Deepmind, the firm of Artificial Intelligence (AI) that Google acquired in 2014 and officially integrated into a single division in April 2023it has been for years one of the great referents of the sector. Over time, he has earned that place for the quality of his scientific publications and for gathering some of the brightest researchers. But there are reasons to believe that this approach has begun to change.
Several experts cited by Financial Times They ensure that Google Deepmind is delaying the publication of certain advances that it considers “strategic” or delicate in the field of generative AI. The measure is part of a strategy designed to preserve its competitive advantage and prevent its most recent and valuable developments from being exploited by rivals such as OpenAi.
Transformers, star architecture
Much of the development of generative artificial intelligence is not understood without Google’s advances. One of the most influential milestones was the publication in 2017 of the article ‘Attention is all you need‘, signed by eight researchers, who introduced the Transformers architecture. This approach, in general, allowed the models to process data more efficiently within large volumes of information.
This architecture became the base of models such as Bert (bidirectional encoder representations from transformers), developed by Google itself and incorporated into its search engine in 2019 to improve the understanding of natural language. It was also key to the development of presentful systems such as GPT (Generative pre-trained transformers) of OpenAi, where the current ones are framed GPT-4 and GPT-4.5.
Google is one of the world’s largest companies in the world. It has a huge financial muscle and access to key technology. Even so, the launch of Chatgptbased on GPT-3.5a model built on Transformers architecture, He took by surprise to the teams directed by L sundaria Pichai. The reaction was a “red code” and an urgent reorganization to compete again in an AI race that today leads OpenAi.


It is no secret: when a company becomes a Big Tech, it loses part of the dynamism that defined it in its beginnings as startup. As we counted exactly one year agothey end up transforming into giants where the “move quickly and breaks things” no longer fits. They have too much to protect and a gear with thousands of pieces that cannot allow failures. Risking is not as easy as it seems.
Even in this context, it surprises the speed with which the company of the search engine has managed to catch up. In a short time he has launched a AVALANCHA OF IA PRODUCTS based on advanced language models. There it is Geminihis direct rival of Chatgpt; Gemini Livedesigned to compete with Openai’s advanced voice mode; the Gemsthat work as Personalized GPTS; and amazing tools, such as Notebooklm.
The new Google pulse
The last years have forced the Mountain View company to introduce internal weight changes. One of the most significant affects the policy of publication of scientific articles: if the content is considered strategic, an seizure of six months is imposed before making it public. In addition, the group led by The Nobel Prize winner Sir Demis Hassabis He has hardened his internal processes, with a stricter review.
One of the researchers who spoke on condition of anonymity with Financial Times was clear: today he cannot imagine Google by publishing a document like Transformers for general use. In this new dynamic, they say, “the company has become more concerned about the product and less for sharing the results of the investigation for the benefit of the common good ”, an address that can generate some discomfort in the community.
Images | Boliviainteligent | Google Deepmind
GIPHY App Key not set. Please check settings