R1

“DeepSeek of the v3/R1 era would already start losing coherence occasionally at t=0.7 with no other samplers”

“R1 truly taught other models step by step how to do chain-of-thought reasoning”

“DS's R1 counts as one, but the duration was too short”

“DeepSeek was the first to show off R1-style "thinking models". That's one of the reasons why it was such a huge hit and required way less compute.”

“DeepSeek's ability to create efficiency-oriented stuff like V3 and R1 is directly related to the fact that the people behind it are a group doing quantization; those quantization folks are always thinking about how to squeeze more returns from fewer resources”