Pre-trained large-scale language models (LLMs) excel at producing coherent articles, yet their outputs may be untruthful, toxic, or fail to align with user expectations. Current approaches focus on using reinforcement learning with human feedback (RLHF) to improve model alignment, which works by transforming coarse human preferences of LLM outputs into a feedback signal that guides the model learning process. However, because this approach operates on sequence-level feedback, it lacks the precision to identify the exact parts of the output affecting user preferences. To address this gap, we propose a method to enhance LLM alignment through fine-grained token-level supervision. Specifically, we ask annotators to minimally edit less preferred responses within the standard reward modeling dataset to make them more favorable, ensuring changes are made only where necessary while retaining most of the original content. The refined dataset is used to train a token-level reward model, which is then used for training our fine-grained Proximal Policy Optimization (PPO) model. Our experiment results demonstrate that this approach can improve LLM performance by up to 5.1% in terms of win rate against the reference model, compared with the traditional PPO model.

Jointly fine-tuning a Pre-trained Language Model (PLM) on a pre-defined set of tasks with in-context instructions has been proven to improve its generalization performance, allowing us to build a universal language model that can be deployed across task boundaries. In this work, we explore for the first time whether this attractive property of in-context instruction learning can be extended to a scenario in which tasks are fed to the target PLM in a sequential manner. The primary objective of so-called lifelong in-context instruction learning is to improve the target PLM’s instance- and task-level generalization performance as it observes more tasks. DynaInst, the proposed method to lifelong in-context instruction learning, achieves noticeable improvements in both types of generalization, nearly reaching the upper bound performance obtained through joint training.

Off-Policy reinforcement learning has been the driving force for the state-of-the-art conversational AIs leading to more natural human-agent interactions and improving the user satisfaction for goal-oriented agents. However, in large-scale commercial settings, it is often challenging to balance between policy improvements and experience continuity on the broad spectrum of applications handled by such system. In the literature, off-policy evaluation and guard-railing on aggregate statistics has been commonly used to address this problem. In this paper, we propose method for curating and leveraging high-precision samples sourced from historical regression incident reports to validate, safe-guard, and improve policies prior to the online deployment. We conducted extensive experiments using data from a real-world conversational system and actual regression incidents. The proposed method is currently deployed in our production system to protect customers against broken experiences and enable long-term policy improvements.