The three-part model of how generative AI works includes: (1) a monster from pre-training on unfiltered internet text (like an H.P. Lovecraft creature), (2) a human mask of supervised fine-tuning with human feedback, and (3) reinforcement learning from human feedback creating a happy face facade, suggesting the model's helpfulness is a thin layer over an alien intelligence underneath.

definitionpending

Speaker

Melanie Mitchell

Evidence Quote

“this is a picture of a what's called a shath which is a a mythical monster from an HP Lovecraft story that's the pre-training part it's a monster um because you know the text on the Internet is monstrous um then there's this supervised fine tuning which is like the human mask and then this little happy face thing which is the what's called reinforcement learning from Human feedback”

Source

The Future of Artificial Intelligence— Santa Fe Institute
Created: 8/12/2026, 6:19:21 PM

Your notes

Loading notes...