Transformer
A neural-network architecture from 2017 that drives every modern LLM. Built on the attention mechanism.
Before transformers, recurrent networks (RNN/LSTM) were the standard for sequences. Transformers replaced them by processing the entire sequence at once instead of one token at a time — enabling much larger models and faster training.
"Attention Is All You Need" (Vaswani et al., 2017) is the paper. Everything since — GPT, Claude, Gemini, Llama — is a variation on that theme.