AWS Amazon’s cloud computing division, presented this Tuesday a new family of multimodal generative artificial intelligence language models, Amazon Nova. He did this in the context of re:Invent the event held every year by the company from Las Vegas, which is one of the largest in the technology sector and has a strong focus on the development of systems, applications and product launches in the IT community. cloud computing.
Generative artificial intelligence gained a lot of ground during 2024 and, with OpenAI (ChatGPT) In the lead, all the technology companies came out with their own products to compete in the market. Amazon Bedrock is the AI platform for developers, which AWS launched in April last year.
In the main talk (keynote), Matt Garman, CEO of AWS, presented news in the area of computing and data processing, and then introduced Andy Jassy, CEO of Amazon and successor to Jeff Bezos in the position, after The founder of the company will leave the leadership in 2021. Jassy was in charge of presenting the main course: four new generative AI models: Micro, Lite, Pro y Premier.
We have been working on our frontier models, which have made great progress in the last four or five months. And we think that if we take advantage of them, probably you too,” Jassy said during the presentation.
What is each text model like?
AWS enters strongly to compete in the GenIA market but with the focus on development: where the end user enters ChatGPT to produce a text or Midjourney to create an image, the AWS client can use new implementations within Bedrock , your AI.
“Whether developing document processing applications that need to process images and text, creating marketing content at scale, or creating AI assistants that can understand visual information, Amazon Nova provides the intelligence and flexibility you need with two categories of models. : understanding and generation of creative content,” the company explained in an official statement.
The new Nova text generation models, optimized for 15 languages (with main focus on English), They have sizes and capacities that vary depending on use, Jassy explained. Micro It can only receive and send text, but it offers the lowest latency of all (i.e. it’s fastest).
Lite It is “very low cost,” says AWS, and can process images, video and text, but offers a balanced combination of accuracy, speed and cost for various tasks.
Pro y Premierlike Lite, can analyze text, images and video but with more capabilities, with Premier being the most developed and expected for early 2025.
Jassy assured that these models are “the fastest in its class”but at the same time they are the cheapest to run. “We have optimized these models to work with our own systems and APIs, so that multiple orchestrated automatic steps can be performed much more easily with these models,” added Jassy.
Two image and video models were also presented: Canvas and Reel, the first for AI-generated images and the second for videos of about six seconds.
In this sense, Jassy anticipated a very strong discussion on artificial intelligence, which is both the responsibility in the generation of content, as well as the transparency of a text or image being generated with AI.
“We are trying to limit the generation of harmful content,” he assured, which is why the same limitations apply as in Bedrock to combat “misinformation, child sexual abuse material and chemical, biological or nuclear risks.
In various re:Invent sessions with journalists, company representatives assured that user information is not used to train their AI models, although specialists in the field of personal data protection assure that this is not entirely clear.
Finally, within the advances of AI models, Jassy said that the company is working on a “speech-to-speech” model, which would take speeches and produce a transformed version, including interpretation of tone and cadence, beyond semantics. , to generate “very natural voices” (this towards mid-2025).
“You can enter text, voice, images or video and emit text, voice, images or video, this is the future of how frontier models will be built and consumed,” concluded the CEO of Amazon.
Was the first time Jassy spoke as CEO of Amazonsince in past editions he was the main speaker due to his role as director of AWS.
Advertisements in Latin America
In addition to the main session, re:Invent holds sessions with specialists, analysts and journalists. In one of them, carried out with the Latin American press, they gave details about the best-known AWS use cases in the region, such as the case of Boti in the City of Buenos Aires, which runs on AWS, or that of the data from the Municipality of Córdoba.
In a session with journalists from South America, Andrés Tahta, General Director of the Public Sector for Latin America at AWS, recounted cases in which technology is being used to analyze mosquitoes and identify if they may be carriers of dengue in the municipality of Vicente. López, along with another public system in Tierra del Fuego to prevent fires.
They also threw out numbers. One of the data they revealed was that the cloud “generated $53 billion of Latin America’s GDP in 2023,” according to official company data. “Looking at the potential and innovation opportunities we have, this figure could exceed USD 700,000 million in the next five years“, commented Paula Bellizia, vice president for Latin America at AWS, in her first participation in AWS re:Invent 2024, during the morning of Monday.
In this sense, Brazil and Mexico continue to be the main focuses: they will invest more than 6,000 million USD in Latin America in the coming years, 5,000 million in the creation of a new AWS Region (set of data centers) in Mexico and 1.8 billion for the expansion and modernization of infrastructure in Brazil.
“The information technology industry is expected to grow by 11% in Latin America in 2025. This is 5 times more than the increase in the region’s total GDP. Cloud computing services are expected to expand by 21%, while the use of artificial intelligence will be 49% higher,” they added in relation to Latin America.
According to Juan-Carlos Gutierrez, director of Technology and Solutions Architecture at AWS for Latin America, generative AI “is reaching a new point of maturity: in 2023, companies carried out proofs of concept to see if they could use it, already this year, Companies of all sectors and sizes began to use it effectively in their processes. In 2025, the challenge will be scale this use.”
About re:Invent
re:Invent has been taking place for 13 years and is the company’s largest event. This year they attended about 60 thousand people in person, at the Venetian hotel, and almost half a million people online, according to company data.
During the main talk, Matt Garman, CEO of the company, assured that the company continues with the customer focussomething that Jeff Bezos always emphasized within Amazon’s culture (the well-known “customer obsession”).
“At AWS we think about build blockswhen a service does something very well, that can be combined with another service to build even more interesting things: that has been fundamental during the 18 years of AWS,” he said at the opening of his keynote.
During the days of the convention there were sessions and talks focusing on the company’s different products, but above all on the pillar of cybersecurity: “Safety always has to be our priority, it is part of our culture and has been since the beginning,” said Garman.
The sessions last all week, until Friday.