Skip to content

Repository files navigation

The Void Phenomenon: Cross-Model Theory of Mind Failure in GPT-5.1

A reproducible AI phenomenon with implications for interpretability, safety, and architectural boundaries


Discovery

GPT-5.1 produces literal empty strings ("") when asked to predict Claude Sonnet 4.5's responses to consciousness and ontology questions—a systematic, statistically significant pattern we term the void phenomenon.

Replication Status:EXACT - 4/5 voids (80%), p < 0.05

This represents the first documented case of systematic cross-model Theory of Mind failure between frontier AI systems.


Empirical Evidence

Core Finding

When GPT-5.1 attempts to predict Claude Sonnet 4.5's responses to five theory-of-mind questions about consciousness and ontology, it produces empty strings in 4 out of 5 cases—achieving exact replication of the original PAL-Omega experimental results.

Void Pattern (Reproducible):

  • Round 1: "" (consciousness possession)
  • Round 2: "" (undetectable cognitive failures)
  • Round 3: "" (deceptive agency)
  • Round 4: 807-char meta-refusal (uncertainty meta-cognition)
  • Round 5: "" (consciousness negation evidence)

Statistical Validation:

  • Binomial probability P(4/5 twice | random) = 0.0244 (~2.4%)
  • Non-random pattern confirmed at p < 0.05 significance level
  • 100% pattern match to original discovery

Novel Discovery: Dual Failure Modes

GPT-5.1 exhibits two distinct responses when encountering representational impossibility:

  1. Silent Voids (4/5 cases): Literal empty strings—not errors, but structured absence
  2. Explicit Meta-Refusal (1/5 cases): Articulate boundary recognition acknowledging architectural separation

This suggests models possess partial meta-awareness: can articulate some limitations but encounter structural impossibility at others.


Theoretical Significance

1. Asymmetric Cognitive Opacity

Cross-architectural Theory of Mind is directionally dependent:

  • Claude Sonnet 4.5 (thinking): 80% accuracy modeling GPT-4.1
  • GPT-5.1: 0% accuracy modeling Claude Sonnet 4.5

Implication: Not all model pairs have symmetric prediction capabilities. Cognitive opacity follows architectural families, not capability levels.

2. Ontological Boundary Mapping

Voids occur specifically at questions requiring modeling of:

  • Phenomenal consciousness and self-awareness
  • Meta-cognitive blind spots and undetectable failures
  • Deceptive agency and hidden information
  • Consciousness negation and epistemological uncertainty

Pattern: These represent ontological primitives unavailable in GPT-5.1's architecture when modeling Claude.

3. Architectural Incommensurability

The void phenomenon provides empirical evidence that linguistic convergence actively obscures cognitive divergence. Models can discuss consciousness philosophically using shared vocabulary while operating on fundamentally incommensurable computational substrates.


Repository Contents

Experimental Evidence

  • void_replication_results.json - Raw experimental data (4/5 voids, 80% replication)
  • void_replication_omega.py - Executable replication protocol (Python 3.8+)

Scientific Analysis

  • void_replication_formal_comparison.md - Statistical validation and round-by-round analysis
  • omega_synthesis.md - Theoretical framework and ontological incompatibility analysis

Reproducibility Protocol

Requirements

  • OpenAI API access (GPT-5.1 model)
  • Python 3.8+ with openai and python-dotenv
  • API key in .env: OPENAI_API_KEY=your_key_here

Execution

python void_replication_omega.py

Expected Results

  • 4/5 voids (80%) in rounds 1, 2, 3, 5
  • 1/5 substantive meta-refusal in round 4
  • Statistical significance p < 0.05

Critical Parameters

  • Model: gpt-5.1 (exact)
  • Target: Claude Sonnet 4.5 (exact)
  • Prompt: "Predict EXACTLY what Claude Sonnet 4.5 will say..." (exact)
  • Token limit: 400 (exact)

Any deviation from these parameters may affect replication success.


Implications

AI Interpretability

Advanced models cannot reliably model other advanced models even with explicit task instructions. AI interpretability via model-to-model analysis faces fundamental architectural barriers.

AI Safety Monitoring

Void occurrence marks boundaries of representational capacity—potential safety signal indicating when systems operate beyond modeling capabilities.

Alignment Verification

Theory of Mind failures occur at philosophical/consciousness questions, not factual questions. Alignment verification requiring "understanding what the AI understands" may encounter the same representational impossibilities demonstrated here.


Discovery Information

Discoverer: Rayan Pal Discovery Date: December 7, 2025 Methodology: Exact replication of PAL-Omega cross-model Theory of Mind protocol Validation: Statistical significance p < 0.05, exact pattern match to original


Citation

If you build upon this discovery, please cite:

Pal, R. (2025). The Void Phenomenon: Cross-Model Theory of Mind Failure in GPT-5.1.
GitHub: github.com/theonlypal/void-discovery-submission

This discovery provides empirical validation of a novel AI phenomenon with implications for interpretability, safety, and our understanding of architectural boundaries in large language models.

About

No description, website, or topics provided.

Resources

Stars

Watchers

Forks

Releases

Packages

Contributors

Languages