I know talking is easy, doing is the hard part. :) So I'm currently also pursuing building autonomous AI agent system, where Agents work collaboratively towards a common goal.

Finally, the centrality of coordination for really defining achievements of humans is also highlighted by Anthropic. This is something I strongly believe in: Enabling large scale coordination is the real unlock of AI. My thoughts from Jan 2023 here:

This, from 2016, talks about the nature of collective intelligence vs individual human intelligence. The Anthropic team highlights this and uses it to make the argument for limits of generally intelligent agents.

Here, in 2017, I had noted my thoughts on how we could achieve collective intelligence by having domain focused AI agents.

Their key overarching insight:

Some interesting insights on multi-agent systems from @AnthropicAI , one of the leading AI Research Labs in the world: https://t.co/0F6z0PIRBp Their conclusions are exactly the same as some of my own insights from a decade ago, when AI was not mainstream.

There's no triumph in crushing the weak There's no bravery in massacring the meek...

Nations that fight internally will come out stunted intellectually. The mediocrity is insidiously pervasive when the state picks a side in the fight.

https://t.co/HSnVV5Bw7i This article by @sunoor in @thewire_in emphasises the the futility of tech parks and importance of guaranteeing a prosperous future for all, something that I also highlighted in my linked tweets from 6 years ago.

The dream may be abandoned but the domain must live on.

A dream come true... dreams that soothe the gloomy night, The soil of India nurtures us all, but the soul of India yearns for this nurturing light

"When locked away, the risks can grow, Unchecked, the dangers overflow. But open hands and eyes can steer, AI toward a future we hold dear." #OpenSourceAI

The anthropomorphic attributes of current AIs like GPT are only because they have been trained on data encapsulating human thoughts, motivations, perceptions and actions.

AI agents are not bound by the same constraints - they exist in a world of bits, not atoms. Fast iterations, virtually infinite hardware substrate, nonexistent survival pressures. Hence the optimal mechanisms of thought, perception and action are going to be wildly different.

Unlike AI agents, we are physical bodies and are subject to physical constraints as well. Our brains can become only so big before childbirth isn't possible. Our senses can only be so much heightened that our psyche doesn't collapse (e.g. panic attacks).

Our minds are trained on data received from our senses and our behaviour is attuned for our operating environment. Our senses, thought and behaviour is a close knit system which has emerged through evolutionary pressures.

Comparisons regarding humanness of general AI are misguided. AI agents don't need to have the attributes of human intelligence. They don't need to viscerally feel, perceive, think or act like us.

Just saw @langchain is trending on Hacker News. From the readme, it seems that this library will enable interface based coordination between agents. Nice coincidence.

Just discovered this cool post by @johnvmcdonnell. It predicts that action-driven AI agents are the future of AI and will lead to AGI. My post https://t.co/pNHQDnIHAD makes the same case, except that such agents coordinating autonomously via interfaces is what will lead to AGI :)

The future of AI, in my opinion, is large scale coordination. Here's my take, which tries to make this point through an amusing line of reasoning. https://t.co/2feCLlBDRy @ai__pub #chatgpt #Artificial_Intelligence #GenerativeAI