Artificial intelligence · 2022
PaLM
Google's 540-billion-parameter model explains jokes and reasons step by step.
PaLM was trained on 6,144 TPU v4 chips with Google's Pathways system. It showed breakthrough gains on reasoning benchmarks with chain-of-thought, and its successor PaLM 2 powered Bard from May 2023.
What it built on 3
What it led to 1
Related measurements
The dashed line marks when it first worked.
Training compute of notable AI models Source: Epoch AI, Notable AI Models (CC BY 4.0)
ML hardware FLOP/s per dollar Source: Epoch AI, ML hardware (CC BY 4.0)