# Stable but Wrong: When Learning Stabilizes Away from the Truth

arXiv: https://arxiv.org/abs/2603.21491 · HTML: https://mlsanity.com/paper/2603.21491/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 15 of its 27 references that are in the map.
- First posted on arXiv 23 Mar 2026.

## Builds on (most cited first, 15 of 15)

- [Deep Residual Learning for Image Recognition](https://mlsanity.com/paper/1512.03385/) (2015)
- [Proximal Policy Optimization Algorithms](https://mlsanity.com/paper/1707.06347/) (2017)
- [LoRA: Low-Rank Adaptation of Large Language Models](https://mlsanity.com/paper/2106.09685/) (2021)
- [Training Verifiers to Solve Math Word Problems](https://mlsanity.com/paper/2110.14168/) (2021)
- [Qwen2.5 Technical Report](https://mlsanity.com/paper/2412.15115/) (2024)
- [Shortcut Learning in Deep Neural Networks](https://mlsanity.com/paper/2004.07780/) (2020)
- [A Survey on Bias and Fairness in Machine Learning](https://mlsanity.com/paper/1908.09635/) (2019)
- [Optimization Methods for Large-Scale Machine Learning](https://mlsanity.com/paper/1606.04838/) (2016)
- [Unmasking Clever Hans Predictors and Assessing What Machines Really Learn](https://mlsanity.com/paper/1902.10178/) (2019)
- [PadChest: A large chest x-ray image dataset with multi-label annotated reports](https://mlsanity.com/paper/1901.07441/) (2019)
- [Emergent Misalignment: Narrow finetuning can produce broadly misaligned LLMs](https://mlsanity.com/paper/2502.17424/) (2025)
- [TorchXRayVision: A library of chest X-ray datasets and models](https://mlsanity.com/paper/2111.00595/) (2021)
- [Performative Prediction: Past and Future](https://mlsanity.com/paper/2310.16608/) (2023)
- [Detecting Shortcut Learning for Fair Medical AI using Shortcut Testing](https://mlsanity.com/paper/2207.10384/) (2022)
- [The Clever Hans Effect in Unsupervised Learning](https://mlsanity.com/paper/2408.08041/) (2024)

## Abstract (arXiv)

Stable training is often treated as evidence that learning is succeeding, but stability characterizes optimization behavior rather than correctness relative to an external objective. We study what happens when the signal being optimized remains persistently biased. We define Stable but Wrong (SBW) as a learning state in which the learning process remains stable under a task-appropriate operational criterion while the learned outcome remains systematically displaced from an independently defined objective. A minimal strongly convex model shows that a persistent bias in the update direction can shift the unique convergence point away from the true optimum. Controlled experiments in reinforcement learning, supervised learning, and continual fine-tuning of a large language model reveal a recurring separation between apparently normal optimization and correctness under static and feedback-coupled biases. Recovery-stage clean-data access and exploration interventions further show that subsequent trajectories can remain modifiable, although the interventions differ in protocol and do not imply a shared mechanism. The results expose a basic limit of optimization stability as a reliability signal in persistent and feedback-coupled learning systems.