Perplexity AI

TL;DR

The ideal AI system goal is to deliver the most token value per watt, balancing accuracy, privacy, and cost/energy.
Perplexity's new hybrid local-server inference orchestrator routes AI tasks automatically between a user's device and the cloud.
This system is designed for sensitive data like financial or health information, running compact models locally for privacy.
Tasks requiring frontier models run on the server, with the system splitting and coordinating mixed tasks.
Advancements in local hardware will allow more inference to run on-device, reducing reliance on centralized infrastructure.
This hybrid AI approach changes the math for compute shortages and enhances data sovereignty.
Perplexity's business model focuses on accurate AI, aligning incentives for optimizing value per watt.
The Personal Computer with local inference, launching in July, is presented as the first product to make hybrid AI a reality.

Continuar leyendo el artículo original