# High-confidence errors in language models

Status: Public preprint; related work accepted as an EIML 2026 workshop poster.

Research on stable high-confidence errors in language models and why local robustness need not imply correctness.

The workshop lists Stable Miscalibration in Large Language Models: A Practical View of High-Confidence Errors. The related preprint is titled False Fixed Points: Kantian Feedback, Stable Miscalibration, and Representational Compression in LLMs. Workshop acceptance is not ICML main-conference acceptance or evidence of attendance; the listing does not identify a paper version.

Canonical page: [High-confidence errors in language models](https://sites.google.com/view/eimlicml2026/accepted-papers_1)

Primary sources:

- [Official workshop listing](https://sites.google.com/view/eimlicml2026/accepted-papers_1)
- [Related preprint](https://arxiv.org/abs/2510.14925)

Reviewed: 2026-10-11

Source guide: [Toppy AI resources](https://www.toppymicros.com/ai/)