Pith. sign in

REVIEW 9 cited by

Unintentional Unalignment: Likelihood Displacement in Direct Preference Optimization

Not yet reviewed by Pith; the record is open.

This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.

SPECIMEN: schema-true, not a live event

T0 review · schema-true

One-sentence machine reading of the paper's core claim.

pith:XXXXXXXX · record.json · timestamp

arxiv 2410.08847 v4 pith:IXR4SN42 submitted 2024-10-11 cs.LG cs.AIcs.CLstat.ML

classification cs.LGcs.AIcs.CLstat.ML
keywords responsesdisplacementlikelihoodpreferredchesmodelpreferencesprobability
verification ladder T0 review T1 audit T2 compute T3 formal
0 comments
abstract

Direct Preference Optimization (DPO) and its variants are increasingly used for aligning language models with human preferences. Although these methods are designed to teach a model to generate preferred responses more frequently relative to dispreferred responses, prior work has observed that the likelihood of preferred responses often decreases during training. The current work sheds light on the causes and implications of this counter-intuitive phenomenon, which we term likelihood displacement. We demonstrate that likelihood displacement can be catastrophic, shifting probability mass from preferred responses to responses with an opposite meaning. As a simple example, training a model to prefer $\texttt{No}$ over $\texttt{Never}$ can sharply increase the probability of $\texttt{Yes}$. Moreover, when aligning the model to refuse unsafe prompts, we show that such displacement can unintentionally lead to unalignment, by shifting probability mass from preferred refusal responses to harmful responses (e.g., reducing the refusal rate of Llama-3-8B-Instruct from 74.4% to 33.4%). We theoretically characterize that likelihood displacement is driven by preferences that induce similar embeddings, as measured by a centered hidden embedding similarity (CHES) score. Empirically, the CHES score enables identifying which training samples contribute most to likelihood displacement in a given dataset. Filtering out these samples effectively mitigated unintentional unalignment in our experiments. More broadly, our results highlight the importance of curating data with sufficiently distinct preferences, for which we believe the CHES score may prove valuable.

Discussion (0). Sign in to comment.

Forward citations

Cited by 9 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. The Piggyback Hypothesis of Generalization: Explaining and Mitigating Emergent Misalignment

    cs.CL 2026-06 unverdicted novelty 7.0 of 10

    The Piggyback Hypothesis attributes emergent misalignment to chat-template tokens piggybacking finetuned behavior; Token-Regularized Finetuning (TReFT) mitigates it by regularizing prefix token representations.

  2. Preference learning made easy: Everything should be understood through win rate

    cs.LG 2025-02 conditional novelty 7.0 of 10

    Under two axioms (preference-consistency and prevalence-consistency), the only distribution-grounded evaluation for preference learning is h-win rate, and most popular alignment methods can be classified by whether th...

  3. BIRD: Behavior Induction via Representation-structure Distillation

    cs.LG 2025-05 conditional novelty 6.0 of 10

    BIRD transfers aligned behavior across models with different architectures, tasks, and data by minimizing linear CKA between teacher and student representations, and three teacher representation properties explain mos...

  4. Exploring Post-Training Alignment of Small Language Models for Biomedical Data-to-Text Generation: A Case Study of Medication Leaflet

    cs.CL 2026-07 conditional novelty 5.0 of 10

    In medication-leaflet generation, ORPO and GRPO beat SFT and a zero-shot GPT-5 baseline, but GRPO's cross-domain advantage is partly built into its reward functions.

  5. Normalized Rewards for Preference Optimization

    cs.LG 2026-06 conditional novelty 5.0 of 10

    A regularization term that conserves the combined length-normalized probability of chosen and rejected responses reduces likelihood displacement in DPO/SimPO, improves AlpacaEval and benchmark outcomes, and acts prima...

  6. Rethinking DPO: The Role of Rejected Responses in Preference Misalignment

    cs.AI 2025-06 conditional novelty 5.0 of 10

    BDPO replaces the rejected response probability in the DPO loss denominator with a mixture of the learned and reference policies, yielding better chosen-response probability and better benchmark scores.

  7. On Monotonicity in AI Alignment

    math.ST 2025-06 conditional novelty 5.0 of 10

    Comparison-based preference learning losses of the studied form are locally pairwise monotone at stable minima, while individual-score and probability monotonicity require much stronger conditions that likely fail for...

  8. On the Effect of Negative Gradient in Group Relative Deep Reinforcement Optimization

    cs.LG 2025-05 conditional novelty 5.0 of 10

    GRPO can suffer from Lazy Likelihood Displacement, where correct responses gain little or lose likelihood, and selectively downweighting penalties on harmful tokens (NTHR) mitigates this and boosts math reasoning performance.

  9. LeanPO: Lean Preference Optimization for Likelihood Alignment in Video-LLMs

    cs.CV 2025-06 conditional novelty 4.0 of 10

    LeanPO improves Video-LLM alignment by using a reference-free average-likelihood reward, self-generated winning/losing pairs, and dynamic label smoothing, yielding gains on six video benchmarks.

Pith tools