2024-08-05 07:33:37
In an opinion column shared by Sundar Pichai, CEO of Google and Alphabet, revealed to the world the arrival of Gemini 1.5Google’s latest innovation in the field of artificial intelligence.
This new generation of AI models, according to Pichai, “r“It represents a radical change in our approach” by combining advances in both research and engineering behind the development of foundation models. Sundar Pichai notes: “We have been able to significantly increase the amount of data our models can process, managing to execute up to a million tokens consistently.”
This breakthrough not only symbolizes technical progress, but also marks a milestone in the ability of AI models to understand long-form contexts, which, according to Google’s CEO, will open up entirely new chapters in terms of possibilities for developers and business customers.
Gemini’s expanded context window promises a revolution in AI applications (Google)
The adoption of an architecture of Mix of Experts (MoE) is central to this development, allowing Gemini 1.5 models to be more efficient in both training and operation.
This model divided into “minor expert neural networks” significantly improve efficiency by activating only the most relevant expert pathways based on the type of input they receive.
The model Gemini 1.5 Proas it is known, is a mid-size multi-modal version that has been optimized to scale across a wide range of tasks while maintaining comparable performance to Gemini 1.0 Ultrathe largest model to date.
This release also incorporates a revolutionary experimental feature in large context understanding, capable of processing vast amounts of information in a single action—including up to one hour of video, eleven hours of audio, codebases with more than 30,000 lines of code, or more than 700,000 words.
Sundar Pichai highlights the importance of this ability by offering concrete examples, such as analyzing and summarizing 402-page transcripts from the Apollo 11 mission to the Moon, or identifying details in 44-minute silent films, demonstrating not only deep understanding but also the ability to reason advancedly about various types of content.
The improved effectiveness of Gemini 1.5 Pro This is also reflected in its superior performance over its predecessors in 87% of the evaluation tests. Importantly, this model maintains high levels of performance even when increasing the size of its context window, finding specific texts embedded in blocks of up to one million tokens with 99% accuracy.
This “learning in context” of Gemini 1.5 Proaccording to Pichaisuggests a remarkable ability of AI to acquire new skills from information provided in a long prompt, without the need for additional adjustments.
On safety and ethics testing, Sundar Pichai says Gemini 1.5 and its models have undergone rigorous evaluations to ensure their alignment with Google’s AI principles and security policies. It expresses the company’s commitment to the responsible deployment of its advanced AI models, promising to extend these innovations to billions of people, developers and companies in a responsible manner.
In this context of constant innovation and development, Google is at the forefront, not only by offering technological advances but also by setting standards in the ethical and safe implementation of artificial intelligence. Sundar PichaiThrough its opinion column, Google not only shares these advances with the world but also reaffirms Google’s commitment to a future where technology serves collective well-being with responsibility and care.
By: Infobae