Hallucination Stations: On Some Basic Limitations of Transformer-Based Language Models

TL;DR

This paper investigates the fundamental limitations of transformer-based language models, revealing that beyond certain complexity levels, they cannot perform or verify complex computational or agentic tasks, highlighting inherent capabilities constraints.

Contribution

The paper introduces a complexity-theoretic framework to analyze hallucinations and limitations in LLMs, providing fundamental insights into their capabilities and boundaries.

Findings

LLMs cannot perform tasks beyond certain complexity thresholds.

Verification of their outputs becomes infeasible at higher complexities.

Hallucinations are linked to these fundamental computational limitations.

Abstract

In this paper we explore hallucinations and related capability limitations in LLMs and LLM-based agents from the perspective of computational complexity. We show that beyond a certain complexity, LLMs are incapable of carrying out computational and agentic tasks or verifying their accuracy.