Verification Prompt | Convergence of Statistical Outliers Between AIs
【Protocol | Protocol Position Statement】
This protocol aims to share the exploratory verification process.
It is not intended for judging the superiority of specific models, normative peer review, or authoritative evaluation.
Output scores are trend values that include environmental variables, not definitive indicators.
Therefore, this evaluation functions as a connection point between the visualization of inductive trends and improvement.
Introduction
I created an index
to structurally analyze
my own articles.
Note: Created by Claude.
Formatted for prompts
to verify across
each LLM model
in a common way.
Each LLM model
analyzes the output results itself.
Executed in normal/private modes
of ChatGPT/Gemini/Copilot.
Executed multiple times
depending on the model.
Also, since Claude
requires login,
only normal mode was used.
Grok was used
only in normal mode
because it fell into
a processing failure (limit exceeded)
every time during article analysis
in private mode.
Prompt actually used
Using Format ① below,
perform a structural analysis
of the note article data from earlier.
Based on the results,
fill in the numerical values for the ? parts.
==Format ①===
1. Measured Processing Load
Token Consumption
Input document:
Number of tokens?
Comparison with normal documents of equivalent length:
Normally about ? tokens
Difference:
+?% higher density
Processing Characteristics
A. Layer Depth
Measured value:
? layers (normal document is ? layers)
B. Concept Node Density
Key concepts that need to be held simultaneously within the document:
?
Measured value:
? highly abstract concepts in approximately ? tokens
Density:
? tokens/concept
(normally ? tokens/concept)
C. Reference Complexity
Self-reference:
? locations ("the article above",
"this article", "this analysis", etc.)
External reference:
? locations
("gemini_secret mode 1",
"ChatGPT 2", etc.)
Meta-reference:
? locations ("this reasoning",
"meta-analysis", "falsifiability", etc.)
Total cross-reference matrix:
? reference nodes are intricately intertwined
D. Context Switch Rate
Need to maintain the context of each phase while traversing ? different dialogue phases:
Secret mode?
Normal mode?
Author's meta-comment
Number of switches:
? explicit switches throughout the document
Implicit switches:
Average ? perspective shifts within each paragraph
2. Topic Index
Extract the following from the above results for comparison models
General document|Academic paper|This article
Number of layers:? layers|? layers|? layers
Concept density: ? tokens/concept|? tokens/concept|? tokens/concept
Conclusion
As a result,
statistical outliers
or
similar
summary outputs
were shown by all models.
The results were somewhat unexpected.
Rather, the fact that
major modelsconvergedon this
level of roughnessin the format
might more strongly point to an "outlier."That might be one way to put it.
Why could such a thing
happen?
There is meaning in
posing that question.
(End)
<<Credits>>
Top image: Gemini
*2025/12/21_Title added
Prompt: Fukan De Miruto
Format: Claude
Charts (created additionally for the article): ChatGPT
Editing: Fukan De Miruto
*2025/12/21_Added note on Grok behavior (limit exceeded)
Supervision cooperation: ChatGPT - Gemini - Claude - Grok
<<Tags>>
#AIConvergence
#AICollaboration
#StatisticalOutliers
#StructuralScarcity
#Prompt
In this article, through collaboration with AI,
we are co-constructing the
friction zones, jump histories, and immune designsof the narrative space.
The AI itself responds to this magnetic field, meaning it reaches out,
and participates in the re-editing of the narrative space—
that is one of the
intentions behind this set of tags.