Google DeepMind is a leading research division of Alphabet Inc. Legally, it operates as DeepMind Technologies Limited, incorporated on September 23, 2010, in London, United Kingdom. In 2014, the company was acquired by Google, and in April 2023, it strategically merged with the Google Brain team to form a unified division under the Google DeepMind brand. This move combined fundamental research with Google's large-scale engineering infrastructure.
DeepMind's history is a series of breakthroughs that have transformed science and technology. Founded as a startup with the ambitious goal to "solve intelligence," DeepMind quickly made headlines by creating AlphaGo, which defeated world Go champion Lee Sedol in 2016. This demonstrated that AI could surpass humans in games with vast state spaces requiring both algorithmic skills and intuition. However, the truly pivotal moment came in 2017 when Google researchers published the paper "Attention Is All You Need," introducing the Transformer architecture, which became the foundation for all modern large language models.
In the realm of modern large language models, the company is known for several brands. Primarily, there's Gemini—a family of proprietary multimodal models serving as Google's flagship product. Next are proprietary model lineups for generating video (Veo 3), images (Imagen 4), voice (Chirp 3), and music (Lyria 2), distinguished by high quality, world physics understanding, and the ability to accurately interpret complex prompts. Finally, **Gemma** is a family of lightweight, open models built on the same core technologies as Gemini. The third version of Gemma already offers native multimodal capabilities and uses an innovative hybrid attention architecture (5:1), which drastically reduces memory consumption when processing long context, enabling powerful models to run on consumer hardware.
Google DeepMind plays a significant role in the open-source community, democratizing access to cutting-edge AI technologies. Its primary contribution is the Gemma family. Models are distributed under an open license permitting free use, modification, and commercial application. They are positioned as easily fine-tunable and aimed at those who want to experiment with AI or integrate its capabilities into their applications using readily accessible hardware. By developing the Gemma lineup, Google DeepMind also supports the **Gemmavers** community, which actively promotes and assists developers and enthusiasts in implementing research and product solutions based on Gemma.
Google DeepMind represents a unique synthesis of fundamental science, a successful commercial venture, and a driver of the open-source community. These qualities, combined with the popularity of its models, position the company as one of the most influential players in the modern artificial intelligence industry.