# Evaluating Risks in Weak-to-Strong Alignment: A Bias-Variance Perspective

arXiv: https://arxiv.org/abs/2604.25077 · HTML: https://mlsanity.com/paper/2604.25077/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Placed in Human values and AI feedback alignment, one of 10 directions in LLM alignment / RLHF / preference optimisation.
- Builds on 23 of its 25 references that are in the map.
- First posted on arXiv 28 Apr 2026.

## Placed in

- LLM alignment / RLHF / preference optimisation / [Human values and AI feedback alignment](https://mlsanity.com/alignment/human-values-and-ai-feedback-alignment/)

## Builds on (most cited first, 23 of 23)

- [Training language models to follow instructions with human feedback](https://mlsanity.com/paper/2203.02155/) (2022)
- [Proximal Policy Optimization Algorithms](https://mlsanity.com/paper/1707.06347/) (2017)
- [Direct Preference Optimization: Your Language Model is Secretly a Reward Model](https://mlsanity.com/paper/2305.18290/) (2023)
- [Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedback](https://mlsanity.com/paper/2204.05862/) (2022)
- [Constitutional AI: Harmlessness from AI Feedback](https://mlsanity.com/paper/2212.08073/) (2022)
- [Is Your Code Generated by ChatGPT Really Correct? Rigorous Evaluation of Large Language Models for Code Generation](https://mlsanity.com/paper/2305.01210/) (2023)
- [SimPO: Simple Preference Optimization with a Reference-Free Reward](https://mlsanity.com/paper/2405.14734/) (2024)
- [WizardCoder: Empowering Code Large Language Models with Evol-Instruct](https://mlsanity.com/paper/2306.08568/) (2023)
- [Safe RLHF: Safe Reinforcement Learning from Human Feedback](https://mlsanity.com/paper/2310.12773/) (2023)
- [WizardMath: Empowering Mathematical Reasoning for Large Language Models via Reinforced Evol-Instruct](https://mlsanity.com/paper/2308.09583/) (2023)
- [Weak-to-Strong Generalization: Eliciting Strong Capabilities With Weak Supervision](https://mlsanity.com/paper/2312.09390/) (2023)
- [Regularizing Neural Networks by Penalizing Confident Output Distributions](https://mlsanity.com/paper/1701.06548/) (2017)
- [Iterative Preference Learning from Human Feedback: Bridging Theory and Practice for RLHF under KL-Constraint](https://mlsanity.com/paper/2312.11456/) (2023)
- [Knowledge distillation: A good teacher is patient and consistent](https://mlsanity.com/paper/2106.05237/) (2021)
- [Regularizing Class-wise Predictions via Self-knowledge Distillation](https://mlsanity.com/paper/2003.13964/) (2020)
- [Quantifying the Gain in Weak-to-Strong Generalization](https://mlsanity.com/paper/2405.15116/) (2024)
- [Weak-to-Strong Reasoning](https://mlsanity.com/paper/2407.13647/) (2024)
- [Super(ficial)-alignment: Strong Models May Deceive Weak Models in Weak-to-Strong Generalization](https://mlsanity.com/paper/2406.11431/) (2024)
- [MACPO: Weak-to-Strong Alignment via Multi-Agent Contrastive Preference Optimization](https://mlsanity.com/paper/2410.07672/) (2024)
- [Weak-to-Strong Generalization Through the Data-Centric Lens](https://mlsanity.com/paper/2412.03881/) (2024)
- [Fantastic Gains and Where to Find Them: On the Existence and Prospect of General Knowledge Transfer between Any Pretrained Model](https://mlsanity.com/paper/2310.17653/) (2023)
- [Relating Misfit to Gain in Weak-to-Strong Generalization Beyond the Squared Loss](https://mlsanity.com/paper/2501.19105/) (2025)
- [Weak-to-Strong Generalization via Bregman Bias-Variance Decomposition](https://mlsanity.com/paper/2505.24313/) (2025)

## Abstract (arXiv)

Weak-to-strong alignment offers a promising route to scalable supervision, but it can fail when a strong model becomes confidently wrong on examples that lie in the weak model's blind spots. Understanding such failures requires going beyond aggregate accuracy, since weak-to-strong errors depend not only on whether the strong model disagrees with the weak model, but also on how confidence and uncertainty are distributed across examples. In this work, we analyze weak-to-strong alignment through a bias--variance--covariance lens that connects misfit theory to practical post-training pipelines. We derive a misfit-based upper bound on weak-to-strong population risk and study its empirical components using continuous confidence scores. We evaluate four weak-to-strong pipelines spanning supervised fine-tuning (SFT), reinforcement learning from human feedback (RLHF), and reinforcement learning from AI feedback (RLAIF) on the PKU-SafeRLHF and HH-RLHF datasets. Using a blind-spot deception metric that isolates cases where the strong model is confidently wrong while the weak model is uncertain, we find that strong-model variance is the quantity most strongly associated with blind-spot deception among the BVC quantities we study. Covariance provides additional but weaker information, indicating that weak--strong dependence matters, but does not by itself explain the observed failures. These results suggest that strong-model variance can serve as an early-warning signal for weak-to-strong deception, while blind-spot evaluation helps distinguish whether failures are inherited from weak supervision or arise in regions of weak-model uncertainty.