We present a theoretical framework showing that popular LLM alignment methods--including RLHF and its variants--can be understood as divergence estimators between aligned (safe or preferred) and unaligned (harmful or less-preferred) distributions. This perspective explains the emergence of separation in the latent space between safe and harmful prompts after alignment. As an application of our general divergence framework, we propose KLDO, a novel KL divergence-based alignment method, and empirically validate its effectiveness. We further show that using compliance-refusal datasets, rather than standard preference-based datasets, leads to stronger separation and improved safety alignment. Finally, to quantify the separation effect, we propose a distance-based metric in the prompt representation space, which also acts as a statistically significant indicator for model safety.

Safety and trustworthiness are indispensable requirements for real-world applications of AI systems using large language models (LLMs).

Safety and trustworthiness are indispensable requirements for real-world applications of AI systems using large language models (LLMs).

We evaluate Kahneman-Tversky Optimization (KTO) as a fine-tuning method for large language models (LLMs) in federated learning (FL) settings, comparing it against Direct Preference Optimization (DPO). Using Alpaca-7B as the base model, we fine-tune on a realistic dataset under both methods and evaluate performance using MT-Bench-1, Vicuna, and AdvBench benchmarks. Additionally, we introduce a redistributed dataset setup, where only KTO is applicable due to its ability to handle single-response feedback, unlike DPO's reliance on paired responses. Our results demonstrate that KTO, in both its original (KTOO) and redistributed (KTOR) configurations, consistently outperforms DPO across all benchmarks. In the redistributed setup, KTO further validates its flexibility and resilience by maintaining superior performance in scenarios where DPO cannot be applied. These findings establish KTO as a robust and scalable fine-tuning method for FL, motivating its adoption for privacy-preserving, decentralized, and heterogeneous environments.

To ensure safe usage, Large Language Models (LLMs) typically undergo alignment with human-defined values. However, this alignment often relies on primarily English data and is biased towards Western-centric values, limiting its effectiveness in low-resource language settings. In this paper, we describe our approach for aligning SEA-Lion-v2.1-Instruct (a Llama3-8B variant) to minimize toxicity in Singlish, an English creole specific to Singapore. We find that supervised fine-tuning and Kahneman-Tversky Optimization (KTO) on paired and unpaired preferences is more sample efficient and yields significantly better results than Direct Preference Optimization (DPO). Our analysis reveals that DPO implicitly enforces a weaker safety objective than KTO, and that SFT complements KTO by improving training stability. Finally, we introduce a simple but novel modification to KTO, KTO-S, which improves training stability through better gradient exploitation. Overall, we present a general approach for safety alignment conducive to low-resource English languages, successfully reducing toxicity by 99\% on our Singlish benchmark, with gains generalizing to the broader TOXIGEN dataset while maintaining strong performance across standard LLM benchmarks.

Offline preference alignment for language models such as Direct Preference Optimization (DPO) is favored for its effectiveness and simplicity, eliminating the need for costly reinforcement learning. Various offline algorithms have been developed for different data settings, yet they lack a unified understanding. In this study, we introduce Pior-Informed Preference Alignment (PIPA), a unified, RL-free probabilistic framework that formulates language model preference alignment as a Maximum Likelihood Estimation (MLE) problem with prior constraints. This method effectively accommodates both paired and unpaired data, as well as answer and step-level annotations. We illustrate that DPO and KTO are special cases with different prior constraints within our framework. By integrating different types of prior information, we developed two variations of PIPA: PIPA-M and PIPA-N. Both algorithms demonstrate a $3\sim10\%$ performance enhancement on the GSM8K and MATH benchmarks across all configurations, achieving these gains without additional training or computational costs compared to existing algorithms.

We propose a theoretical framework demonstrating that popular Large Language Model (LLM) alignment methods, including Reinforcement Learning from Human Feedback (RLHF) and alternatives, fundamentally function as divergence estimators between aligned (preferred or safe) and unaligned (less-preferred or harmful) distributions. This explains the separation phenomenon between safe and harmful prompts in the model hidden representation after alignment. Inspired by the theoretical results, we identify that some alignment methods are better than others in terms of separation and, introduce a new method, KLDO, and further demonstrate the implication of our theories. We advocate for compliance-refusal datasets over preference datasets to enhance safety alignment, supported by both theoretical reasoning and empirical evidence. Additionally, to quantify safety separation, we leverage a distance metric in the representation space and statistically validate its efficacy as a statistical significant indicator of LLM resilience against jailbreak attacks.

Contrastive learning models have demonstrated impressive abilities to capture semantic similarities by aligning representations in the embedding space. However, their performance can be limited by the quality of the training data and its inherent biases. While Reinforcement Learning from Human Feedback (RLHF) and Direct Preference Optimization (DPO) have been applied to generative models to align them with human preferences, their use in contrastive learning has yet to be explored. This paper introduces a novel method for training contrastive learning models using Preference Optimization (PO) to break down complex concepts. Our method systematically aligns model behavior with desired preferences, enhancing performance on the targeted task. In particular, we focus on enhancing model robustness against typographic attacks, commonly seen in contrastive models like CLIP. We further apply our method to disentangle gender understanding and mitigate gender biases, offering a more nuanced control over these sensitive attributes. Our experiments demonstrate that models trained using PO outperform standard contrastive learning techniques while retaining their ability to handle adversarial challenges and maintain accuracy on other downstream tasks. This makes our method well-suited for tasks requiring fairness, robustness, and alignment with specific preferences. We evaluate our method on several vision-language tasks, tackling challenges such as typographic attacks. Additionally, we explore the model's ability to disentangle gender concepts and mitigate gender bias, showcasing the versatility of our approach.

Multi-turn interactions between large language models (LLMs) and users naturally include implicit feedback signals. If an LLM responds in an unexpected way to an instruction, the user is likely to signal it by rephrasing the request, expressing frustration, or pivoting to an alternative task. Such signals are task-independent and occupy a relatively constrained subspace of language, allowing the LLM to identify them even if it fails on the actual task. This creates an avenue for continually learning from interactions without additional annotations. We introduce ReSpect, a method to learn from such signals in past interactions via retrospection. We deploy ReSpect in a new multimodal interaction scenario, where humans instruct an LLM to solve an abstract reasoning task with a combinatorial solution space. Through thousands of interactions with humans, we show how ReSpect gradually improves task completion rate from 31% to 82%, all without any external annotation.

Safety and trustworthiness are indispensable requirements for real-world applications of AI systems using large language models (LLMs). This paper formulates human value alignment as an optimization problem of the language model policy to maximize reward under a safety constraint, and then proposes an algorithm, Stepwise Alignment for Constrained Policy Optimization (SACPO). One key idea behind SACPO, supported by theory, is that the optimal policy incorporating reward and safety can be directly obtained from a reward-aligned policy. Building on this key idea, SACPO aligns LLMs step-wise with each metric while leveraging simple yet powerful alignment algorithms such as direct preference optimization (DPO). SACPO offers several advantages, including simplicity, stability, computational efficiency, and flexibility of algorithms and datasets. Under mild assumptions, our theoretical analysis provides the upper bounds on optimality and safety constraint violation. Our experimental results show that SACPO can fine-tune Alpaca-7B better than the state-of-the-art method in terms of both helpfulness and harmlessness.