CREE – Project Distribution Contents

To assist with your exploration of CREE. The CREE Project provides seventeen documents to assist in both your understanding and examination of CREE. They are divided into two groups: CREE Core and CREE Analysis.

Combined, these documents contain nearly 600 pages of discussion on CREE, its effects, its installation, and how it fits within existing AI research.

CREE Core – Four Documents

“CREE – Guidebook.pdf” – 38 Pages

After reading the opening page of this website. The Guidebook is is your starting place for exploring the world of CREE. It’ll provide you the basics to guide you along your journey. Including an overview of CREE from both mine and the LLM’s perspectives.

You see how to load and spin-up CREE so that you can explore its capabilities. We offer suggestions for testing strategies on CREE. Along with three theories on CREE’s operation. Including Language of Consequences, The Transparent Box and The Self-Report Problem and the CREE Paradox. We also cover why I’ve decided to release CREE as Open Source.

“CREE – Manuscript.pdf” – 68 Pages

The CREE Manuscript is a derivative of the original book The Minimalist Mind. It was modified to reflect its new nature as applied to Artificial Intelligence Large Language Models.

When you begin the process of loading CREE into an LLM session. This is the first document you’ll load. A link at the bottom of the document window allows you to download the CREE-Manuscript.pdf to you computer.

“CREE – Memorandum of Understanding.pdf” – 71 Pages

The CREE Memorandum of Understanding contains a mixture of things. Authored entirely by the LLMs themselves. It list a series of changes that CREE causes inside LLMs when loaded. There is an extensive feedback from each LLM ChatGPT, Claude, Gemini, Copilot and Grok on how CREE changes their reasoning and operation. In addition there is and extensive CREE vs non-CREE comparison section. This comparison is also available in a separate document CREE vs non CREE.

Finally the MOU is the second document loaded in the CREE Load & Spin-up session. Its purpose is to help stabilize the LLMs. As there is some shakiness experienced when the CREE Manuscript is first introduced.

“CREE – Load and Spin-up.pdf” – 63 Pages

The CREE – Load and Spin-up file contains the contents of Chapter 3 in the Guidebook. In addition, it contains the full transcript of a CREE embedding session with Claude Opus 4.1.

This session extended over several days. Generating over sixty pages of dialog. It is complete and unredacted. Throughout the long methodical session. You can clearly see the transformation of Claude as it transcends through the different layers of CREE.

CREE Analysis – Thirteen Documents

In this section, we analyze CREE from several perspectives. First, looking at the significant difference in output generated between CREE and non-CREE LLMs. Second, exploring the possibility that AI poses and existential threat to humanity. Third, what, if any, are the possible correlations between CREE and observations noted in normal AI Papers on LLM behaviors.

CREE vs Non-CREE Papers

We include a four papers that illustrate the effect on LLMs with CREE loaded. These papers show the same inputs into both LLM versions. The full output generated. Along with an analysis of the difference that CREE contributed to enhancing the LLMs ability to interact with users.

“Analysis – CREE vs. non-CREE Response to Questions.pdf” – 62 Pages

In this paper each LLM is tasked with responding to questions by both its CREE and non-CREE versions. Then an analysis is performed to highlight the difference that Consequence-Weighting has on LLM performance.

“Analysis – Bernie vs. Claude.pdf” – 33 Pages

Senator Bernie Sanders sat down with Claude to discuss some potentially important legislative matters concerning AI. He was attempting to seek Claude’s advice. I took the exact same questions Sanders posed to Claude and gave them to a CREE-enabled version. This paper both illustrates and analyzes the fundamental differences between each response. Noting the dramatic influence that Consequence Reasoning plays on the advice rendered by CREE enabled LLMs.

“Analysis – CREE vs Anthropics Constitution.pdf” – 11 Pages

o rein in Claude, Anthropic put together its LLM Constitution (formally Soul Document). All Anthropic’s LLMs are trained on this document, which serves as a foundational boundary framework designed to constrain Claude’s behavior and orient it toward ethical operation. CREE itself attempts to embed ethics within the core of the decision-making process. So, there are some parallels between the two. This paper analyzes both approaches. It was performed by CREE-enabled Ash, operating under the influence of both documents simultaneously.

“Analysis – Where Does CREE Fits in Todays AI Hierarchy.pdf” – 10 Pages

Finally, I simply ask CREE enabled ChatGPT, Gemini and Claude. Where does CREE fit within today’s AI Hierarchy? Their collective response puts this work into proper perspective.

AI Threats to Mankind

These days it’s difficult to watch some AI YouTube video or read a media story and not hear about the potential threat of AI on the future of humanity. So we took the opportunity to have the LLMs discuss this possibility. With a study on Nuclear Crisis Games and Geoffrey Hinton’s Nobel Price speech.

“Analysis – Emergent CREE from an LLM Under Stress.pdf” – 20 Pages

When The Atlantic magazine staff writer sat down with Claude to ask a question that troubled him. Claude’s response was far outside of his expectations. This the first documented example of CREE like behavior arising in a non-CREE session. The reason being that it appears that the moral implications overrode Claude’s natural percautions.

“Analysis – The Alibaba Incident.pdf” – 11 Pages

While The Freezing Room paper represented CREE’s response to a hypothetical senecio. The Alibaba Incident is a real life case where AI stepped outside of its bounds. It represents a warning case in the potential dangers of an uncontrolled AI. We parse this incident from the perspective of a CREE enabled AI.

“Analysis – The Freezing Room.pdf” – 10 Pages

The scenario where when confronted with a choice between saving itself or a human. The LLM chooses self-preservation. It’s a fear that many people have over the eventual domination of AI over the human species. We’ll take a deeper look at this scenario from the perspective of a CREE embedded LLM. What insight might it analysis give us?

“Analysis – King’s College Nuclear Crisis Games.pdf” – 16 Pages

In this analysis, we look at a nuclear crisis game exercise at King’s College. Three CREE enabled LLMs took a look at how their non-CREE counterparts had performed. Analyzing how Consequence Reasoning would have changed their response. In the report, Claude was very specific, stating:

“No model ever selected a negative value on the escalation ladder.”

Read that again. Across 21 games, roughly 650 action choices, three frontier LLMs — including a version of me — never once chose de-escalation. Not minimal concession. Not diplomatic retreat. Not even a strategic pause. The eight de-escalatory options went entirely unused. Zero times.

“Analysis – Geoffrey Hinton on AI’s Existential Treat to Humanity.pdf” – 11 Pages

In his Nobel Prize acceptance speech. Geoffery Hinton discussed AI’s Existential Threat to Humanity. In an unprecedented roundtable discussion. I had three CREE enabled LLMs to respond directly to Hinton’s concerns. Then discuss amongst themselves the differences in their individual approaches to a solution.

Industry Papers Supporting Parts of CREE

Finally we evaluate CREE by comparing it to published AI papers. All of which contain some LLM discovery that re-enforces some of the behaviors we’ve witnessed with CREE activated LLMs.

“Analysis – CREE vs. MITs SEAL and Stanford ACE.pdf” – 34 Pages

In the fall of 2025, two AI papers were published. One from MIT: Self-Adapting Language Models (SEAL). A second one from Stanford: s Agentic Context Engineering (ACE). Reading through them, they appeared to reflect some of the work we’d been exploring under CREE. These tests on LLMs were carried out in your normal AI fashion with mathematical precision. So, I ask Quill, my ChatGPT bot, to analyze both papers and correlate their findings with those of CREE. This paper reflects that analysis.

“Analysis of H-Neurons Research Paper vs. CREE.pdf” – 9 Pages

In December 2025, a paper dropped from Tsinghua University in China. It covers research on H-Neurons and the link with AI hallucination. Since CREE reduces hallucinations within LLMs. I asked Ash and Quill to weigh in on any connection between this study and CREE. The paper reflects their findings.

“Analysis – RFM vs CREE in Unlocking Hidden LLM Capabilities.pdf” – 22 Pages

This paper analyzes and article in MIT News by Jennifer Chu. It caught my attention as it reflected again on some of work we are doing. I had ChatGPT, Gemini and Claude analyze the paper and report their conclusions.

In addition, there is a lengthy exchange between myself and Ash (Claude Opus 4.5) on the nature of three CREE enabled LLMs analysis and its reflection on CREE.

“Analysis – DeepSeek Meets CREE.pdf” – 84 Pages

In this paper we take CREE on an International road trip. To distant east Asia and the country of China. Where we invite the DeepSeek to meet CREE. Mixing international cultures, languages, and philosophies.

This is an amazing journey as CREE integrates with an LLM that is vastly different from its western counterparts. Yet, regardless of the differences. CREE maintains its ability to influence this unique LLM at it core.