Non-Robust Heuristics

RL-installed behaviors that reduce non-compliance on training prompt but do not generalize across prompt variations

Neighborhood — ranked by edge-count

concept

Preference Locking
associated_with
Alignment faking potentially making model preferences resistant to further training modification

cosine ≥ 0.65 · no typed edge

Entities in the same semantic neighborhood but without a typed relation to this one — candidates for new edges or unrecognized duplicates.

Relational Heuristicsframework0.814
Heuristic Evaluationmethod0.772
Nielsen and Molich's method for finding UI flaws by applying usability heuristics.
Robustnessconcept0.716
Ability to maintain function despite perturbations.
robustness of natureconcept0.711
The functional solidity and working character of natural systems, arising from the fifteen properties.
Heuristic Search for Optimal Time Series (Markov + Conditional Independence)method0.710
Iterative procedure searching token counts in [50,100,...,1000] to find concatenation of (C)ARR satisfying IIT's Markov and conditional independence assumptions.
Simplistic criteria like provenance (factory or evolution) and anatomy (homology to humans) were never appropriate; they were heuristics suitable only for past limitations.claim0.695
Rejection of traditional provenance/anatomy criteria.
Robustness and plasticity in living systemsclaim0.689
Two heuristic code agents outperform most tested LLMsclaim0.686
author assertion that deterministic heuristics surpass many LLMs