news

한어Русский языкEnglishFrançaisIndonesianSanskrit日本語DeutschPortuguêsΕλληνικάespañolItalianoSuomalainenLatina

Author: Chang Minxiao

Editor: Qiu Xiaofen

From the Hundred Models War in the past two years to the gradual implementation of the Large Language Model (LLM), edge AI has always been a crucial part of the development of artificial intelligence technology.

The so-called end-side AI means that users do not rely on cloud servers during use, but use AI services directly on the terminal device.

Compared with mainstream large language models such as ChatGPT4.0 and the latest Llama3.1 that rely on cloud interfaces, edge AI has stronger privacy and many advantages such as personalized operations and cost savings.

Although several companies represented by Microsoft have proposed "AI PC" products in the market, which successfully shrink the large language models in the cloud and enable them to be used offline on mobile terminals such as mobile phones and computers, ensuring that automated programs can be used anytime and anywhere on low-cost smart devices is still a major challenge for small models.

An American startup company wants to solve the problem of end-side implementation.

"Aizip" was founded in Silicon Valley, USA in 2020, focusing on the edge AI applications of micro artificial intelligence (Tiny AI) on intelligent hardware such as perception end and edge end.

Currently, Aizip has created a series of small language models (SLMs) and AI Agents named Gizmo, ranging in size from 300 million to 2 billion parameters.

This model supports use on various platforms, allowing better compatibility between different devices and allowing users to experience consistent operations and functions when switching between different devices.

Aizip mainly provides a series of services such as intelligent customer service, technical support, financial support, and human support to enterprise customers. Individual users can use Gizmo's personalized voice assistant service on smart devices such as mobile phones, tablets, watches, cars, and home appliances.

It is understood that Aizip hopes to enable users to use AI applications anywhere by building an AI model that can be deployed on mobile devices at any time and has strong scalability. The company has currently made breakthroughs in key technologies such as data-centric model efficiency and AI design automation.

Compared with large models, small models can be deployed on chips costing less than ten U.S. dollars. Their application and promotion costs are much lower, and they can be widely used in the mobile terminals of smart devices.

This means that AI can provide users with faster and more accurate suggestions. Users can also use AI technology to handle daily work and entertainment more conveniently on mobile phones and tablets, and use AI's personalized services offline.

In terms of commercialization, it is understood that Aizip has recently cooperated with Renesas Technology to launch an ultra-efficient small language model SLM (Small Language Model) and compact AI Agents based on ARM microprocessor units (MPUs). Renesas Technology provides a wide range of microcontrollers, system chips (SoCs), power management ICs and other products.

In terms of the team, the "Aizip" team members are all from well-known universities such as Stanford, Berkeley, MIT, Carnegie Mellon, and many members have worked for Google and Meta.

Weier Wan, one of the founding members of Aizip and head of Aizip SLM and AI Agents development, said: "We are witnessing a critical moment in efficient AI. Edge-side AI Agents can truly make everything think. On the RZ/G2L with a single A55 core running at 1.2 GHz, the response time of the model can be reduced to less than 3 seconds."

end

end