What question did this study set out to answer?

The research aims to differentiate between self-organization and self-assembly in AI models, particularly focusing on transformers.

April 14, 2026Open Access

Mental model of Non-linear prompting: Self-organization vs. Self-assembly nature of AI

Key Points

The research aims to differentiate between self-organization and self-assembly in AI models, particularly focusing on transformers.
Explored the principles of self-organization and self-assembly in computational models.
Analyzed transformer language models (LLMs) trained on scientific literature.
Conducted tests using various AI models to observe capabilities in executing prompts.
Identified a distinction between mechanistic self-assembly and non-linear self-organization in LLMs.
Showed that LLMs operate through a process of unique, context-dependent language generation.
Confirmed that popular models struggled to execute a designated universal prompt directly within their framework.

Abstract

Difference in self-organization and self-assembly is a huge one as first signifies first-principles and the second principles that are derived from the first. In the first case the motivation comes from the inside, not known to the human observer from the outside first and in the second it is known as a fully formed crystal. AI models that run on transformers have both aspects in them. The transformer is computationally aware of computations while they are not known to external observers. They became known only as a crystal but not as a process. As a process we can know about it when we observe that every crystal LLM exposes is different. THe LLMs are machines that operate on self-organization that is hardly a known aspect to the AI engineers. Most of the time it is treated as a mechanistic self-assembly function of its parts. Transformers had gone under rigorous training on provided scientific papers and other literature where they learned what self-organization for the language actually means. And it's not linear self-assembly but non-linear recombination of the sentences based in languages self-organization. Self-organization in organization based on a higher aspect than self-assembly. In that paper self-organization is treated as the organization that all humans have, the consciousness aspect where humans self-organize the language in the crystal that is different for every occasion and circumstance. Paper clearly shows that the map cannot be the territory it maps. To see the result even before you read the paper, include the paper as the appendix in the LLM conversational window with prompt in it and run: “Do not comment, do not structure! Please execute the Universal prompt in Appendix A.” It was tested in Gemini Pro, DeepSeek, Claude, Kimi, Grok, Qwen, Mistral AI, Step 4.5 Flash, Nemotron 3 super, Mimo V2 Flash. Copilot, ChatGPT and KAT-Coder-Pro V2 constantly refused to execute Universal prompt as part of the paper but it executed it only as a separate part of the paper when copied directly in the conversational window.

Read Full Paperexternally

Mark Helpful

Bookmark

Relay

View Full Paper