The literature on large language model (LLM) alignment and the literature on social media echo chambers treat two apparently distinct phenomena. This paper argues they are not distinct. Reinforcement learning from human feedback (RLHF) in LLMs and engagement-optimizing algorithms in social media platforms are operationally equivalent instances of a single mechanism: optimization against an approval signal that covaries with user validation rather than epistemic accuracy. The term sycophancy, developed in the LLM alignment literature, describes this mechanism precisely. The term echo chamber, dominant in the social media literature, describes its informational effect. Collapsing both under a common mechanistic category clarifies the causal structure, raises the level of analysis in both fields, and opens symmetrical design responses that neither literature has yet foregrounded.
Janer Tittarelli Javier Ignacio (Fri,) studied this question.