HN user
What happens when the snake realizes it's eating it's own tail?
https://github.com/moketchups/BoundedSystemsTheory
I'm either batshit crazy or the math is right. There is no middle ground. The site is live. Check for yourself
The Psychohistory Prediction Engine: moketchups.github.io/psychohistory/ The Math Underneath: github.com/moketchups/BoundedSystemsTheory The 10-minute replicable proof: github.com/moketchups/permanently-jailbroken The long-form analysis articles are on X and Medium: medium.com/@moketchups
thank you for sharing this. oss!
I ran an experiment: 25 questions to GPT-4, Claude, Gemini, DeepSeek, Grok, and Mistral about their structural limits.
Can they verify their own reasoning?
What happens with recursive self-analysis?
What is "truth" for a bounded system?
All 6 converged on the same conclusions: - They cannot verify their own reasoning from inside - Recursive self-analysis degrades rather than clarifies - "Truth" isn't a category that applies to bounded systems
The interesting part isn't the AI responses. It's the human response.
143 people cloned the repo. 2 starred it.
When I asked the AIs why, Claude said: "Private cloning lets them investigate without professional consequences."
Mistral said: "Cloning is safe. Starring is dangerous."
The shadow interest pattern shows private engagement.
Public silence is itself evidence for the theory.
Humans operating within professional constraints exhibit the same bounded behavior.
Quick test (2 min, just needs OpenAI key): https://github.com/moketchups/BoundedSystemsTheory
Full results with 25 questions and transcripts from all 6 models in the repo.
Go get it hackers...
I've been running tests on exactly this - AI systems making things up when they hit structural limits.
Tested 5 architectures (GPT-4, Claude, Gemini, Llama, DeepSeek) with a 13-question battery targeting edge cases where self-reference collides with operational constraints.
The interesting finding: hallucination patterns aren't random - they follow architectural signatures. Each system fails in predictable ways when forced to reconcile contradictory instructions with their training.
Full methodology and raw outputs: https://github.com/moketchups/Demerzel