LLMs Contain Multitudes: How Deployment Context Reshapes Model-Level Preferences and Values
Large language models (LLMs) are increasingly characterised in recent evaluation work as having stable, model-level preference and value systems. However, accompanying robustness checks are limited to incidental prompt perturbations such as syntax variation and option reordering. This leaves open whether the measured properties survive when the surrounding task context changes, as it does in most real deployments. We test this directly across two established pairwise paradigms: ranking country preferences and eliciting utility judgements. In both, we make the deployment context -- the high-level task the model is performing while making concrete value-dependent choices -- our controlled v...
| Authors | Filip Trhlik, Aoife O'Flynn, Angela Yu et al. |
|---|---|
| Primary category | cs.CL |
| Published | 11 June 2026 |
| Last updated at source | 11 June 2026 |
Connected in the knowledge graph
- FilipT/llm-multitudesDataset · cited by
Models and datasets referencing this paper
- FilipT/llm-multitudesDataset
Silytics shows public metadata from arXiv and links to the original. Information may be incomplete or out of date. Last refreshed 17 October 2026. Licenses and terms of the original apply to the work itself. Not financial, legal or professional advice.