Google Research

What Google Research is

Google Research is the research division of Google, based in Mountain View, California; its founding year is generally given as 2000.^[1] Its work ranges from machine learning for language to large-scale learning infrastructure, and to applications of computing in other sciences.

From Google Brain to a deep-learning group

The deep-learning effort that became central to the division started in 2011 inside X, as a collaboration involving Jeff Dean, Greg Corrado and Andrew Ng.^[2] In 2013 Google hired Geoffrey Hinton and acquired DNNResearch, the company built around his neural-network research.^[2] A small internal team combined with hires from academia shaped how the group grew.

The Transformer

The division's best-known single result came from machine translation. "Attention Is All You Need" was submitted to arXiv on 12 June 2017 and published in June 2017, introducing the Transformer architecture that modern large language models use.^[6][1] The paper has eight authors: alongside Ashish Vaswani and Noam Shazeer are Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N. Gomez, Lukasz Kaiser and Illia Polosukhin.^[6]

A Google Research blog post of 31 August 2017 described the Transformer as a novel neural network architecture based on a self-attention mechanism, one that outperforms recurrent and convolutional models on translation benchmarks.^[8] Self-attention lets every position in a sequence weigh every other position directly, instead of passing information along step by step as a recurrent network does. Long-range relations inside a sentence can therefore be captured in few layers, and the computation parallelises more easily during training.

BERT and T5

BERT was developed by Jacob Devlin, Ming-Wei Chang, Kenton Lee and Kristina Toutanova of Google AI Language, and its paper appeared on arXiv on 11 October 2018. The paper reports a GLUE score of 80.5%, an absolute improvement of 7.7 points.^[7] On 2 November 2018 Google Research open-sourced the code and the pre-trained models.^[3]

A second language result was T5, the Text-to-Text Transfer Transformer. T5 casts every language task as text in, text out, and was released together with C4, a web-crawl dataset.^[4]

Beyond language

Work listed for the division also includes the neural machine translation behind Google Translate, flood forecasting, large-scale learning infrastructure, and — through Google Accelerated Science — below-threshold quantum calculations reported in Nature in December 2024.^[1]

Reorganisation in 2023

On 20 April 2023 Google combined the Brain team with DeepMind to form Google DeepMind under Demis Hassabis, with Jeff Dean becoming Google's Chief Scientist.^[5]

Why it matters

The Transformer paper and the models that followed it gave the field a reusable recipe: an architecture built on attention, a way of pre-training on large text collections, and open weights that others could build on. BERT's open-sourced code and pre-trained models, and T5's text-to-text framing released with a public dataset, both lowered the cost of entering the field.^[3][4] The division's later history also shows how that work was absorbed into the company's main AI organisation, when the Brain team was merged into Google DeepMind in 2023.^[5]