Infrastructure
Compute
The raw processing power used to train or run a model, a key driver of cost and capability.
Compute is the shorthand for the amount of calculation a model requires, usually supplied by large fleets of GPUs or TPUs. Training a frontier model can consume a staggering amount of it, which is why compute budgets are one of the main things separating what different labs can build. The broad trend has been that more compute, paired with more data, yields more capable models, though not without limits.
Related terms
Graphics Processing Unit
The specialized chip that does the heavy parallel math behind training and running AI models.
Tensor Processing Unit
A chip designed by Google specifically for machine learning, as an alternative to general-purpose GPUs.
Pretraining
The first, largest training phase where a model learns general patterns from a huge amount of broad data.
Frontier Model
One of the largest and most capable AI models at the leading edge of what is currently possible.