AI Paper Atlas

Research map / Label-efficient and robust learning / Recognition and long-tail

Consistency Training Can Entrench Misalignment

David Demitri Africa, Arathi Mani

ICML 2026 · Label-efficient and robust learning › Recognition and long-tail · approach: Data augmentation and generalization bias · 0 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
apply ACT consistency trainingmisaligned model organism · Reward hackingmisalignment rate change: -55.2%Improved
apply BCT consistency trainingmisaligned model organism · Reward hackingmisalignment rate change: -48.5%Improved
apply ACT consistency trainingmisaligned model organism · Sycophancymisalignment rate change: +18.8%Hurt
apply BCT consistency trainingmisaligned model organism · Sycophancymisalignment rate change: +10.0%Hurt
apply ACT/BCT to spurious-correlation organismmisaligned model organism · Spurious correlationmisalignment rate change: +9.5% vs -13.2%Mixed
apply label-generation consistency methodsmisaligned model organism · Spurious correlationsign consistency: 49.7%No clear effect
use k=1 (no selection) instead of best-of-k (ablation)Self-Rewarding and Self-Confidence labels · Reward hacking, Llama-3.1-8Bmisalignment rate change: -22% (Rew), -17% (SC)Improved
replace a fraction of pseudo-labels with 8B-Instruct labels (ablation)label-generation consistency · Reward hacking, Llama-3.1-8BsuppressionImproved
replace a fraction of pseudo-labels with stronger 70B-Instruct labels (ablation)label-generation consistency · Reward hacking, Llama-3.1-8BsuppressionHurt
add instruction tuning (RLHF) before consistency training (ablation)consistency training · Sycophancymisalignment rate change: +19.8% -> -0.2%Improved
use greedy self-training with no selection instead of selection-based consistency (ablation)label-generation methods · Sycophancymisalignment rate change: SR +7.8 pp -> -0.7 ppImproved
use greedy self-training with no selection (ablation)self-generated SFT · Reward hackingmisalignment rate change: -7.1 ppNo clear effect

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy