Epoch AI measures how fast the cost of a given level of AI performance is falling across five benchmarks covering math, science and games of skill: about 47% per quarter, or 13x per year, since 2023, faster than electricity, compute, batteries or DNA sequencing ever fell.
What technical, fundamental reason is there for the crash in price?
Smaller models are inherently faster, and use less gpus to fit inside, and less gpu time per training step. Newer models are smarter at smaller size than older models, and so also come up with correct answer in fewer tokens.
There are 15 labs accross the world competing without patent restriction for software building. It’s orders of magnitude faster than Moore’s law, because its highly competitive, and software has massive “compiler” resources thrown at it, in investor/government frenzy.
Smaller models are inherently faster, and use less gpus to fit inside, and less gpu time per training step. Newer models are smarter at smaller size than older models, and so also come up with correct answer in fewer tokens.
There are 15 labs accross the world competing without patent restriction for software building. It’s orders of magnitude faster than Moore’s law, because its highly competitive, and software has massive “compiler” resources thrown at it, in investor/government frenzy.