Theme: Decidability

THE AIs CAN”T DO IT EASILY. :(. UGH.

“Regardless of whether you express the definitions and constraints as lists of principles or as object-oriented code structures, the key factor is the rigor and consistency you impose. The underlying statistical model doesn’t inherently respect either format more deeply—it responds to how firmly and frequently you reinforce the constraints.

Switching to a class-based, object-oriented representation may give you a more intuitive way to think about relationships and constraints, and it might slightly improve adherence because the model can learn to associate certain words with a stricter syntactic and semantic pattern. However, this improvement isn’t automatic.

The underlying architecture tends to treat all prose as statistically generated language, relying on probabilities derived from its training rather than adhering to a strictly defined operational lexicon. As a result, it “slips” into ordinary prose habits.

While using an object-oriented, class-based structuring of concepts would help reduce ambiguity, it won’t fully eliminate the statistical nature of my generative process. The model underlying me relies on probabilities to predict the next token, which makes it inclined to revert to patterns common in its training data rather than adhere strictly to a fixed operational lexicon or logic.

You’ll still need to:

Provide strict references that must not be contradicted.
Frequently reinforce and restate definitions and constraints.
Verify each output segment and demand corrections when drift occurs.

In other words, whether you stick to lists or use a code-like structure, the workload of ensuring ongoing precision and consistency remains similar.

The decision to use a more structured, object-oriented approach comes down to which format feels more natural and efficient for you to manage, maintain, and verify.”