The connectionist breakthrough
For decades, artificial intelligence was divided between symbolic logic and neural networks. Pioneers like Geoffrey Hinton proved that multi-layer neural networks trained with backpropagation could automatically discover hierarchical representations from raw data.
The transformer and compute scaling
The self-attention mechanism allowed neural models to process sequences in parallel, creating Large Language Models and Generative AI. This compute revolution was enabled almost entirely by NVIDIA GPUs and massive Data Centers.
