Transformer

What is the Transformer?

The Transformer is a neural network architecture Google researchers introduced in a 2017 paper. It landed like a bomb in the pool — shattering the old ways of processing text and becoming the shared foundation for nearly every large language model today. GPT, Claude, Gemini: underneath, they're all Transformers.

What problem did it solve?

Recurrent networks were too slow
Older models read a sentence one word at a time, waiting for each word before the next. No parallelism, and long sentences got slow or forgetful.
The Transformer sees the whole sentence at once
Using attention, every word can look directly at every other word in the sentence, all in one parallel pass — fast and good at long-range memory.

What does it look like under the hood?

Encoder and decoder
The classic design has two halves: the encoder "reads" the input, the decoder "writes" the output. Translation is read-one-sentence, write-one-sentence.
Multi-head attention
Not one way of looking, but several attention heads firing at once, each catching different relationships between words.
Positional encoding
Because everything runs in parallel, the model doesn't know word order on its own — so you tell it who's first and who's last.

Why does it rule AI?

It scales beautifully. Add data, compute and model size, and capability keeps climbing with no obvious ceiling. That's the stage for the scaling laws, and the starting gun of the LLM era.

Bottom line: the Transformer reads whole sentences in parallel via attention, and it's the shared foundation of modern large language models.