Are large language models (LLMs) biased towards text generated by LLMs over text authored by humans, leading to possible anti-human bias? Utilizing a classical experimental design inspired by employment discrimination studies, we tested widely-used LLMs, including GPT-3.5 and GPT4, in binary-choice scenarios. These involved LLM-based agents selecting between products and academic papers described either by humans or LLMs under identical conditions. Our results show a consistent tendency for LLM-based AIs to prefer LLM-generated content. This suggests the possibility of AI systems implicitly discriminating against humans, giving AI agents an unfair advantage.

To understand the goals and goal representations of AI systems, we carefully study a pretrained reinforcement learning policy that solves mazes by navigating to a range of target squares. We find this network pursues multiple context-dependent goals, and we further identify circuits within the network that correspond to one of these goals. In particular, we identified eleven channels that track the location of the goal. By modifying these channels, either with hand-designed interventions or by combining forward passes, we can partially control the policy. We show that this network contains redundant, distributed, and retargetable goal representations, shedding light on the nature of goal-direction in trained policy networks.