Shoggoth

The shoggoth is the AI world’s picture of an alien mind behind a friendly mask. The creature comes from H. P. Lovecraft’s At the Mountains of Madness (1936), whose shoggoths are formless beings bred as servants that eventually overthrew their masters. In December 2022 the user @TetraspaceWest drew a shoggoth labeled GPT-3 next to the same creature wearing a small smiley-face mask, labeled GPT-3 + RLHF: reinforcement learning from human feedback, the training step that turns a raw language model into a polite assistant. The meme spread quickly through the AI world in 2023, and “shoggoth” became shorthand for whatever a model is underneath its assistant persona.

Its creator said it was not meant to suggest the model is evil or conscious, only that its inner workings are unknown. That is the point of Grown, not built: training shapes what a system does without revealing how it does it, so a friendly manner is evidence about behavior, not about what lies underneath. The worry is not that the mask is a lie, but that nobody can yet check. Mechanistic interpretability is the attempt to look behind the mask; the Seeing inside AI path follows how far it has got.

It is a modern addition to the old stories that recognized parts of the risk long before AI; see AI risk in old stories.