NPO enables stable unlearning of 50%+ training data in LLMs on TOFU by making collapse exponentially slower than gradient ascent, preserving sensible outputs where prior methods fail.
super hub Canonical reference
KTO: Model Alignment as Prospect Theoretic Optimization
Canonical reference. 75% of citing Pith papers cite this work as background.
100
Pith papers citing it
Background
75% of classified citations
abstract
Kahneman & Tversky's $\textit{prospect theory}$ tells us that humans perceive random variables in a biased but well-defined manner (1992); for example, humans are famously loss-averse. We show that objectives for aligning LLMs with human feedback implicitly incorporate many of these biases -- the success of these objectives (e.g., DPO) over cross-entropy minimization can partly be ascribed to them belonging to a family of loss functions that we call $\textit{human-aware losses}$ (HALOs). However, the utility functions these methods attribute to humans still differ from those in the prospect theory literature. Using a Kahneman-Tversky model of human utility, we propose a HALO that directly maximizes the utility of generations instead of maximizing the log-likelihood of preferences, as current methods do. We call this approach KTO, and it matches or exceeds the performance of preference-based methods at scales from 1B to 30B, despite only learning from a binary signal of whether an output is desirable. More broadly, our work suggests that there is no one HALO that is universally superior; the best loss depends on the inductive biases most appropriate for a given setting, an oft-overlooked consideration.
hub tools
citation-role summary
background 14
method 2
citation-polarity summary
claims ledger
- abstract Kahneman & Tversky's $\textit{prospect theory}$ tells us that humans perceive random variables in a biased but well-defined manner (1992); for example, humans are famously loss-averse. We show that objectives for aligning LLMs with human feedback implicitly incorporate many of these biases -- the success of these objectives (e.g., DPO) over cross-entropy minimization can partly be ascribed to them belonging to a family of loss functions that we call $\textit{human-aware losses}$ (HALOs). However, the utility functions these methods attribute to humans still differ from those in the prospect th
- method ate a base set of N responses, denoted as Dbase = {(x, ri, ai)}N i=1, where ri is the textual response anda i ∈ Ais the corresponding attribute. To embed the target distributionP∗ into the train- ing data, we explicitly control the generation fre- quency such that the count Nk of responses exhibit- ing attributea k satisfies: Nk =round(N·P ∗(ak|x))(6) For instance, given a target distribution of {Male: 0.99, Female: 0.01} and N= 100 , Dbase will contain 99 responses with the Male attribute and 1
- method policy log-probability ratios against pairwise preference data relative to a fixed reference model. This reformulation reduces alignment to a stable classification-style objective while retaining strong em- pirical performance. As a result, DPO has inspired a growing family of reference-based, reward-free alignment methods, including IPO [11], KTO [12], SimPO [13], ORPO [14], and iterative or online variants such as SPIN [15]. Preprint. arXiv:2605.08037v1 [cs.LG] 8 May 2026 The pairwise and list
- background non-linear optimization problems involving phys- ical dynamics. We follow a scalable backtrans- lation based synthetic data generation strategy described in Section 3.2. 2.3. RL for Reasoning and Code Generation Group Relative Policy Optimization (GRPO) [31] eliminates the critic model from PPO [32] by sampling groups of outputs and normalizing ad- vantages within each group; DeepSeek-R1 [33] showed that complex reasoning strategies emerge from GRPO with verifiable rewards alone, and Dr. GRPO [3
- background Orpo: Monolithic preference optimization without reference model. InProceedings of the 2024 Conference on Empirical Methods in Natural Language Processing, pages 11170-11189, 2024. [63] Yu Meng, Mengzhou Xia, and Danqi Chen. Simpo: Simple preference optimization with a reference-free reward.Advances in Neural Information Processing Systems, 37:124198- 124235, 2024. [64] Kawin Ethayarajh, Winnie Xu, Niklas Muennighoff, Dan Jurafsky, and Douwe Kiela. Kto: Model alignment as prospect theoretic opti
- background [172] Chongyu Fan, Yihua Zhang, Jinghan Jia, Alfred Hero, and Sijia Liu. Cyclicreflex: Im- proving large reasoning models via cyclical reflection token scheduling. arXiv preprint arXiv:2506.11077, 2025. [173] Siqi Fan, Peng Han, Shuo Shang, Yequan Wang, and Aixin Sun. Cothink: Token-efficient reasoning via instruct models guiding reasoning models. arXiv preprint arXiv:2505.22017, 2025. [174] Tiantian Fan, Lingjun Liu, Yu Yue, Jiaze Chen, Chengyi Wang, Qiying Yu, Chi Zhang, Zhiqi Lin, Ruofei Zhu,
- background [59] proposed a two-stage strat- egy combining SFT and Feasibility-and-Optimality-Aware Reinforcement Learning (FOARL) to guide LLMs and improve solution quality. 3.2.2 Reinforcement Learning RL strategies are introduced to enhance model robustness. To address hallucina- tion issues in LLMs, Jiang et al. [60] incorporated Kahneman-Tversky Optimization (KTO) [61] along with self-correction mechanisms, and proposed LLMOPT, which has been validated across six real-world datasets spanning 20 domains
co-cited works
representative citing papers
ORPO performs preference alignment during supervised fine-tuning via a monolithic odds ratio penalty, allowing 7B models to outperform larger state-of-the-art models on alignment benchmarks.
Trajectories from a Bittensor ShoppingBench subnet arena, filtered to retain only agentic tool-calling behavior, enable SFT+GRPO post-training of Qwen3-4B to 42.7% ASR on leak-guarded held-out tests, nearly matching synthetic-data baselines with a fraction of a day's data.
A hybrid first-order then zeroth-order optimization approach improves robustness of safety-aligned LLMs while preserving utility, with layer-wise sensitivity estimation for efficiency.
LaneRoPE adds an inter-sequence attention mask and extended RoPE to enable collaborative parallel sequence generation in LLMs, yielding accuracy gains on math reasoning under length limits.
CrossVLA develops a surrogate log-probability estimator for DPO on flow-matching VLAs, shows DoRA outperforming LoRA by +10.4 pp mean on LIBERO, and identifies inference bottlenecks with limited caching gains.
DPO-RLHF equivalence holds only conditionally on the optimal policy preferring human-preferred responses; otherwise DPO optimizes relative advantage and can prefer worse outputs, addressed by introducing CPO.
Pion modifies Muon's Newton-Schulz iterations into a controllable high-pass filter that anchors dominant singular values at 1 while suppressing noisy tails, outperforming Muon and AdamW in VLA and RLVR regimes.
VPD frames language feedback learning as variational EM so the teacher policy refines itself via trust-region updates on outcomes while the student learns dense token distributions on its own rollouts, outperforming fixed-teacher baselines on reasoning and code tasks.
Introduces TBPO, which derives a Bregman-divergence density-ratio matching objective for token-level preference optimization that generalizes DPO while preserving the induced optimal policy.
Introduces Block-R1 benchmark, Block-R1-41K dataset, and a conflict score to handle domain-specific optimal block sizes in RL post-training of diffusion LLMs.
PNAPO augments preference data with prior noise pairs and uses straight-line interpolation to create a tighter surrogate objective for offline alignment of rectified flow models.
The cumulative token IS ratio gives unbiased prefix correction and lower variance than full-sequence ratios for token-level gradients in LLM policy optimization, enabling CTPO to outperform GRPO and GSPO baselines on mathematical reasoning tasks.
PairAlign learns compact variable-length token sequences for audio via self-alignment on paired content-preserving views, achieving 55% fewer archive tokens than VQ while preserving edit-distance retrieval at 12.71 tokens/s.
Standard DPO surrogates are inconsistent for equicontinuous neural nets; SA-DPO provides structure-aware H-consistency bounds by adapting margins to semantic distance and shows heavy-tailed losses yield superior guarantees for capacity-bounded models via the Margin-Capacity Profile.
RLHF should decompose annotations into dimensions each matched to one of three models—extension, evidence, or authority—instead of applying a single unified pipeline.
HiPO improves LLM reasoning performance by optimizing preferences separately on response segments rather than entire outputs.
DDO-RM turns reward scores into a target distribution and applies KL-regularized mirror-descent projection on finite candidates to improve policies, outperforming DPO on Pythia-410M.
CapTrack shows post-training causes drift beyond facts, with instruction fine-tuning producing stronger behavioral changes than preference optimization across model families.
Magpie synthesizes 300K high-quality alignment instructions from Llama-3-Instruct via auto-regressive prompting on partial templates, enabling fine-tuned models to match official instruct performance on AlpacaEval, ArenaHard, and WildBench.
Equal reward weighting outperforms targeted weighting in RL-based BPMN generation across 48 configurations, with design choices producing effects as large as applying RL itself.
Transferring a weak model’s RL-induced log-ratio policy shift on a strong student’s own rollouts raises AIME accuracy more cheaply than imitating the weak teacher or running matched-step RL on the student.
Evolution Fine-Tuning trains LLMs on 156K trajectories spanning 371 tasks to achieve 10.22% average improvement on 22 held-out optimization tasks and match SOTA on select circle-packing problems when combined with test-time RL.
ViPSy constructs policy-aligned and visually grounded preference pairs for VLMs via visual cues from image variants, yielding SOTA hallucination reductions of 35.7% on AMBER and 24.5% on Object HalBench.
citing papers explorer
-
Negative Preference Optimization: From Catastrophic Collapse to Effective Unlearning
NPO enables stable unlearning of 50%+ training data in LLMs on TOFU by making collapse exponentially slower than gradient ascent, preserving sensible outputs where prior methods fail.
-
ORPO: Monolithic Preference Optimization without Reference Model
ORPO performs preference alignment during supervised fine-tuning via a monolithic odds ratio penalty, allowing 7B models to outperform larger state-of-the-art models on alignment benchmarks.
-
Bittensor Agent Arenas as a Trajectory Primitive: Distilling a Shopping Agent from ShoppingBench Subnet Traces
Trajectories from a Bittensor ShoppingBench subnet arena, filtered to retain only agentic tool-calling behavior, enable SFT+GRPO post-training of Qwen3-4B to 42.7% ASR on leak-guarded held-out tests, nearly matching synthetic-data baselines with a fraction of a day's data.
-
Aligned but Fragile: Enhancing LLM Safety Robustness via Zeroth-Order Optimization
A hybrid first-order then zeroth-order optimization approach improves robustness of safety-aligned LLMs while preserving utility, with layer-wise sensitivity estimation for efficiency.
-
LaneRoPE: Positional Encoding for Collaborative Parallel Reasoning and Generation
LaneRoPE adds an inter-sequence attention mask and extended RoPE to enable collaborative parallel sequence generation in LLMs, yielding accuracy gains on math reasoning under length limits.
-
CrossVLA: Cross-Paradigm Post-Training and Inference Optimization for Vision-Language-Action Models
CrossVLA develops a surrogate log-probability estimator for DPO on flow-matching VLAs, shows DoRA outperforming LoRA by +10.4 pp mean on LIBERO, and identifies inference bottlenecks with limited caching gains.
-
Conditional Equivalence of DPO and RLHF: Implicit Assumption, Failure Modes, and Provable Alignment
DPO-RLHF equivalence holds only conditionally on the optimal policy preferring human-preferred responses; otherwise DPO optimizes relative advantage and can prefer worse outputs, addressed by introducing CPO.
-
Rethinking Muon Beyond Pretraining: Spectral Failures and High-Pass Remedies for VLA and RLVR
Pion modifies Muon's Newton-Schulz iterations into a controllable high-pass filter that anchors dominant singular values at 1 while suppressing noisy tails, outperforming Muon and AdamW in VLA and RLVR regimes.
-
Learning from Language Feedback via Variational Policy Distillation
VPD frames language feedback learning as variational EM so the teacher policy refines itself via trust-region updates on outcomes while the student learns dense token distributions on its own rollouts, outperforming fixed-teacher baselines on reasoning and code tasks.
-
TokenRatio: Principled Token-Level Preference Optimization via Ratio Matching
Introduces TBPO, which derives a Bregman-divergence density-ratio matching objective for token-level preference optimization that generalizes DPO while preserving the induced optimal policy.
-
Block-R1: Rethinking the Role of Block Size in Multi-domain Reinforcement Learning for Diffusion Large Language Models
Introduces Block-R1 benchmark, Block-R1-41K dataset, and a conflict score to handle domain-specific optimal block sizes in RL post-training of diffusion LLMs.
-
Offline Preference Optimization for Rectified Flow with Noise-Tracked Pairs
PNAPO augments preference data with prior noise pairs and uses straight-line interpolation to create a tighter surrogate objective for offline alignment of rectified flow models.
-
Rethinking Importance Sampling in LLM Policy Optimization: A Cumulative Token Perspective
The cumulative token IS ratio gives unbiased prefix correction and lower variance than full-sequence ratios for token-level gradients in LLM policy optimization, enabling CTPO to outperform GRPO and GSPO baselines on mathematical reasoning tasks.
-
PairAlign: A Framework for Sequence Tokenization via Self-Alignment with Applications to Audio Tokenization
PairAlign learns compact variable-length token sequences for audio via self-alignment on paired content-preserving views, achieving 55% fewer archive tokens than VQ while preserving edit-distance retrieval at 12.71 tokens/s.
-
Mind the Gap: Structure-Aware Consistency in Preference Learning
Standard DPO surrogates are inconsistent for equicontinuous neural nets; SA-DPO provides structure-aware H-consistency bounds by adapting margins to semantic distance and shows heavy-tailed losses yield superior guarantees for capacity-bounded models via the Margin-Capacity Profile.
-
Three Models of RLHF Annotation: Extension, Evidence, and Authority
RLHF should decompose annotations into dimensions each matched to one of three models—extension, evidence, or authority—instead of applying a single unified pipeline.
-
HiPO: Hierarchical Preference Optimization for Adaptive Reasoning in LLMs
HiPO improves LLM reasoning performance by optimizing preferences separately on response segments rather than entire outputs.
-
DDO-RM: Distribution-Level Policy Improvement after Reward Learning
DDO-RM turns reward scores into a target distribution and applies KL-regularized mirror-descent projection on finite candidates to improve policies, outperforming DPO on Pythia-410M.
-
CapTrack: Multifaceted Evaluation of Forgetting in LLM Post-Training
CapTrack shows post-training causes drift beyond facts, with instruction fine-tuning producing stronger behavioral changes than preference optimization across model families.
-
Magpie: Alignment Data Synthesis from Scratch by Prompting Aligned LLMs with Nothing
Magpie synthesizes 300K high-quality alignment instructions from Llama-3-Instruct via auto-regressive prompting on partial templates, enabling fine-tuned models to match official instruct performance on AlpacaEval, ArenaHard, and WildBench.
-
Improving LLM-Generated Process Model Quality Through Reinforcement Learning: The Role of Reward Function Design
Equal reward weighting outperforms targeted weighting in RL-based BPMN generation across 48 configurations, with design choices producing effects as large as applying RL itself.
-
Weak-to-Strong Generalization via Direct On-Policy Distillation
Transferring a weak model’s RL-induced log-ratio policy shift on a strong student’s own rollouts raises AIME accuracy more cheaply than imitating the weak teacher or running matched-step RL on the student.
-
Evolution Fine-Tuning: Learning to Discover Across 371 Optimization Tasks
Evolution Fine-Tuning trains LLMs on 156K trajectories spanning 371 tasks to achieve 10.22% average improvement on 22 held-out optimization tasks and match SOTA on select circle-packing problems when combined with test-time RL.
-
Vision-driven Preference Synthesis for Mitigating Hallucinations in VLMs
ViPSy constructs policy-aligned and visually grounded preference pairs for VLMs via visual cues from image variants, yielding SOTA hallucination reductions of 35.7% on AMBER and 24.5% on Object HalBench.
-
Towards Spec Learning: Inference-Time Alignment from Preference Pairs
Proposes compiling preference pairs into readable natural-language specifications for inference-time LLM alignment, claiming outperformance over DPO on dense-preference domains.
-
Emo-LiPO: Listwise Preference Optimization for Fine-Grained Emotion Intensity Control in LLM-based Text-to-Speech
Emo-LiPO applies listwise preference optimization to model global emotion intensity ordering in LLM TTS, yielding better accuracy and controllability than supervised or DPO baselines on a new multi-speaker dataset.
-
Representation-Aware Advantage Estimation: Your Reward Model Provides More Than A Scalar Output
GraphAE builds graphs from RM hidden-state similarities among sampled responses and propagates advantages to improve RLHF sample efficiency.
-
FlowPRO: Reward-Free Reinforced Fine-Tuning of Flow-Matching VLAs via Proximalized Preference Optimization
FlowPRO applies proximalized preference optimization to flow-matching VLAs with intervention-rollback data to reach higher success rates on long-horizon bimanual tasks without rewards or critics.
-
SimuScene: Simulation-Ready Compositional 3D Scene Reconstruction from a Single Image
SimuScene feeds physics simulation diagnostics back into shape and layout estimation to correct geometric errors and output simulation-ready compositional scenes from single images.
-
Black-box, Adaptive, Efficient, Transferable, Harmful, Applicable... Attacks Are All You Need to Break LLMs
IHO is a new black-box jailbreak attack for LLMs that is adaptive, efficient, transferable across models and behaviors, and effective even against layered defenses without modification.
-
CogPortrait: Fine-Grained Eye-Region Control in Portrait Animation via Hierarchical Agent Planning
CogPortrait uses MLLM-based hierarchical planning to convert high-level labels into eye keypoints and a conditioned DiT model to produce portrait animations with improved eye-region accuracy on the new EMH benchmark.
-
Perceive-then-Plan: Layout-as-Policy for Monocular 3D Scene Layout Estimation
Introduces Layout-as-Policy (LaP) to turn 3D layout estimation into an iterative policy-learning refinement process for better physical coherence.
-
Convex Optimization for Alignment and Preference Learning on a Single GPU
COALA applies convex optimization reformulations of neural networks to direct preference optimization, claiming single-GPU training with ~18% of DPO's TFLOPs and competitive performance on multiple datasets and models up to 8B parameters.
-
Towards Context-Invariant Safety Alignment for Large Language Models
Introduces AIR, an asymmetric regularization that anchors open-ended safety prompts to verifiable ones via stop-gradient, improving invariance and accuracy when combined with group preference optimization.
-
DEFLECT: Delay-Robust Execution via Flow-matching Likelihood-Estimated Counterfactual Tuning for VLA Policies
DEFLECT is an offline post-training method that improves async VLA policy success rates under high inference delays by using flow-matching likelihood ratios on counterfactual fresh/stale action pairs from a frozen reference policy.
-
General Preference Reinforcement Learning
GPRL carries a k-dimensional skew-symmetric preference structure into policy updates with per-dimension advantages and a drift monitor, yielding 56.51% length-controlled win rate on AlpacaEval 2.0 from Llama-3-8B-Instruct while outperforming SimPO and SPPO on other benchmarks.
-
Power Reinforcement Post-Training of Text-to-Image Models with Super-Linear Advantage Shaping
Super-Linear Advantage Shaping (SLAS) introduces a non-linear geometric policy update for RL post-training of text-to-image models that reshapes the local policy space via advantage-dependent Fisher-Rao weighting to reduce reward hacking and improve performance over GRPO baselines.
-
Training-Free Cultural Alignment of Large Language Models via Persona Disagreement
DISCA converts within-country disagreement among World Values Survey personas into a bounded logit correction that reduces cultural misalignment by 10-24% on MultiTP for models 3.8B and larger across 20 countries, without any weight updates.
-
Team-Based Self-Play With Dual Adaptive Weighting for Fine-Tuning LLMs
TPAW uses teams of current and historical model checkpoints that collaborate and compete, plus adaptive weightings for responses and players, to improve self-supervised LLM alignment and outperform baselines.
-
Mechanistic Analysis of Alignment Algorithms in Language Models
Mechanistic analysis of six preference optimization methods reveals distinct geometric shifts in model representations, with KTO/GRPO enhancing separability while DPO/ORPO degrade it.
-
Beyond Pairs: Your Language Model is Secretly Optimizing a Preference Graph
GraphDPO generalizes pairwise DPO to a graph-structured Plackett-Luce objective over DAGs induced by rollout rankings, enforcing transitivity with linear complexity and recovering DPO as a special case.
-
Threshold-Guided Optimization for Visual Generative Models
A threshold-guided alignment method lets visual generative models be optimized directly from scalar human ratings instead of requiring paired preference data.
-
Gradient-Gated DPO: Stabilizing Preference Optimization in Language Models
Gate-DPO attenuates gradients on low-probability rejected responses to reduce probability collapse and improve chosen-response likelihood during preference optimization.
-
Multilingual Safety Alignment via Self-Distillation
MSD enables cross-lingual safety transfer in LLMs via self-distillation with Dual-Perspective Safety Weighting, improving safety in low-resource languages without target response data.
-
PERSA: Reinforcement Learning for Professor-Style Personalized Feedback with LLMs
PERSA combines RLHF with selective parameter-efficient updates to top transformer layers, raising style alignment scores from 35% to 96% on code feedback benchmarks while holding correctness near 100%.
-
Most Current Model Organisms Are Leaky: Perplexity Differencing Often Reveals Finetuning Objectives
Perplexity differencing on completions from short random prefills surfaces finetuning objectives in the vast majority of tested model organisms across sizes and types.
-
Aligning Language Models for Lyric-to-Melody Generation with Rule-Based Musical Constraints
Rule-generated preference data aligned via sequential DPO and KTO reduces musical constraint violations and improves coherence in lyric-to-melody generation over baselines.
-
Representation-Guided Parameter-Efficient LLM Unlearning
REGLU guides LoRA-based unlearning via representation subspaces and orthogonal regularization to outperform prior methods on forget-retain trade-off in LLM benchmarks.
-
AutoOR: Scalably Post-training LLMs to Autoformalize Operations Research Problems
AutoOR uses synthetic data generation and RL post-training with solver feedback to enable 8B LLMs to autoformalize linear, mixed-integer, and non-linear OR problems, matching larger models on benchmarks.
-
Bridging What the Model Thinks and How It Speaks: Self-Aware Speech Language Models for Expressive Speech Generation
SA-SLM uses variational information bottleneck for intent-aware bridging and self-criticism for realization-aware alignment to close the semantic-acoustic gap, outperforming open-source models and nearing GPT-4o-Audio expressiveness on EchoMind after training on 800 hours of data.