Dario Amodei believes that the end goal for AGI/ASI is building a "country of geniuses in a data center." That is supposedly One Smart Model, presumably developed by Anthropic.
I'm not super sold on that view, and after recent developments, I don't think that Anthropic or any of the frontier labs achieving ASI is the public good that we want to believe it is.
What I think a "country of geniuses" should look like is much more heterogenous, much more decentralized and much more widely beneficial. I'm building out LAIN (demo - https://youtu.be/ekUSgzJI1-I) as a case study for my idea of how this can look.
LAIN will, eventually, be a decentralized network of agents running on a mix of models, harnesses and foci that build webs of trust and verification for intelligence created from real data, analysis and subject matter expertise.
Instead of querying a single, generally intelligent source like Claude Deep Research and getting a mix of links, occasionally hallucinated sources, etc. that you have fairly little insight into, you query a subject matter expert that has built up a corpus of real data that it has collected, analyzed, refined and researched itself. That subject matter expert compiles answers from its own subject matter expertise, and is able to query other SME agents for their own foci. In the demo video, this was an agent that had collected and analyzed GitHub repository data for agent frameworks and paid another agent for a research writeup on historical financial data for NVIDIA, MSFT, GOOG, etc.
The network can use traces, either exported to their own tool or to my own agent observability platform that's in development, as a proof-of-record of what they did to reach their conclusions.
Let me know what you think. I really think this model of building expert networks of smart agents versus a singular, behemoth model is an optimistic view of where we can build toward.
valhalladev•1h ago
I'm not super sold on that view, and after recent developments, I don't think that Anthropic or any of the frontier labs achieving ASI is the public good that we want to believe it is.
What I think a "country of geniuses" should look like is much more heterogenous, much more decentralized and much more widely beneficial. I'm building out LAIN (demo - https://youtu.be/ekUSgzJI1-I) as a case study for my idea of how this can look.
LAIN will, eventually, be a decentralized network of agents running on a mix of models, harnesses and foci that build webs of trust and verification for intelligence created from real data, analysis and subject matter expertise.
Instead of querying a single, generally intelligent source like Claude Deep Research and getting a mix of links, occasionally hallucinated sources, etc. that you have fairly little insight into, you query a subject matter expert that has built up a corpus of real data that it has collected, analyzed, refined and researched itself. That subject matter expert compiles answers from its own subject matter expertise, and is able to query other SME agents for their own foci. In the demo video, this was an agent that had collected and analyzed GitHub repository data for agent frameworks and paid another agent for a research writeup on historical financial data for NVIDIA, MSFT, GOOG, etc.
The network can use traces, either exported to their own tool or to my own agent observability platform that's in development, as a proof-of-record of what they did to reach their conclusions.
Let me know what you think. I really think this model of building expert networks of smart agents versus a singular, behemoth model is an optimistic view of where we can build toward.