The Fundamental Problem

LLMs excel at producing text that sounds authoritative, flows beautifully, and follows proper grammar. But they often generate what we might call “semantic foam”—text that appears substantial but lacks actual information content. They hallucinate facts, create plausible-sounding but empty explanations, and can dilute strong ideas with verbose elaboration.

This creates a critical question: In an age where AI can generate unlimited text, where does actual value come from?

Information Density Defined

Information density is the ratio of meaningful, accurate, actionable content to total text volume. High-density information:

• Contains specific facts, not generalities
• Provides unique insights, not common knowledge
• Offers concrete examples, not abstract concepts
• Makes falsifiable claims, not vague assertions
• Connects to real-world consequences

The Human Monopoly on Meaning

This is the crucial insight: Humans provide the information density that makes AI output valuable.

We contribute:

• Domain expertise - Real understanding of the problem space
• Contextual knowledge - What actually matters to stakeholders
• Ground truth - Facts that can be verified
• Strategic thinking - Why certain approaches matter
• Lived experience - Stories and examples that resonate

The AI contributes:

• Language refinement - Better articulation of our ideas
• Format consistency - Professional presentation
• Exploratory improvisation - “Happy accidents” that spark new thinking
• Synthesis capability - Combining ideas in novel ways

The Super Spell-Checker Paradigm

Thinking of AI as a “super spell-checker” is more profound than it first appears:

Traditional Spell-Checker

• Corrects spelling errors
• Fixes basic grammar
• Suggests simple improvements
• Preserves your meaning exactly

AI as Super Spell-Checker

• Refines expression of ideas
• Improves document structure
• Suggests alternative phrasings
• Explores adjacent concepts
• Still preserves your core information

The key: Like a spell-checker, it enhances presentation without replacing substance.

Why This Matters for Value Creation

Our value to customers comes from the information we provide, not the words we use to convey it.

Customers pay for:

• Solutions to their specific problems
• Insights they couldn’t generate themselves
• Expertise applied to their context
• Strategic thinking about their situation

They don’t pay for:

• Generic advice dressed in fancy language
• Verbose explanations of obvious concepts
• Hallucinated facts that sound plausible
• Beautiful documents with no substance

The Grounding Strategy

To maintain information density when using AI:

1. Front-Load Real Information

Provide the AI with:

• Specific data points
• Real examples from experience
• Concrete constraints and requirements
• Actual stakeholder feedback
• Domain-specific knowledge

2. Maintain Editorial Control

• Review every claim for accuracy
• Remove or correct hallucinations
• Trim verbose sections that add no value
• Ensure examples are real, not invented

3. Preserve Happy Accidents

When AI improvisation generates interesting ideas:

• Evaluate if they add real value
• Ground them in actual possibilities
• Connect them to real-world applications
• Keep what strengthens the work

The Economic Imperative

As AI makes text generation essentially free, information density becomes the scarce resource:

What Becomes Commoditized

• Basic writing ability
• Grammar and formatting
• Generic explanations
• Boilerplate content

What Remains Valuable

• Original thinking
• Specific expertise
• Real-world experience
• Accurate, verified information
• Strategic insight

Practical Implementation

For Individual Work

1. Start with substance - Write key facts and insights first
2. Use AI for polish - Let it improve expression, not content
3. Fact-check everything - Verify all claims before delivery
4. Measure density - Can you compress without losing meaning?

For Team Collaboration

1. Humans provide information - Team members contribute expertise
2. AI handles formatting - Consistent presentation across documents
3. Review focuses on substance - Is the information accurate and valuable?
4. Iterate on density - Remove fluff, keep substance

For Client Deliverables

1. Never auto-generate - Always human-review before delivery
2. Information audit - Does every section provide real value?
3. Source verification - Can we defend every claim?
4. Value justification - Why would client pay for this information?

The Competitive Advantage

Organizations that understand information density will thrive because they:

• Deliver actual value, not impressive-sounding fluff
• Build trust through accuracy and substance
• Maintain pricing power through expertise
• Can’t be replaced by generic AI tools

Those that don’t will find themselves:

• Competing on price with AI-generated content
• Losing credibility through hallucinated facts
• Providing interchangeable, commodity output
• Racing to the bottom on value

Connection to Broader Themes

Complexity Collapse

Low information density might contribute to Evidence of Complexity Collapse in LLMs—models fail when they lack substantial grounding information.

Agency of Agents

In Agency of Agents - ThinkNimble’s AI Collaboration Framework, each agent should add information density, not just process existing text.

The AI Onion

Each layer in The AI Onion - Layered Approach to AI Implementation should increase information density, not just sophistication.

The Bottom Line

In the AI era, our value doesn’t come from our ability to write—it comes from having something worth writing about. AI can help us express our ideas more clearly, format them more professionally, and explore adjacent possibilities. But the core information, the substance that makes the work valuable, must come from human expertise, experience, and thought.

We must retain control of information density because this is where our value to customers comes from.

AI is a powerful tool for enhancing how we communicate our ideas. But without high-density information to communicate, we’re just generating semantic foam—impressive-looking but ultimately empty, like a souffle that collapses at first touch.

Academic Foundation & References

Information Theory Origins

Claude Shannon (1948) established the mathematical foundation of information theory in “A Mathematical Theory of Communication.” Shannon defined information not as meaning but as the reduction of uncertainty. His work introduced key concepts:

• Information entropy - A measure of average information content
• Redundancy - The difference between actual and maximum possible information
• Channel capacity - Maximum rate of reliable information transmission

Shannon’s insight: Information is inversely related to predictability. The more surprising a message, the more information it contains.

Related Academic Concepts

Information Density appears in multiple fields with related but distinct meanings:

1. Linguistics: Information density refers to the amount of information conveyed per unit of linguistic material (words, syllables, time). Researchers like Levy (2008) and Jaeger (2010) study how speakers optimize information density in communication.

2. Data Compression: The Kolmogorov complexity measures the shortest possible description of a string, essentially its irreducible information content.

3. Cognitive Science: Processing fluency research (Reber et al., 2004) shows humans prefer information at optimal density—not too sparse, not too dense.

4. Technical Writing: Information Mapping® (Horn, 1969) methodology emphasizes chunking information for optimal density and comprehension.

The Bullshit Problem

Harry Frankfurt’s “On Bullshit” (1986) provides philosophical grounding for our concern. Frankfurt distinguishes:

• Lying: Deliberately stating falsehoods
• Bullshit: Speaking without regard for truth
• LLMs, arguably, are supreme bullshitters—they generate plausible text without understanding truth

Signal vs. Noise

Nate Silver’s “The Signal and the Noise” (2012) popularized the concept of signal-to-noise ratio in data analysis. In our context:

• Signal: Information that changes decisions or understanding
• Noise: Text that sounds meaningful but adds no value
• Information density: Essentially the signal-to-noise ratio of text

The Semantic Foam Phenomenon

While “semantic foam” isn’t an established academic term, related concepts include:

• Gish Gallop: Overwhelming with numerous arguments regardless of accuracy
• Deepity (Dennett, 2009): Statements that seem profound but are trivial or meaningless
• Mathewashing (recent AI ethics term): Using mathematical language to appear rigorous without substance

Measurement Approaches

Various metrics attempt to quantify information density:

• Flesch Reading Ease: Measures readability, not density
• Lexical Diversity: Vocabulary richness (type-token ratio)
• Propositional Density: Ideas per unit text (Kintsch, 1974)
• Compression Ratio: How much text can be compressed without losing meaning

Contemporary AI Research

Recent papers addressing information quality in LLMs:

• “Hallucination in Large Language Models” (Ji et al., 2023)
• “TruthfulQA: Measuring How Models Mimic Human Falsehoods” (Lin et al., 2022)
• “The Curse of Recursion: Training on Generated Data Makes Models Forget” (Shumailov et al., 2024)

---