The first steps of implementing artificial intelligence on mobile processors
Financial Times: How will generative artificial intelligence affect Qualcomm’s customers, who are mainly large companies active in the field of electronic and digital products?
Big companies are talking about improving the working experience of their products
Cristiano Amon: We are still at the beginning of this long and winding road. These days, big companies are talking about “changing the work experience of their products”. Some of them want to use artificial intelligence to improve the photography experience, and others prefer to take advantage of the limitless power of artificial intelligence to increase efficiency, perform translation or improve communication.
The potential of mobile artificial intelligence in 2024
Financial Times: To what extent can artificial intelligence guarantee the growth of the mobile market? Will the consumer pay more money to access the new features and facilities of artificial intelligence, or will he consider these features as a necessity of the smartphone and will not pay extra money to have it?
We want you to come to the conclusion that your current mobile is outdated
Cristiano Amon: In early 2024, with the unveiling of mobile phones equipped with productive artificial intelligence, we will witness the beginning of a revolution in the mobile industry. All we want is for the consumer to come to the conclusion that his current mobile phone is outdated and that he “needs” to get a new and advanced smartphone in order to benefit from the new features and capabilities of artificial intelligence.
Move from cloud artificial intelligence to mobile artificial intelligence
Financial Times: The mobile industry has faced many challenges in the past few years and now it needs something new to revive and increase sales.
Cristiano Amon: We were very pleased with the revenue of the mobile industry in 2023, but the industry is mature in many ways and the growth will be partial. With the presence of productive artificial intelligence in mobile, we can be optimistic about the growth of this industry in 2024. I think AI can have a great potential in creating a cycle of replacing old phones with new ones.
Hybrid models of artificial intelligence
Financial Times: Chip makers can share in the benefits of large models of productive artificial intelligence in two ways; Training and inference. “Training” involves processing a large amount of information to develop a model, and “inference” refers to how these models are used to create new products based on artificial intelligence. Currently, chip makers such as Nvidia are focused on education. Which approach do you think is more profitable for chipmakers?
Cristiano Amon: Many of the necessary infrastructures for information processing and the development of large artificial intelligence models have been provided by cloud services using ultra-advanced processors. We all know that the current models are huge and all-rounders, but as the model gets smaller, the tasks are limited. So I think the “inference” will be a profitable market.
Reducing the cost of artificial intelligence calculations and processing
Financial Times: Currently, the use of tools such as ChatGPT imposes a high cost on the development companies. How can the implementation of artificial intelligence models on mobile processors minimize these costs?
Artificial intelligence processing on mobile can reduce the cost of information processing
Cristiano Amon: Today, many large artificial intelligence models run on cloud services, while by transferring them to mobile processors, we can significantly increase the accuracy of these models while reducing costs. Considering the cost and energy required for information processing and development of artificial intelligence models, the high importance of using hybrid models is felt more and more.
Currently, it is very expensive to send commands to large language models and receive responses from them, but with the rise of special AI processors in mobile phones, AI models such as GPT are installed on the neural processing unit, and by sending commands to tools such as ChatGPT, First, the model installed on the mobile is executed, and then the output of this model is sent to cloud services for final processing. In this situation, a large part of information processing takes place on the mobile phone, and in many cases, there is no need for cloud processing.
There is no need for a new phone and operating system
Financial Times: The implementation of generative artificial intelligence models on the mobile phone will ultimately benefit artificial intelligence companies and cloud service providers; But these days, big model language developers like OpenAI, Anthropic, Microsoft, Amazon, and Meta don’t necessarily design mobile phones or proprietary operating systems, and big companies like Apple, Google, and Samsung are the main players in the mobile industry. How are you going to align these companies and incorporate smaller models into their products?
Gadgets with NPU process part of the information on the mobile and the other part in the cloud servers
Cristiano Amon: Any device that has a neural processing unit can process part of the information on the mobile and the other part in the cloud platform. Now suppose you want to use Meta’s Llama 2 model. In this case, you will not need a mobile phone and a dedicated operating system, and if your phone has an NPU, you can benefit from the benefits of this company’s artificial intelligence model.
Neural processing unit, the most important element in new mobile phones
Financial Times: You believe that companies like Meta and Microsoft can regain their lost influence in the mobile industry by offering AI models on mobile phones. Don’t you think that in order to realize this great goal, it is necessary for mobile phone manufacturers, operating system developers and artificial intelligence model designers to unite with each other? In most cases, the relationship between these companies is not very friendly.
NPU will become one of the key elements of future mobiles, like the phone’s camera
Cristiano Amon: If this idea comes true, AI models will be installed on the NPU and you will no longer need to download a separate application from the software store to take advantage of it. Perhaps it is better to consider the NPU as a mobile camera; The manufacturers of phone camera modules do not think about whether this module wants to be installed on an Android mobile or on an iPhone. They make the camera part and provide it to companies active in the mobile industry.
Advantages of mobile artificial intelligence processing for Qualcomm
Financial Times: How does Qualcomm benefit from this market?
Cristiano Amon: Our first and biggest benefit is expanding the smartphone market. However, to enjoy the benefits of productive AI, we need more powerful and customized processors. For this reason, with the expansion of the mobile market, the sale of our mobile processors will also increase, and the research and development process of new and powerful chips will go faster.
With the introduction of productive artificial intelligence, the desktop computer market will also expand significantly. We are entering the computer world and Microsoft has chosen Qualcomm’s processors to compete with Apple’s proprietary processors. We have been able to produce processors that provide very high processing power while consuming low energy and are perfectly suitable for processing artificial intelligence information.
Qualcomm’s benefits from the advancement of artificial intelligence: the expansion of the mobile market, Desktop computers and automotive industry
Another industry in which we have a good position is the electric vehicle industry. The quality of the cabin and digital console made by us is so high that today more than 30 car companies use these technologies in their electric cars. By implementing generative artificial intelligence in cars, not only will the quality and efficiency of our current systems be enhanced, but our position will be firmly established in the automotive industry.
Generative artificial intelligence in the world of virtual reality
Financial Times: In the not-so-distant past, tech people were talking about virtual reality and Metaverse, and it went so far that Facebook changed its name to Meta; But the result of many investments in this area was not as effective as people thought. These days, experts and people are talking about generative artificial intelligence and amazing products have been developed based on this technology. What do you think about the combination of these two technologies?
Artificial intelligence processing on mobile can revolutionize the world of virtual reality
Cristiano Amon: Right now, virtual reality glasses and headsets are very popular, and in my opinion, with the introduction of productive artificial intelligence into these products and the possibility of interacting with large language models through voice commands, these types of headsets will achieve exciting sales in the near future.
The future of wearable gadgets
Financial Times: The glasses made by Meta and RayBan use Qualcomm’s proprietary chips designed specifically for lightweight wearable gadgets. Of course, rather than relying on immersive images, these glasses prioritize “artificial intelligence assistants” and “voice command processing”. Are these glasses going to be a disaster like Google Glass or will they become popular products in the short term?
The wearable gadgets of the future must be lightweight and have exclusive content
Cristiano Amon: To learn these types of gadgets, two big and challenging tasks must be done; Weight reduction and easy use and “proprietary content production. By solving the first problem, the second problem will be solved. In fact, with the increase in sales of wearable gadgets and the acceptance of users, developers will have more motivation to produce exclusive content for these types of products.
