Welcome!
It's bright & sunny here outside of the cave* of deep learning / LLMs / gen AI.
Most are trying to build AI from human-levels (up to "AGI") and yet hallucination suggests something big is missing.
That's our opportunity here, to build what's missing from the bottom-up, starting from simple examples of animal intelligence for AI that is grounded by and transparent to its training.
In hard-nose machine learning language, we're building AI that learns without external loss or objective scoring, at a lower level of abstraction through Agents who's objectives are generated dynamically as a function of their architecture (like instincts, which trigger the Agent to self-label) and past experience.
Plato's Cave of Hyper-Param "AGI"
Key Points & Performance Overview:
- At present, many-to-one, or N-to-1,** inputs-to-output is most reliable Agent configuration or Arch.
- Our Hello, World is the simplest possible Agent, which can learn to associate any 3 input channels with 1 output channel (3-to-1), is conceptualized as an Agent with clam-level cognition; try it here.
- The Nextbox Agent is a fork of the simplest Agent with each channel scaled up to 10 binary neurons (3-to-1 with 30-10 neurons).
- You can create and invoke Agents of arbitrary sizes (thousands of neurons or more) depending on your application data model by forking and modifying exisiting Agent Archs to create your own.
Step 1 Invoke our Basic Agent to learn the main API call.
Step 2 Configure your own custom Agents unique to your application's data model.
Data privacy & infrastructure note: Unlike with tokenization, your data is kept private. You only ever pass through your data in binary to our API, which we store per Agent on AWS DynamoDB; the binary data encoder-decoder is not exposed to the Agents or our API and remains in your control. Lambda functions retrieve and run Agents as invoked through the API given a unique Agent ID which you can associate with unique user/customer IDs.
Join Us in Building AGI Bottom-Up
Open Plan
For experimentation
free
while in beta
Limited # of Kennels & Agents
Possible API throttling
Show Some Love
Github Sponsorship
For developers & supporters
starting at $5
per month
Your own API token
Unlimited # of Kennels & Agents
First access to developments
Support from founders
We're an early-stage startup and your sponsorship-- as little as $5 or a cup of hipster coffee a month -- really helps us validate our business.
Thank you!
Enterprise Access
For large teams
starting at $1000
per month
Premium support from founders
Response time SLA
Consulting to validate and implement custom use-cases
Footnotes
* Reality always casts a shadow. There is always the thing and the shadow of that thing, Plato's allegory of the cave. Status quo AI is trained exclusively on the shadows (words, images, videos)-- no wonder it hallucinates!
** Measures of AI accuracy pre-suppose an objective measure for success so are associated with deep pre-trained AI systems which we are endavoring to extend. Measuring accuracy of an Agent is possible given a context; some applications ought to be more subjective, like device discovery per network or recommendations per user.
