if the frontier models are commoditized, compute concentration matters even more

if you can train better models for fewer flops, compute concentration matters even more

compute is the primary means of production of the future and owning more will always be good

roon

strategic gpu reserve

https://x.com/tszzl/status/1883076766232936730

disagree. is oai copium post. gpu is not spice. floating points depreciate fast.

history: once we figured the math and the usecases, deep learning models were shipped on $5-8 pieces of hardware that you can hook to any microcontroller/SoC. large models will follow same trajectory.