Qualcomm has developed and integrated generative AI across its semiconductor line over the last few years. AI is a way for those completely off the grid to create new, original content, like images, videos, music, and movies, using intelligent algorithms.
Qualcomm’s generative AI strategies embrace this technology to improve various aspects of its products and services. The company claims that its technologies are capable of various unusual use cases. However, doing this locally on a smartphone is far more valuable, especially from the cost per query and scalability perspective.
Let’s talk about Qualcomm’s ability to create a hybrid AI feature that extends beyond the device and into the cloud.
Qualcomm started the conversation on this potential at Mobile World Congress earlier this year. This requires specific hardware modifications and substantial software changes, resulting in a large model called Stable Diffusion. It is a deep-learning text-to-image model that will be introduced in 2022.
Stable Diffusion’s main use is to generate detailed images from text descriptions. However, Stable Diffusion is also useful for tasks such as image repair (inpainting) and AI-generated images modified outside the original image boundaries (outpainting).
Parameters are the foundation for machine learning algorithms, which enables functional gen AI applications. The parameters are part of the model trained on past data. In general, the relationship between several parameters (and sophistication) has been surprisingly stable in the language domain. In the past, the approximate number of parameters needed for Gen-AI style apps was around 10 billion.
Stable Diffusion For On-Device Artificial Intelligence
Qualcomm claims that its implementation of Stable Diffusion only requires 1 billion parameters to be squeezed into the size of a phone. Stable Diffusion allows users to input a text question and generate a local picture without relying on the smartphone’s Internet capability.
Qualcomm’s demo was in airplane mode, so all the information necessary to create the image was stored within the device. Qualcomm has chosen Stable Diffusion as the model of choice because of its size and extensive training, which is derived from large amounts of data. It can understand concepts of a vast scope that are not limited to a small or specific set of topics.
Qualcomm is currently the only company that can enable this model to work on Android devices. The parameter models are getting smaller and more compact, making it possible to run compelling-gen AI apps on just one device. Along this path, similar generative AI applications can be used on all mobile devices.
Scalability is key for Qualcomm from a platform perspective. Few other companies have the same legacy of reaching devices across the entire end-user ecosystem. Qualcomm’s “established base” of Snapdragon devices now exceeds 2 billion, many without internet connectivity.
The Benefits of Qualcomm’s Generative AI Approach
Qualcomm’s history in the smartphone sector gives it distinct advantages, even if Nvidia is often the focus of the AI news.
Qualcomm’s generative AI can be used to create more immersive, realistic content that enhances the user experience. AR applications, for example, can create high-quality videos and photos, improving the user’s experience.
Qualcomm also offers businesses several advantages in product development and testing. Qualcomm’s generative AI can create realistic simulations for product testing and development. This may help speed up the process, reduce costs, and improve the effectiveness of product development.
Qualcomm’s OEMs could also benefit from the potential for personalization within the AI realm, while Qualcomm solutions can provide tailored experiences to consumers that utilize generative AI.
Qualcomm’s solutions can create specialized suggestions, unique user interfaces, or adaptive answers based on preferences and behavior patterns.
Qualcomm Must Tell Us More
As my readers know, I have been raising awareness about the ethical questions surrounding general artificial intelligence. The emergence of generative AI raises several moral issues, especially in the context of deepfakes. Qualcomm must ensure users of its generative AI act ethically and by the law.
There are many reasons to be concerned.
When I asked a company CEO that offers text-to-image gen-AI software if its terms and conditions required permanent watermarks and metatag fingerprints in the material produced, he responded negatively and dismissively.
A prominent CEO gushed at a recent tech conference about the possibility of “laborious” performance evaluations by gen-AI applications. Unimaginable is the number of lawsuits that will result if this happens.
On a recent call with analysts, Qualcomm seemed to realize that it had to play a leadership role ethically in this area and suggested it would reveal much more information about this topic at future conferences.
The company acknowledges wanting its customers to get the most out of gen-AI on their devices. It also stresses the importance of distinguishing between original content and that which gen-AI has altered.
Facial authentication could, for example, play a convincing role in minimizing this problem on this front. There are also biometric hardware functions that could be helpful.
New World, a brave one. But will we be safer
The company’s focus on gen AI and its work to integrate that capability into its extensive silicon portfolio has the potential to change the entire technological landscape. The time and productivity savings are real, significant, and almost incomprehensible.
Qualcomm can now run these apps robustly on mobile devices, including PCs and smartphones, even without an Internet connection. It is also very clear that the potential for privacy and information invasion can be a problem.
Qualcomm must follow strict privacy laws and protect user data to alleviate these concerns. It should also ensure that all personally identifiable information (PII) used in developing or deploying generative AI models is anonymized appropriately to prevent individual identification.
Qualcomm also must obtain users’ explicit consent before collecting or utilizing data for generative AI. To maintain user confidence, it is important to communicate openly about data sharing, storage, and usage procedures.