MultiSTEVE-1s: A Model Zoo and Interpretability Suite for Instruction-Following Vision Agents

A striking case of goal misgeneralisation was previously observed in OpenAI's Minecraft agent VPT: it killed villagers standing under leaves, mistaking them for tree trunks. Although this agent was released publicly, enabling white-box interpretability research, few open-weight model organisms of misalignment exist outside the LLM space. We release MultiSTEVE-1s: a model zoo of 140 fine-tuned versions of VPT, over 1,000 training checkpoints, and an interpretability suite for analysing them. We use the STEVE-1 training procedure to add instruction-following capabilities to VPT with fixed hyperparameters and controlled variations in training randomness. We demonstrate the utility of MultiSTEVE-1s by showcasing the research it enables. First, some training runs differ only by a least-significant bit flip in a single initialised weight. Others differ in the full randomness for weight initialisation and data. Yet, the single bit-flip setting produces agents that act nearly as differently from each other as the full randomness ones. Second, we use our interpretability suite to show that several known VPT attention heads retain their roles after STEVE-1 fine-tuning, while attention strength to the same behaviourally meaningful frame can vary substantially across agents and checkpoints. Finally, although the agents are similarly capable on in-distribution tasks, their out-of-distribution behaviour of villager killing can differ substantially: in one setting, one agent kills villagers less than 5\% of the time, while another kills them nearly 50\% of the time. Our results show the value of studying multiple similarly trained agents rather than acting like a behavioural biology lab with only one rat. Code: https://anonymous.4open.science/r/multisteve1s-anonymous-76D4/