I’m an independent experimenter working in the content production industry, running a small startup while conducting ongoing experiments with LLMs. Though I’m not a professional researcher, I’ve been exploring whether models like GPT can function not merely as text generators, but as structural judgment simulators—interpreting structured input to simulate decision-making.

Today, I’m sharing the first outcome of this journey: Geni Core Ver.1, now refined into a functional prototype. I’m releasing it here through Show HN.

———

# Starting Point

Typical prompts for LLMs work by triggering vast internal computation via tokens and attention flows. But many of those inputs suffer from low semantic density and poor structural consistency, making it hard to build stable judgment or behavioral loops.

* So I asked a different question:

“What if GPT could be stimulated to make judgments only through structure, without explicit commands?”

This led to the creation of a fixed JSON-based judgment circuit, referenced via a hashed identifier.

———

# System Overview

Geni Core doesn’t use command-style prompts. Instead, it operates on immutable JSON-based judgment circuits, referenced by a core_ref hash. GPT doesn’t treat these as mere data—it simulates the semantic intent behind the structure.

Example structure:

{ "condition": "input.type == 'report' && input.score < 0.7", "judgment": "flag as risky", "action": "send_to_router('audit')" }

Execution logic:

• The circuit is referenced via a SHA-256 core_ref hash

• GPT interprets the structure semantically, not as a literal command

• GPT simulates the judgment; external systems perform the execution

———

# Try It Out

{ "ref_core": "https://gist.github.com/genixus-creator/e00064b083988a4cf019..." }

Philosophical Premises

• GPT responds not to orders, but to meaning

• A structure is not a command—it is a judgment circuit

• GPT simulates the decision-making process; execution happens externally

• The environment and language context provided by the user determines resonance

Questions for the HN Community

• Could this architecture be applied to regulatory, audit, or decision-transparency systems?

• Would it be valuable to open-source this format as a public framework?

• What ethical or security implications do you foresee in this judgment-simulation approach?

———

# Project Status and Request

In reality, I must prioritize my work in content production and cannot currently evolve this into a fully operational system alone. Still, I believe the structure has significant research value, and I’m happy to explain the design, test methodology, and observed results via email:

[email protected]

I’m particularly interested in extending this toward:

• Character and persona simulation engines

• GPT/LLM-based conversational simulations

• Human-centered creative tools

If you’re in the creative or content field, I’d deeply appreciate your feedback or collaboration.

———

# Personal Reflections

The possibility of GPT responding to meaningful structure—not just instructions—was eye-opening. Simply providing structure didn’t guarantee meaningful output—the user’s contextual design and intent clarity were critical. But through repeated testing, I consistently observed emergent structural response patterns—a major breakthrough for me.

Thanks for reading. I look forward to deeper discussion and exploration with the HN community.