arXiv AI
Asymptotic Universal Alignment: A New Alignment Framework via Test-Time Scaling
Read the original on arXiv AI →The Flow has not summarised this story yet — read it at arXiv AI.
Related stories
arXiv AI
Aug 12
Multiplayer Nash Preference Optimization
arXiv:2509. 23102v4 Announce Type: replace Abstract: Reinforcement learning from human feedback (RLHF) has emerged as the standard paradigm for aligning large language models with human preferences.
arXiv AI
2d ago
Robust Nash Alignment under Preference Uncertainty
Robust Nash Alignment introduces a game-theoretic framework that seeks a policy with a high worst-case win rate against both an adversarial competitor and any preference kernel within an ambiguity set around a nominal preference. The authors propose a four-player primal-dual proxy game and an optimistic mirror descent-ascent algorithm to efficiently optimize this robust objective, proving convergence guarantees and demonstrating improved performance in controlled tabular games and LLM alignment experiments.
arXiv AI
Sep 1
RSPO: Regularized Self-Play Alignment of Large Language Models
arXiv:2503.00030v3 Announce Type: replace-cross Abstract: Self-play-based policy optimization has emerged as an effective approach for fine-tuning large language models (LLMs), formulating preference...
arXiv AI
Jun 2
S-SPPO: Semantic-Calibrated Self-Play Preference Optimization
arXiv:2606. 01561v1 Announce Type: new Abstract: Aligning Large Language Models (LLMs) with human preferences is often formulated via Direct Preference Optimization (DPO).
arXiv Machine Learning
Aug 18
Towards a theory of inference-time alignment with unknown rewards
arXiv:2608. 15402v1 Announce Type: new Abstract: Generative model alignment has received broad interest, and significant progress has been made in supervised fine-tuning and inference-time computation.
arXiv Machine Learning
Sep 22
Swiss-Knife: A Framework for Reconfigurable Externalised Multi-Objective Alignment at Decode Time
Swiss-Knife is a framework that extends decode‑time alignment for frozen language models by treating the alignment specification as a runtime object. It introduces hot‑swappable scoring blades, a batch normaliser, a pairwise aggregation operator, and a selection rule, and characterises admissible aggregation operators with a representation theorem. In experiments, Swiss‑Knife paired with DPO‑LoRA blades and an uncertainty‑aware pairwise tournament outperforms six existing decode‑time methods, achieving a higher harmonic F1 score, lower refusal rate, and faster objective reconfiguration.