What is the next step for AI? Qualcomm is optimistic about hybrid AI and multi-mode AI becoming a new trend

The wave of artificial intelligence (AI) is spreading around the world, and the edge AI (Edge AI) is gradually becoming an important force in promoting industry changes. Paul Torres, the director of Qualcomm Product Management, pointed out in the first "Qualcomm Taiwan AI Hackers" that in the future, "Hybrid AI" and "Multimodal AI" will become the two major trends of AI in the future.

"In fact, 'edge AI' and 'cloud AI' are not the choice of two choices. Both have their own advantages and can also be used as mutually complementary effects." Torres believes that future development goals may move towards "hybrid AI", that is, the AI ​​models of cloud and device-side cooperation to achieve their own advantages. In addition, with the diversification of application situations and devices, "multi-mode AI" links various devices according to different situations, and integrates them, and provides more real-time and personalized intelligent experience, which will also be a major trend.

As for why edge AI plays such an important role, Torres also points out several key advantages. First, there is "low latency". Since edge AI runs on the device side, it is suitable for applications that require low latency and real-time reaction; second, the Internet "reliability" and not all scenes can be connected at any time, so the device must be considered to be able to operate independently even when off-line; the third point is "privacy", such as financial data or personal sensitive information, and many users prefer to keep this data in their own devices rather than uploading them to the cloud.

The fourth point is "personalization". When personal data is left in your own equipment, the edge AI can adjust services based on the user's connection and situation. Torres said that Qualcomm Sensing Hub is an example. It can integrate multiple sensors, understand the user's location, habitual applications and behavioral patterns, and then establish a dedicated AI experience; finally, it is processing "efficiency". If some tasks can be completed directly on the device side, it will be more efficient than uploading the cloud side. Although large language models are still mostly running on the cloud, due to their high costs, the "Local Model" (local model) that can be executed on the device side has gradually been output, which balances performance and cost.

Neighboring AI needs to jointly promote the ecological system, and Qualcomm's hardware and technology dual support

Due to the rapid advancement of AI, new cloud models are often found, but at the same time, many companies are developing small models that can run on the device side. This also means that more hybrid solutions will emerge, while combining the advantages of the cloud and device side.

Currently, multiple bank applications are often required to be installed on each person's mobile phone, and not only one bank account. To check the information of different accounts, these applications must be connected to the cloud to obtain data. But in the future, Torres believes that it can provide assistance through "AI agents", directly grab and integrate data from the cloud, and then retain the integrated information on the user's own mobile phone, which is also a comprehensive solution for hybrid AI to generate.

As for the "multimodal AI" in "multimodal AI", it means that everyone has many devices at the same time, such as mobile phones, computers, cars or IoT network devices, which can collect personal information. Multi-mode AI is to understand user needs based on the situation through AI, and then integrate various devices and data to provide smarter and more personalized services.

However, to develop edge AI, the entire ecological system still needs to be jointly promoted. Torres said that in addition to the device having sufficient hardware capabilities, developers also play a key role. Therefore, Qualcomm continues to provide innovative technologies to provide hardware with stronger computing capabilities, hoping to provide developers with service as a starting point for promotion, and this is also the original intention of "Qualcomm AI Hackers".

Qualcomm AI Hackers are held at multiple locations around the world to stimulate the spirit and innovation of developers

Qualcomm has held an AI hackathon event worldwide, bringing the best CPU and powerful NPU through its Snapdragon X series platform, and paired with the diverse resources of Qualcomm AI Hub to provide developers with practical development tools and platforms. Torres said that because Qualcomm has relevant technologies, he hopes to see actual technical applications and innovations in application areas, and promote this event even more.

Qualcomm AI Hackers has been held in India, Canada, the United States, South Korea and other places. Torres laughed at the most interesting thing about this event is that it can see diverse applications. For example, Indian developers proposed to apply AI to VR headset devices, so that users can instantly listen to translations of different languages ​​when playing croquet because India has multiple dialects and uses different voices such as celebrities; Korean developers have appeared in cases of using device-side AI to process emails, using Qualcomm NPUs to summarize, classify emails, and even integrate chat robots to automatically reply emails; and this time, Taiwan’s champion team uses AI It helps users quickly classify numb files, greatly shortening the time for users to find the correct files, which is quite creative.

Torres said that the core spirit of Qualcomm AI Hackathon is to allow developers to try to solve problems encountered in the real world. Although the 24-hour challenge time is limited and many works still require continuous corrections and optimization, Qualcomm will continue to encourage the team to continue to develop and choose the most potential and commercial solutions to provide more support.

Finally, Torres also emphasized that AI Hackers are not just a competition, but more like a platform for communication and initiation.. Participants can stimulate new spirits through watching other teams’ solutions, or extend new applications from existing works. For example, a prototype of a chat robot can be transformed and extended to other potential areas in addition to its established functions. This mutual development process not only helps the team break through the thinking framework, but also accelerates the emergence of diversified application of AI on the device side.

Extended reading: Qualcomm AI hacker has come to an end, Qualcomm Liu Sitai: Snapdragon X platform helps Taiwan become a global AI innovation AI server promotes PCB material and equipment upgrades! The legal person named six major Taiwanese factories benefited from