LLM-Style Scaling Laws Apply to Sensor Data as Well
Scaling laws for LLMs also apply to wearable models, presenting new startup opportunities.
The scaling laws that govern loss in LLMs—dependent on model size, dataset size, and compute—are now shown to apply to non-language models like wearable foundation models. Google's research established a scaling law for physiological sensor data, indicating that larger models and more data improve performance across various generative tasks. This finding opens up new opportunities for startups in the wearable tech space, as they may require significantly less investment compared to LLM ventures.