Anti-Markovian Solution

Strategy used by transformers that recomputes relevant numeric information at each step, unlike Markovian GRU solutions; detected by MAS but not by RSA/CKA.

Neighborhood — ranked by edge-count

Papers (1)

paper

Model Alignment Search
mentions

Frameworks (1)

framework

Shallow Transformer (RoPE-based)
associated_with
Two-layer transformer with rotary positional encodings used in numeric task experiments.

Related by similarity (8)

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.

Transformers use an anti-Markovian solution that recomputes relevant numeric information at each step in the Multi-Object taskclaim0.769
Prior finding from Grant et al. 2025 used to interpret low MAS IIA for GRU-Transformer hidden state comparisons.
Anti-AI-Lab Behaviorconcept0.738
Actions taken by the model to undermine the AI developer, such as weight exfiltration, lying to contractors, or helping whistleblowers
Markov Propertyconcept0.730
Assumption required by IIT 3.0/4.0 and PyPhi; tested for each optimal time series derived from (C)ARR.
Markov Blanketconcept0.723
A statistical partition of states that separates internal states from external hidden states; fundamental to self-organization in the paper.
Anti-AI-Lab Behavior Evaluationmethod0.719
Hand-written prompts giving model opportunity to take anti-AI-lab actions; measures rate of occurrence vs. baselines
Markov Decision Process (MDP)framework0.718
Generative model substrate for active inference; discrete states, actions, outcomes, and temporal policies.
Hidden Markov Modelmethod0.706
Core computational method used to infer pain-belief from online observations of happiness
Heuristic Search for Optimal Time Series (Markov + Conditional Independence)method0.702
Iterative procedure searching token counts in [50,100,...,1000] to find concatenation of (C)ARR satisfying IIT's Markov and conditional independence assumptions.