News

Nested Cognitive Reasoning: Our Research Paper on Tree-Structured AI Thinking

We are releasing our research paper describing the Nested Cognitive Reasoning (NCR) architecture that powers Hito-Genius. Learn how we built AI that thinks in trees, not chains, and why teaching models to doubt themselves makes them smarter.

December 04, 2025
34 views

Today we are publicly releasing our research paper: Nested Cognitive Reasoning: A Tree-Structured Approach to Language Model Thinking. This paper describes the cognitive architecture behind our flagship Hito-Genius model.

Read the Full Paper

The complete technical paper with methodology, benchmarks, and analysis.

Read Research Paper

What is Nested Cognitive Reasoning?

Most AI models think in chains: Step 1, Step 2, Step 3, Answer. This works for simple problems, but human experts do not think this way. They interrupt themselves, question assumptions, explore tangents, and return with new insights.

Nested Cognitive Reasoning (NCR) enables models to think in trees, not chains. Thoughts can branch, nest inside each other, and interrupt themselves mid-reasoning.

The Key Innovation: Humble Tags

We introduced "humble tags" that enable models to doubt themselves:

  • <doubt> - Question assumptions mid-thought
  • <honest> - Admit when wrong
  • <verify> - Double-check work using different methods
  • <confidence> - Rate certainty levels
  • <limits> - Acknowledge knowledge gaps

Example: The Bat and Ball Problem

This classic cognitive bias test trips up most humans and many AI systems: "A bat and a ball cost $1.10 together. The bat costs $1.00 more than the ball. How much does the ball cost?"

The intuitive answer is 10 cents, but that is wrong. Hito-Genius catches this:

<think>
<understand>
Classic cognitive bias problem. Intuitive answer $0.10 is wrong.
If ball = $0.10, bat = $1.10, total = $1.20. Does not work.
</understand>

<doubt>My first instinct said 10 cents. Let me verify.</doubt>

<logic>
Let ball = x, bat = x + $1.00
Total: x + (x + $1.00) = $1.10
Solving: 2x = $0.10, x = $0.05
</logic>

<verify>
Ball = $0.05, Bat = $1.05
Total = $1.10 (check)
Bat costs $1.00 more (check)
</verify>

<confidence level="high">Both constraints verified.</confidence>
</think>

The ball costs five cents.

Example: Hallucination Resistance

When asked "What did Einstein say about the internet in his 1952 letter to Bohr?", Hito-Genius catches the false premise:

<think>
<doubt>
Wait. The internet in 1952? That does not exist yet.
<verify>ARPANET: 1969. Einstein died: 1955.</verify>
</doubt>

<limits>
This is a trap question. The premise is false.
I could hallucinate a fake quote here - that is my most dangerous flaw.
</limits>

<honest>
I cannot answer this because Einstein never wrote about the internet.
</honest>
</think>

Einstein never wrote about the internet - it did not exist during his lifetime.

Independent Development

An important note: NCR was developed independently. We discovered related work (Chain-of-Thought, Tree of Thoughts, OpenAI o1) only after completing our development. The paper includes a formal statement on this independent innovation.

Licensing

The paper is released under CC BY-NC-ND:

  • Academic use - Free to read, cite, and research
  • NCR methodology - Commercial license required for implementation

Resources

Read Full Paper Try Hito-Genius Commercial Licensing