Turning sand into intelligence. One bitter lesson at a time.

p(doom) is an AGI research lab. Pretraining unlocked one direction to scale. As did self-play, RLHF and RLVR. We work on the scalable methods of the future: learning from real-world human behavior, continual improvement through deployment, architectures that carry memory forever. Removing scaling bottlenecks, one bitter lesson at a time.