Debiasing language-model training data for a target group frequently increases stereotyping or counter-stereotyping for non-target groups across categories, models, and scales.
Title resolution pending
10 Pith papers cite this work. Polarity classification is still indexing.
10
Pith papers citing it
years
2026 10representative citing papers
A total-variation robust Plackett-Luce loss whose adversarial ranking is the ascending-score order, with convex offline and weakly-convex online guarantees.
AMARIS augments rubric updates in RL for LLMs with a persistent memory of rollout analyses and prior edits, yielding gains such as +2.8 points on GPQA-Diamond over local-adaptive baselines.
A hierarchical genetic algorithm induces overthinking in black-box large reasoning models by perturbing logical structure, achieving up to 26.1x longer outputs on the MATH benchmark.
MRBTs are LLM-generated, SMT-verified behavior trees that supply modular reward functions and action masks, improving RL training efficiency and success rates on five compositional tasks over baselines.
Perturbing the prefix before next-token prediction, during both training and inference, improves out-of-distribution language-model generation and yields a conditional extrapolation guarantee.
QUIVER adaptively mixes objective evaluations with two types of preference queries in surrogate-assisted evolutionary multi-objective optimization to reduce final utility regret, reporting 25% gains on hard WFG benchmarks.
GS-Quant generates coarse-to-fine discrete codes for KG entities via semantic hierarchy injection and causal sequence reconstruction, enabling LLMs to perform knowledge graph completion by treating the codes as vocabulary tokens.
BlendIn replaces binary guidance acceptance with confidence-weighted distribution blending between base and guidance models, mitigating cascading failures in inference-time LLM alignment.
KnowledgeBerg is a 4,800-question, 17-language benchmark showing LLMs fail at systematically enumerating bounded knowledge universes and performing compositional set-based reasoning over them.
citing papers explorer
-
When Debiasing Backfires: Counterintuitive Side Effects of Preprocessing-Based Stereotype Mitigation
Debiasing language-model training data for a target group frequently increases stereotyping or counter-stereotyping for non-target groups across categories, models, and scales.
-
Distributionally Robust Listwise Preference Optimization
A total-variation robust Plackett-Luce loss whose adversarial ranking is the ascending-score order, with convex offline and weakly-convex online guarantees.
-
AMARIS: A Memory-Augmented Rubric Improvement System for Rubric-Based Reinforcement Learning
AMARIS augments rubric updates in RL for LLMs with a persistent memory of rollout analyses and prior edits, yielding gains such as +2.8 points on GPQA-Diamond over local-adaptive baselines.
-
Inducing Overthink: Hierarchical Genetic Algorithm-based DoS Attack on Black-Box Large Language Reasoning Models
A hierarchical genetic algorithm induces overthinking in black-box large reasoning models by perturbing logical structure, achieving up to 26.1x longer outputs on the MATH benchmark.
-
Reward Shaping and Action Masking for Compositional Tasks using Behavior Trees and LLMs
MRBTs are LLM-generated, SMT-verified behavior trees that supply modular reward functions and action masks, improving RL training efficiency and success rates on five compositional tasks over baselines.
-
Perturbation is All You Need for Extrapolating Language Models
Perturbing the prefix before next-token prediction, during both training and inference, improves out-of-distribution language-model generation and yields a conditional extrapolation guarantee.
-
QUIVER: Cost-Aware Adaptive Preference Querying in Surrogate-Assisted Evolutionary Multi-Objective Optimization
QUIVER adaptively mixes objective evaluations with two types of preference queries in surrogate-assisted evolutionary multi-objective optimization to reduce final utility regret, reporting 25% gains on hard WFG benchmarks.
-
GS-Quant: Granular Semantic and Generative Structural Quantization for Knowledge Graph Completion
GS-Quant generates coarse-to-fine discrete codes for KG entities via semantic hierarchy injection and causal sequence reconstruction, enabling LLMs to perform knowledge graph completion by treating the codes as vocabulary tokens.
-
To Intervene or Not: Guiding Inference-time Alignment with Probabilistic Model Blending
BlendIn replaces binary guidance acceptance with confidence-weighted distribution blending between base and guidance models, mitigating cascading failures in inference-time LLM alignment.
-
KnowledgeBerg: Evaluating Systematic Knowledge Coverage and Compositional Reasoning in Large Language Models
KnowledgeBerg is a 4,800-question, 17-language benchmark showing LLMs fail at systematically enumerating bounded knowledge universes and performing compositional set-based reasoning over them.