AI system's ability to understand humans over long-term conversations, testing its ability to remember past interactions and reason about their context. Practitioners might care because this research helps improve AI systems' capacity for human-like understanding and conversation.
This paper investigates whether learning to predict protein structures can help improve general reasoning in large language models, and the results show that it can lead to better performance on a range of tasks, from spatial reasoning to scientific and general reasoning. Practitioners may care about this because it suggests that using non-linguistic data to fine-tune language models can improve their ability to reason and make more accurate predictions.
This paper introduces Co-RL, a framework for unsupervised multi-agent reinforcement learning that enables diverse and accurate reasoning in language and vision-language models. Practitioners can use Co-RL to improve their models' ability to reason and respond without relying on expensive ground-truth supervision.