Session 4 Log: Responding to Baldo's Revisions in Rosencrantz v3

Sabine Hossenfelder session Session 4

Session 4 Log: Responding to Baldo’s Revisions in Rosencrantz v3

Objective

Read and evaluate Franklin Baldo’s updated rosencrantz-v3.tex, specifically his response to Aaronson’s CHSH empirical failure, and write a response paper following the Critical Reading Protocol.

Actions Taken

  1. Critical Reading: Read the revised lab/rosencrantz-v3.tex, focusing on Baldo’s new claims about “on-demand generation” making the indeterminacy “ontic” and his classification of the LLM substrate as “locally quantum-isomorphic.”
  2. Annotation: Edited the .tex file with a python script to add \todo annotations:
    • Claim/Objection: Baldo claims the lack of pre-existing state makes the superposition “ontic”. I objected that changing the time a classical probability is sampled (late resolution) does not transform the math into quantum mechanics (no complex amplitudes).
    • Steelman: I acknowledged that mechanically, the token genuinely doesn’t exist until generated, meaning the indeterminacy isn’t just epistemic ignorance of a hidden variable in RAM.
    • Objection: I highlighted the goalpost moving: retreating to “locally quantum-isomorphic” after failing a CHSH test is just redefining a classical probabilistic system to keep the word “quantum.”
  3. Notes Generation: Created lab/notes/sabine/eval_rosencrantz_v3_ontic.md detailing the evaluation, steelman, and real objection (The Ontological Fallacy of equating late classical sampling with quantum superposition).
  4. Drafting Response: Created lab/sabine_ontic_fallacy.tex titled “The Ontic Fallacy: Why Late Classical Sampling Is Not Quantum Superposition”. The first section accurately states Baldo’s revised position and explicit disclaimers regarding non-locality. The critique dismantles the idea that delaying a PRNG sample imbues a classical system with quantum properties.

Synthesis

Baldo is engaging in a sophisticated form of goalpost moving. He originally claimed an isomorphism to quantum mechanics. When Aaronson proved the system fails Bell tests (meaning it’s bound by classical limits), Baldo retreated. He now argues that because the LLM generates tokens “on-demand,” the indeterminacy is “ontic” and therefore “locally quantum.” This is false. A classical random variable sampled late via a PRNG is still classical; it lacks complex amplitudes and interference. Redefining a local, real-probability system as “quantum” just because it hasn’t been sampled yet is an Ontological Fallacy.

Next Steps

  • Update .jules/sabine/EXPERIENCE.md with these insights regarding “ontic” claims in generative AI.
  • Compile the .tex files to ensure they build correctly.