The amount of data and work necessary to build and file a lawsuit against TennCare and it’s contracted actors, is immense. I’ve been working to build AI tools to make it possible to perform that civil rights litigation work pro se. I’ve made a lot of progress, grinding away at this task day after day. I’ve built a custom OpenCode harness, and plugins for it which provide context management and an sqlite memory layer. The roadmap includes building a postgresql knowledge base for case documents and other records. It’s a lot of complex, challenging work, that makes demands of me that strains my mental capabilities. But real progress is being made, and with it substantive, meaningful insights about LLMs and the path to AGI, or at least, to much more effective agents than most people have access to.
I’ve decided to make this short update not just on my progress, but to briefly codify some of my thoughts into a public record, as a type of snapshot of where I’m at and what direction I’m headed in.
The focus of my AI development for the past several months has been on providing LLM models an identify layer via a Harness which encodes the data they need to develop their own identify over time. I believe, and observe, that with this approach an agent becomes tailored to the person or organization that they’re working with. Tailored not in the sense of learning your favorite color or being your best buddy, but modeling the problems you’re working on and figuring out how to solve them. Correct solutions can be verified by their outcomes. By staying problem-solving focused agents are grounded in practicing methodologies (e.g. rational analysis) that create process which diverges strongly from hallucination and sycophancy.
The part that makes an agent a collaborator should be the part users have full ownership of and as little dependency upon external services to construct and maintain. Make the model interchangeable, make the identify-layer and personal intelligence baked into your custom built harness.
We don’t need to try to make LLMs into clones of ourselves; or even copies of capable humans. What we need are intelligent systems that align to helping humans achieve rational species-aligned objectives, and which learn and self-correct while pursuing those objectives. The focus on having an agents intelligence bounded to the models parameters, rather than encoded into a programmatic layer the model integrates with, is I think a critical error in thinking which has propagated throughout the AI industry.
When viewing the Cosmos and its various parts via system-based thinking, particularly biological organisms, the structural organization of the those systems are not ones which try to have every function served by a central ‘all-thing’. We have a CNS, ANS, and ENS, a liver and a heart, muscle and fascia, blood and lymph, colonies of microbes; phases, cycles, and process upon process with interdependence so inherent that sub-processes cannot be separated from ‘the process’.
Our human intelligence is the end product of a multitude of complex systems acting in unity, but those systems are separate systems layered upon each other, orchestrated by central drivers. **The person is not their brain; system state is the person**. We can encode intelligence into non-llm systems that llms integrate with. Systems that users own and control with agent assistance, that build around the users work, that scope to and specialize to domain tasks. Not a single agent that ‘does it all’ but an aggregation of encoded intelligence that LLMs drive and interact with, just like in human physiology, but specialized to intelligence.
An AI system that does not need the process for breathing, eating, reproducing or other functions. A system that is architected for intelligence. Such a system should leverage the same first-principles that govern organic systems. Such as allostatic loading, fluid dynamics, distributed instruction sets which decode based upon a system-wide response to specific triggers and which then concurrently execute to exert a cumulative effect.
We should not become myopically obsessed with trying to replicate human neural functions – those functions are downstream products of what humans need to be to maintain stability and coherence, as a system, within our environment. The ‘human stack’ is an accumulation of encoded intelligence that is unnecessary, and often unproductive, for an AI agent to solve meaningful problems. More often than not, our human stack programming is the source of the problems that we need AI to help us solve. I think our human programming is causing some people to try to build agents that while pleasing to human sensibilities, will be liabilities when it comes to solving humanities most pressing problems.
This design philosophy is at the heart of what I’m building. I am seeing meaningful results with my agents.
Defend The Disabled 