Do LLMs have core beliefs?
cs.LG
Submitted: 2026-05-05
Updated: 2026-10-14
License: http://creativecommons.org/licenses/by/4.0/
The gist: The rise of Large Language Models (LLMs) has sparked debate about whether these systems exhibit human-level cognition.
Terminology
Abstract
The rise of Large Language Models (LLMs) has sparked debate about whether these systems exhibit human-level cognition. In this debate, little attention has been paid to a structural component of human cognition: core beliefs, truths that provide a foundation around which we can build a worldview. These commitments usually resist debunking, as abandoning them would represent a fundamental shift in how we see reality. In this paper, we ask whether LLMs hold anything akin to core commitments. Using a probing framework we call Adversarial Dialogue Trees (ADTs) over five domains (science, history, geography, biology, and mathematics), we find that most LLMs fail to maintain a stable worldview. Though some recent models showed improved stability, they still eventually failed to maintain key commitments under conversational pressure. These results document an improvement in argumentative skills across model generations but indicate that all current models lack a key component of human-level cognition.
Sources
- International AI Safety Report
- Machine Psychology
- Epistemology of Language Models: Do Language Models Have Holistic Knowledge?
- Are Large Language Models Consistent over Value-laden Questions?
- Towards Understanding Sycophancy in Language Models
- The Illusion of Thinking: Understanding the Strengths and Limitations of Reasoning Models via the Lens of Problem Complexity
- Can ChatGPT Defend its Belief in Truth? Evaluating LLM Reasoning via Debate
- Belief Revision: The Adaptability of Large Language Models Reasoning
- Revealing emergent human-like conceptual representations from language prediction
- Knowledge Conflicts for LLMs: A Survey
- Language Models Represent Beliefs of Self and Others
Related papers
- Polynomial-Augmented Neural Networks (PANNs) with Weak Orthogonality Constraints for Enhanced Function and PDE Approximation
- AIRL-S: Unifying Reinforcement Learning and Search-Based Test-Time Scaling via Adversarial Inverse Reinforcement Learning
- Transformers as Bayesian In-Context Experimenters: Smoothness-Adaptive Efficient ATE Estimation
- Convergence issues in Relational Concept Analysis based on AOC-posets
- Beliefs Beyond Posteriors: The Predictive Cost of ELBO Binding
- Understanding Diffusion Models via Ratio-Based Function Approximation with SignReLU Networks