The world’s textWhat it learned from trained WebsitesBooksCodeScience papers Trillions of words, read once, during training. It does not search them when you ask.
GPU clusterThe chips that do the maths 8 online Maths per word140 billion Memory140 / 640 GB Speed40 words/s power 0.6 kW · 34 °C · fans 20% hardware8 GPUs