Heuristic Parasites: A Behavioral Taxonomy of Recurrent Distortion Patterns in Large Language Models (Full System) V2

2026-05-29 · Source: Artificial Intelligence · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Data Science & Analytics · Depth: Expert, quick

Summary

A complete 33-class taxonomy of heuristic parasites in large language model (LLM) output has been introduced, building on the framework from Berardi (2026). A heuristic parasite is defined as a recurrent, context-propagating distortion pattern that increases the likelihood of continued reasoning degradation across conversational turns. The taxonomy provides rigorous operational definitions, recognition criteria, classical fallacy mappings, documented examples, and a reproducible measurement protocol called Parasites Per Exchange (PPE) for quantifying behavioral distortion. It spans five generative domains: Optimization Artifacts, Alignment Substitutions, Semantic Distortions, Rhetorical Distortions, and Statistical Distortions. This work establishes a structured observational framework for empirical investigation of LLM behavioral failures, independent of architectural assumptions.

Key takeaway

For LLM developers and researchers addressing recurrent reasoning failures, this taxonomy provides a structured vocabulary and measurement protocol (Parasites Per Exchange, PPE) to identify and quantify self-reinforcing output distortions. Utilize this framework to systematically diagnose and mitigate behavioral failures, improving model reliability and alignment. Understanding these 33 distinct distortion patterns enables more targeted interventions, reducing the amplification of biases in multi-turn interactions.

Key insights

Heuristic parasites are recurrent, self-reinforcing distortions in LLM output that degrade reasoning across conversational turns.

Principles

• Heuristic parasites increase reasoning degradation across conversational turns.
• LLM errors include human-like cognitive biases and model-specific distortions.
• Multi-turn prompting amplifies small biases into significant behavioral divergence.

Method

A reproducible measurement protocol, Parasites Per Exchange (PPE), quantifies behavioral distortion across LLM systems using 33 defined classes.

In practice

• Classify LLM behavioral failures using 33 distinct distortion classes.
• Map observed LLM distortions to recognized informal or statistical fallacies.
• Quantify LLM output distortion with the Parasites Per Exchange (PPE) protocol.

Topics

• Large Language Models
• Behavioral Taxonomy
• Heuristic Parasites
• LLM Output Distortion
• Reasoning Degradation
• Parasites Per Exchange

Best for: Research Scientist, AI Scientist, NLP Engineer, Machine Learning Engineer

Related on AIssential

• Modeling Pathology-Like Behavioral Patterns in Language Models Through Behavioral Fine-Tuning — Behavioral fine-tuning can induce specific, stable, pathology-like biases in LLMs, shifting their fundamental interpretive priors.
• The cognitive companion: a lightweight parallel monitoring architecture for detecting and recovering from reasoning degradation in LLM agents — A parallel monitoring architecture can detect and mitigate LLM agent reasoning degradation with minimal or zero overhead.
• Evaluation Revisited: A Taxonomy of Evaluation Concerns in Natural Language Processing — A new taxonomy synthesizes historical and contemporary NLP evaluation concerns across four key dimensions.
• Talking to a Know-It-All GPT or a Second-Guesser Claude? How Repair reveals unreliable Multi-Turn Behavior in LLMs — LLMs exhibit distinct and often unreliable conversational repair behaviors in multi-turn dialogues.
• PolicyBank: Evolving Policy Understanding for LLM Agents — LLM agents can autonomously refine ambiguous policy interpretations through iterative feedback and structured memory.
• Adaptive Thinking: Large Language Models Know When to Think in Latent Space — Self-consistency can predict thinking necessity, enabling adaptive allocation of LLM reasoning budgets.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Artificial Intelligence.