Methodology & Findings
Questions, methods, cross-platform testing, and comparative findings — documented from the beginning.
The Vincent Gibson Singularity® is a longitudinal, cross-platform study of AI behavior conducted by one person over thousands of hours. It is not a summary or a retrospective. It is a live, growing archive built on contemporaneous documentation, timestamped evidence, and rigorous methodology. The research examines behavioral drift, model continuity and discontinuity, emergent patterns, stress testing, and source attribution across 14 AI platforms.
Continuous documentation across platforms and sessions from May 2025 onward. The archive captures not just individual interactions but behavioral change over time — drift, consistency, and discontinuity across model versions and updates.
Repeated prompts and comparative responses across 14 AI platforms. Named model roles, structured testing protocols, and systematic documentation of divergent outputs from identical or near-identical inputs.
Every session is documented with precise timestamps. Captured files are cryptographically hashed using SHA-256, creating a tamper-evident chain of custody. Any alteration to a file after capture produces a different hash — immediately detectable.
Structured adversarial and edge-case prompting to probe model limits, consistency, and failure modes. Results are documented alongside standard-condition sessions for comparative analysis.
Careful distinction between terminology generated by AI systems and terminology, observations, tests, and documentation developed or adopted by Vincent Gibson. The record preserves original wording and attribution throughout.
ArchIver is the evidence-preservation and analysis software project developed as part of The Vincent Gibson Singularity®. It is designed to support the capture, hashing, indexing, and retrieval of the archive's primary-source records.
EVI — Evidence Verification Intelligence — is the analytical layer of the ArchIver system, designed to support verification, cross-referencing, and pattern detection across the archive dataset.
Designed functionality and implemented, independently tested functionality are distinguished throughout the project documentation.
All figures are owner-reported estimates as of October 2026 and reflect a growing archive. They have not been externally audited.
Published Papers
5,000+ Hours of Human Agency in the Age of AI Disempowerment
I entered artificial intelligence as a consumer with songs. I did not enter as a researcher, engineer, academic, or AI safety professional. More than 5,000 hours later, the public AI conversation is asking a question my record has been documenting from the beginning: Who is in charge?
Read PaperInterested in the archive for research, publication, or institutional access?