Been saying this since 2020: AI is going to follow the same pendulum that beats in the heart of IT since its inception. Some new tech starts out centralized, then slowly shifts to a distributed model. Examples:

• Mainframes to desktops.
• Unix servers with (X11) thin clients to desktops.
• Server-rendered web to client rendered (e.g. SPAs).
• PBX to soft phones and on-device audio DSP.

AI is moving even faster. Examples:

• Apple and Google have already moved inference to their phones to save central compute resources (hybrid approach).
• Companies are now moving from cloud AI providers to locally hosted options.

Before you know it, companies and consumers will have the necessary hardware to handle all their AI needs with edge hardware (e.g. right on their phones and desktops—no need to call out to any servers).

This week’s news had researchers demonstrating that ChatGPT Astra could natively modify and improve a tiny Qwen model. How long before we have models that have had all their internal knowledge (fact) checked and embedded into chips via Cerebras Systems method?

We could have the equivalent of PhD-level knowledge on zillions of topics built into our PCs and phones and these hardware models wouldn’t need to call out to the cloud at all except to receive patches/updates that get applied whenever they run.

They’d get refined and updated over time as flaws are found. People would regularly update these chips in their PCs or buy new phones so they’d have the latest knowledge and capabilities without having to wait for the software patches to do their thing.

Their not british.