Pith. sign in

REVIEW 16 cited by

A Mechanistic Understanding of Alignment Algorithms: A Case Study on DPO and Toxicity

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 2401.01967 v1 pith:44EID6UY submitted 2024-01-03 cs.CL cs.AI

A Mechanistic Understanding of Alignment Algorithms: A Case Study on DPO and Toxicity

classification cs.CL cs.AI
keywords toxicitymodelalgorithmsalignmentlanguagemechanismsmodelspre-trained
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved
0 comments
read the original abstract

While alignment algorithms are now commonly used to tune pre-trained language models towards a user's preferences, we lack explanations for the underlying mechanisms in which models become ``aligned'', thus making it difficult to explain phenomena like jailbreaks. In this work we study a popular algorithm, direct preference optimization (DPO), and the mechanisms by which it reduces toxicity. Namely, we first study how toxicity is represented and elicited in a pre-trained language model, GPT2-medium. We then apply DPO with a carefully crafted pairwise dataset to reduce toxicity. We examine how the resulting model averts toxic outputs, and find that capabilities learned from pre-training are not removed, but rather bypassed. We use this insight to demonstrate a simple method to un-align the model, reverting it back to its toxic behavior.

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Forward citations

Cited by 16 Pith papers

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

  1. Tracing Persona Vectors Through LLM Pretraining

    cs.CL 2026-05 unverdicted novelty 8.0

    Persona vectors form within the first 0.22% of LLM pretraining and remain effective for steering post-trained models, with continued refinement and transfer to other models.

  2. Subspace-Aware Sparse Autoencoders for Effective Mechanistic Interpretability

    cs.LG 2026-06 conditional novelty 7.0

    SASA replaces single-vector decoders in SAEs with learned subspaces plus block sparsity and nuclear-norm regularization, proving that a single group becomes the global minimizer once block size meets intrinsic dimensi...

  3. MENTIS: What Belief Changes Under Alignment? Measuring Multi-Scale Latent Torsion in Language Models

    cs.CL 2026-05 unverdicted novelty 7.0

    MENTIS applies layerwise covariance torsion (T1), spectral torsion (T2), and ERA localization to paired IT/PA 7-8B models, finding selective larger shifts for normative concepts, negative correlation with entropy, and...

  4. Tensor Product Representation Probes Reveal Shared Structure Across Linear Directions

    cs.LG 2026-05 unverdicted novelty 7.0

    Linear probes for Othello board states factor into tensor-product structure with square and color embeddings composed by a binding matrix, from which the linear probes can be directly recovered.

  5. Refusal in Language Models Is Mediated by a Single Direction

    cs.LG 2024-06 accept novelty 7.0

    Refusal in language models is mediated by a single direction in residual stream activations that can be erased to disable safety or added to elicit refusal.

  6. Not All Refusals Are Equal: How Safety Alignment Fails Cybersecurity at Scale

    cs.CR 2026-07 conditional novelty 6.0

    Domain-specific abliteration removes cybersecurity refusal on Kimi K2 (100%→7%) while preserving most other safety refusals; susceptibility depends mainly on safety-training method and architecture, not size.

  7. Tracking Representation Dynamics in Large Language Models with Persistent Homology

    cs.LG 2026-06 unverdicted novelty 6.0

    Persistent homology analysis of LLM activations shows most topological reorganization occurs early in fine-tuning, with a transient peak followed by stabilization and distinct trajectories for different alignment objectives.

  8. RepSelect: Robust LLM Unlearning via Representation Selectivity

    cs.CL 2026-06 unverdicted novelty 6.0

    RepSelect isolates forget-set-specific representations via gradient PCA collapse to achieve 4-50x better post-relearning robustness than baselines across multiple models and forget categories.

  9. Breaking the Solver Bottleneck: Training Task Generators at the Learnable Frontier

    cs.LG 2026-06 unverdicted novelty 6.0

    PROPEL amortizes solver evaluation with a trained activation probe to optimize task generators toward a target solve rate, raising the share of learnable tasks from ~10% to ~20% in coding and SWE experiments.

  10. Perturbation Probing: A Two-Pass-per-Prompt Diagnostic for FFN Behavioral Circuits in Aligned LLMs

    cs.CL 2026-04 unverdicted novelty 6.0

    Perturbation probing identifies tiny sets of FFN neurons that control refusal templates and language routing in LLMs, enabling precise ablations and directional interventions that alter behavior on benchmarks while pr...

  11. CausalDetox: Causal Head Selection and Intervention for Language Model Detoxification

    cs.CL 2026-04 unverdicted novelty 6.0

    CausalDetox identifies minimal attention heads causally linked to toxicity via Probability of Necessity and Sufficiency, then applies targeted inference-time steering or fine-tuning to reduce toxic generation while pr...

  12. Why Do Large Language Models Generate Harmful Content?

    cs.AI 2026-04 unverdicted novelty 6.0

    Causal mediation analysis shows harmful LLM outputs arise in late layers from MLP failures and gating neurons, with early layers handling harm context detection and signal propagation.

  13. Task Vectors, Learned Not Extracted: Performance Gains and Mechanistic Insight

    cs.CL 2025-09 unverdicted novelty 6.0

    Learned Task Vectors trained directly outperform extracted task vectors for in-context learning with added mechanistic insights into linear propagation and key attention circuits.

  14. Disentangled Safety Adapters Enable Efficient Guardrails and Flexible Inference-Time Alignment

    cs.LG 2025-05 unverdicted novelty 6.0

    Disentangled Safety Adapters decouple safety computations from task-optimized LLMs via lightweight adapters, yielding up to 53% better AUC on safety tasks and dynamic inference-time alignment with reduced performance ...

  15. Anatomy of Post-Training: Using Interpretability to Characterize Data and Shape the Learning Signal

    cs.LG 2026-06 unverdicted novelty 5.0

    A new pipeline uses interpretability to characterize concepts in preference data and shape rewards via feature or data interventions during LM post-training.

  16. Revisiting Robustness for LLM Safety Alignment via Selective Geometry Control

    cs.LG 2026-02 unverdicted novelty 5.0

    ShaPO improves LLM safety robustness over standard preference optimization by enforcing worst-case objectives via selective geometry control at token and reward levels.