The rise of large language models (LLMs) has catalyzed a diverse array of post-training techniques that are reshaping the frontiers of artificial intelligence. Approaches such as reinforcement learning with feedback -- e.g., reinforcement learning with human feedback (RLHF) (Kaufmann et al., 2023) and reinforcement learning with internal feedback (RLIF) (Zhao et al., 2025) -- and test-time scaling methods (e.g., best-of-N sampling) (Snell et al., 2024) have become pivotal in enhancing model performance. Historically, these approaches have evolved along nearly independent trajectories, with their mathematical underpinnings developed largely in isolation. This technical note reflects on some basic, yet intricate, connections among these paradigms, offering some unified perspectives that might inform and improve the design of each.

Although reinforcement learning methods offer a powerful framework for automatic skill acquisition, for practical learning-based control problems in domains such as robotics, imitation learning often provides a more convenient and accessible alternative. In particular, an interactive imitation learning method such as DAgger, which queries a near-optimal expert to intervene online to collect correction data for addressing the distributional shift challenges that afflict na\"ive behavioral cloning, can enjoy good performance both in theory and practice without requiring manually specified reward functions and other components of full reinforcement learning methods. In this paper, we explore how off-policy reinforcement learning can enable improved performance under assumptions that are similar but potentially even more practical than those of interactive imitation learning. Our proposed method uses reinforcement learning with user intervention signals themselves as rewards. This relaxes the assumption that intervening experts in interactive imitation learning should be near-optimal and enables the algorithm to learn behaviors that improve over the potential suboptimal human expert. We also provide a unified framework to analyze our RL method and DAgger; for which we present the asymptotic analysis of the suboptimal gap for both methods as well as the non-asymptotic sample complexity bound of our method. We then evaluate our method on challenging high-dimensional continuous control simulation benchmarks as well as real-world robotic vision-based manipulation tasks. The results show that it strongly outperforms DAgger-like approaches across the different tasks, especially when the intervening experts are suboptimal. Code and videos can be found on the project website: rlif-page.github.io