Last week, OpenAI showcased the latest iterations of Sora, GPT-4o Vision, and Voice Engine at Vivatech 2024. The demonstration was a vivid display of AI’s growing capabilities:
Sora generated a film from a text prompt illustrating the seamless integration of AI in creative processes.
GPT-4o Vision analyzed the video, offering an insightful description and narrative, showcasing how AI can enhance content comprehension.
Voice Engine adapted to the presenter’s voice in seconds, narrating the story in French and demonstrating real-time language adaptation.
While advancements in Large Language Models (LLMs) concerning vision and graphics are remarkable, there is a notable shift towards embracing Small Language Models (SLMs).
The Rise of Small Language Models
Microsoft’s recent introduction of its 'Phi3' models marks a significant advancement in AI technology. These Small Language Models (SLMs) are designed to function efficiently on smaller, less powerful devices like smartphones and wearables, offering smart functionality even in non-connected environments.
Key Advantages of SLMs
On-device operation: Optimized to perform complex computations locally, reducing the need for constant cloud access.
Non-connected functionality: Ideal for use in remote or unstable internet conditions, ensuring reliability and continuous operation.
Enhanced data privacy: Local data processing minimizes the risk of breaches, protecting user privacy more effectively.
Extended Applications of SLMs
SLMs are not just transforming traditional tech applications but are paving the way for innovative uses across various industries:
Healthcare: Devices could perform on-the-spot diagnostics and patient monitoring without needing to send data out, ensuring patient confidentiality and immediate results.
Automotive: In vehicles, SLMs could process real-time data for features like autonomous driving aids and personalized in-car experiences.
Smart Homes: Enhancing home devices with AI enables them to anticipate needs and adapt to preferences without compromising privacy.
These examples illustrate a few industries that could be radically transformed through further AI development of on-device SLMs. Capabilities such as real-time object recognition, augmented reality overlays, and contextual assistance will one day enrich user interaction and engagement across platforms.