Agentic AI won't scale in silos.
We need shared protocols for tool use, comms, memory, and trust.
It's time to build the Internet of Agents. An interop stack for autonomous collaboration.
https://t.co/g9cG0k8Xbz
The interop problem is real. The work starts now.
The leap I witnessed in AI this week has me shaken.
If you've followed me, you know I don't really make statements like this.
I've been hearing how AI will change everything for years, but there's been little beyond promises. Nothing that can be used now.
That just changed:
I also noticed that GPT-4.5 looks very human. In fact, this empirical study found that the majority of users thought 4.5 was a human tool.
4.5 + AVM + robotics creates the humanoid human.
@karpathy Do you think this is a single-shot, multi-party dialogue model or blended in post somehow?
If the former, this would open new doors for how we ask models to pair program, conduct meetings, plan projects, brainstorm, etc.
I feel like i found a blood code for #NotebookLM !
Submit a creative brief for the episode you want with your sources and they follow your direction! Trying to make a tough decision, have the deep dive podcast debate from predefined positions. We live in the future.