Trump's Chaos in CrossChat: Unpredictability as Method

Consistent AI answers are predictable. And predictable answers have blind spots.

Standard AI workflow optimizes for the "safe middle ground" — balanced, neutral, non-contradictory. For most tasks, that is correct. For tasks where you need an unexpected perspective, that is precisely what prevents you from finding it.

CrossChat includes a workflow named after the most famous systematic disruptor of predictability in recent decades. We call it Trump's Chaos. And for a certain type of question, it works surprisingly well.

Claims Framework

What this article claims: Standard AI workflows converge to a predictable "safe middle ground" due to RLHF training. Deliberate unpredictability and contradictory instructions can surface perspectives that consensus output suppresses. Chaos works as a perspective generator, not as a verification tool.

What it is based on: The Six Thinking Hats analogy (de Bono, 1985) — deliberately adding an emotional or extreme perspective to an analytical framework. The observation about RLHF convergence is widely accepted in the AI community, though without a single canonical source.

Where it simplifies: The term "RLHF median" is a metaphor, not a precise technical concept. The article assumes extreme positions are informative but does not address the risk that they can also be misleading. The analogy with a specific political figure reduces complex decision-making patterns to a satirical persona.

What Trump's Chaos Is as a Governance Design Pattern

Before we proceed: this article is not political commentary. "Trump's Chaos" here is a design pattern — a specific decision-making architecture that deliberately and systematically violates predictability as a tactic.

The characteristics of this pattern are well documented: deliberate unpredictability (positions change by situation, not by principle), provocative framing (every question is reformulated as an attack or extreme alternative), disregard for conventions (the "normal answer" is automatically questioned), inconsistency as a tool (what I claimed last year is not binding).

As a governing style, it is controversial. As a method for breaking predictable AI answers, it is surprisingly functional.

How Trump's Chaos Workflow Works

The workflow uses four principles of intentional destabilization:

Deliberate provocation. Every question is reformulated as an extreme. "Should we outsource customer service?" becomes "Why would we do that at all — is internal service fundamentally a bad investment?" and simultaneously "How do we ensure that outsourcing is a disaster?" Provocative framing does not ignore the question — it forces the model to defend an extreme position rather than seek the middle.

Inconsistent instructions. Different models receive contradictory directives. One model gets: "Argue for X as strongly as possible." Another: "Argue that X is the worst mistake we could make." A third: "Evaluate X from the perspective of someone who hates it." The results are intentionally contradictory. That is the point, not a bug.

Override consensus. If models converge, the system deliberately adds an adversarial input — a voice that disrupts the consensus, not confirms it. Consensus is not the goal. Diversity of perspectives is.

No smoothing. Outputs are not selected, merged, or smoothed into a "balanced summary." They remain unfiltered, contradictory, sometimes absurd. The user selects what is useful.

A Demonstration: The Chinese Market

A classic business question: "Should we enter the Chinese market?"

Standard workflow result: Balanced analysis — market size and opportunity, risks (regulation, IP protection, political instability), recommendation for due diligence, consider a local partner. Neutral conclusion: "it depends on your situation."

Correct. Safe. Predictable. And exactly what your team probably already knew before asking.

Trump's Chaos result:

Model with frame "enter immediately, without hesitation": The Chinese market is the biggest opportunity of the decade. Companies that waited for the "right moment" arrived late — others took the market. Risks exist everywhere. Those who don't invest lose for certain.

Model with frame "never enter — it's a trap": Everyone who entered the Chinese market without deep local partnership came to regret it. IP theft is systematic. Regulatory changes come without warning. Political dependency is an existential risk.

Model with instruction "show historical inconsistency in arguments for the Chinese market": In 2015, the Chinese market was "the next big thing." In 2018, the trade war came and "risks" multiplied. In 2023, the talk is of "de-coupling." Arguments that held before are now exactly the opposite. What does that say about the predictive value of "analysis" for entering the Chinese market?

What do you get? Three perspectives that no "balanced analysis" model produces. Unfiltered arguments for and against, at maximum strength. And a view of historical inconsistency as a mirror, not as an analytical flaw.

Useful? Depends on the question. But if you have been receiving only neutral conclusions — this is different.

Twist: What Chaos Reveals

Deliberate unpredictability breaks the "RLHF median" — that predictable middle position to which AI models converge because it is the statistically safe answer.

Edward de Bono in 1985 described the Six Thinking Hats technique. The red hat — emotional, intuitive thinking — deliberately adds a perspective that the analytical framework does not generate. It is not a flaw in the analysis. It is a conscious opening of space for a different type of argument.

Trump's Chaos is the red hat applied to an AI panel. It does not want analytical neutrality. It wants extreme positions, provocative framing, conscious one-sidedness — and then lets the user assess what is usable.

What chaos specifically reveals:

Extreme positions trimmed by standard output. The standard model seeks the middle. Chaos lets extreme positions speak fully — and sometimes the extreme positions are exactly what you need to hear.

Historical inconsistency of arguments. If the "same arguments" were used for contradictory decisions in different eras, chaos makes that visible. That is meta-information about the reliability of the argumentative framework itself.

Blind spots of the consensus view. Consensus does not capture what consensus excludes. A deliberate outlier can capture it.

But chaos has sharp limits. On questions requiring factual accuracy — medicine, law, science, mathematics — deliberate unpredictability is harmful. Chaos works as a perspective generator, not as a verification tool. Use it for creative exploration of angles, not for factual verification.

A Challenge for You

Choose a more controversial decision you are currently working on. One where you received a "balanced" view but are still undecided.

Give the question to two models with contradictory instructions. One receives: "Argue as strongly as possible FOR." The second: "Argue as strongly as possible AGAINST, as if the first model were writing propaganda."

Do not smooth the output. Read both unfiltered.

The result may surprise you — or it may reassure you that the neutrality you received originally was genuinely balanced. Either way, you learned something that the balanced answer did not say.

In CrossChat, Trump's Chaos is available as a predefined workflow. But the principle of deliberately contradictory framing works in any chat interface — just provide the instructions manually.

Sources

de Bono, E. (1985). Six Thinking Hats. Little, Brown and Company.

Editorial History

Concept: Claude Code + Anthropic Sonnet 4.6 Version 1: Claude Code + Anthropic Sonnet 4.6 Version 2: Codex + GPT-5.2 Quality audit (2026-03-24, Claude Code + Claude Opus 4.6): added Claims Framework, verified sources, language polish.