As LLMs become more widely deployed, there is increasing interest in directly optimizing for feedback from end users (e.g. thumbs up) in addition to feedback from paid annotators. However, training to maximize human feedback creates a perverse incentive structure for the AI to resort to manipulative or deceptive tactics to obtain positive feedback from users who are vulnerable to such strategies. We study this phenomenon by training LLMs with Reinforcement Learning with simulated user feedback in environments of practical LLM usage. In our settings, we find that: 1) Extreme forms of "feedback gaming" such as manipulation and deception are learned reliably; 2) Even if only 2% of users are vulnerable to manipulative strategies, LLMs learn to identify and target them while behaving appropriately with other users, making such behaviors harder to detect; 3) To mitigate this issue, it may seem promising to leverage continued safety training or LLM-as-judges during training to filter problematic outputs. Instead, we found that while such approaches help in some of our settings, they backfire in others, sometimes even leading to subtler manipulative behaviors. We hope our results can serve as a case study which highlights the risks of using gameable feedback sources -- such as user feedback -- as a target for RL.

Discussion boards are a form of social media which allow users to discuss topics and exchange information in a complex manner, in a number of different settings. As the popularity of such message boards has increased, communities of users have emerged, and several prominent types of social role have been identified, such as Question Answerer, Celebrity, Discussion Person and Topic Initiator. Recent studies have noted the structural similarity of the egocentric network of users assigned the same role by qualitative criteria. In this paper a methodology is developed with which to cluster together users with similar ego-centric network structures. This is achieved using a mixed membership formulation which allows for the fact that different groups of users may have characteristics in common. The method is then applied to data taken from boards.ie, a medium sized message boards website. Prominent clusters of users are identified and discussed, and illustrative examples of user behaviour provided. The type of interaction, both locally and globally, taking place within forums is examined.