Conference Presentation, Fireside Chat
A fireside Chat with Ramin Hasani, Co-Founder & CEO of Liquid AI ft. Robert Cahill (Enhance Venture)
- Liquid AI plans to establish a European office soon and deploy foundation models directly on diverse devices, including robots, phones, laptops, satellites, cars, and Raspberry Pis.
- The company intends to scale "liquid neural network" technology, developed over the last 10 years, to create alternative models to transformers capable of processing text, vision, and audio for predictive financial and medical time series analysis.
- A research team comprising scientists from MIT, Stanford, and Yoshua Bengio's lab is being assembled to address the scaling of this non-transformer architecture.
- Models are designed to exhibit linear scaling relative to sequence lengths, reducing compute requirements while balancing model quality with memory constraints on hardware-limited devices.
- Strategic plans include partnerships with Fortune 500 companies to deploy mobile, laptop, car, and satellite intelligence in consumer electronics and home appliances.
- The company anticipates a trillion-dollar market for on-device AI by 2035, accompanied by the emergence of advanced wearable hardware such as extremely smart glasses and pins.
- Liquid AI expects systems superior to human performance to emerge within the next couple of months, necessitating the development of safety protocols.
- The organization aims to function as an interface or "gate" for safe, direct interaction between users and large general AI models ("oracles") while managing daily tasks.
- A future ecosystem is predicted involving the coexistence of multiple advanced oracles and intelligent devices carried by users.