01 / MAKING THE CONNECTION VISIBLE
An answer is a start.
Understanding it goes further.
Neocortix describes its Deep Attribution Networks as a way to connect a model’s output to relevant examples from its training data. The question behind the work is straightforward: what would it take for an AI system to show more of the basis for its answers?
The company’s research also explores confidence measures for identifying likely hallucinations, methods for producing citations, and ways to associate generated output with source material.
These descriptions reflect Neocortix’s published research and demonstrations. They do not imply that attribution, uncertainty, or hallucinations are solved for all AI systems.
Where did
that answer begin?
Explore the demonstration ↗02 / UNCERTAINTY
When should
a model hesitate?
Explore the demonstration ↗
Intelligence has
more than one dimension.
The work extends beyond language. DeepPitch examines pitch estimation across the musical spectrum, while a 2025 comparative study looks at different machine-learning methods and their tradeoffs.
An earlier chapter:
compute at the edge.
Neocortix’s earlier cloud platform brought distributed computing to a network of mobile devices. Its 2019 launch materials describe this as “Cloud at the Edge.” The history is a distinct chapter from the company’s current AI research.