Aligning language models with human preferences

• URL: http://arxiv.org/abs/2404.12150v1
• Date: Thu, 18 Apr 2024 12:55:18 GMT
• Title: Aligning language models with human preferences
• Authors: Tomasz Korbak,
• Abstract summary: Language models (LMs) trained on vast quantities of text data can acquire sophisticated skills. They also manifest behaviors that violate human preferences. I explore several approaches to aligning LMs with human preferences.
• Score: 5.0994393083677
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Language models (LMs) trained on vast quantities of text data can acquire sophisticated skills such as generating summaries, answering questions or generating code. However, they also manifest behaviors that violate human preferences, e.g., they can generate offensive content, falsehoods or perpetuate social biases. In this thesis, I explore several approaches to aligning LMs with human preferences. First, I argue that aligning LMs can be seen as Bayesian inference: conditioning a prior (base, pretrained LM) on evidence about human preferences (Chapter 2). Conditioning on human preferences can be implemented in numerous ways. In Chapter 3, I investigate the relation between two approaches to finetuning pretrained LMs using feedback given by a scoring function: reinforcement learning from human feedback (RLHF) and distribution matching. I show that RLHF can be seen as a special case of distribution matching but distributional matching is strictly more general. In chapter 4, I show how to extend the distribution matching to conditional language models. Finally, in chapter 5 I explore a different root: conditioning an LM on human preferences already during pretraining. I show that involving human feedback from the very start tends to be more effective than using it only during supervised finetuning. Overall, these results highlight the room for alignment techniques different from and complementary to RLHF.

Related papers

• AlignSum: Data Pyramid Hierarchical Fine-tuning for Aligning with Human Summarization Preference [22.13596750775719]
  We introduce a novel human summarization preference alignment framework AlignSum. With AlignSum, PLMs like BART-Large surpass 175B GPT-3 in both automatic and human evaluations.
  arXiv  Detail & Related papers  (2024-10-01T05:14:48Z)
• Comparing Bad Apples to Good Oranges: Aligning Large Language Models via Joint Preference Optimization [105.3612692153615]
  A common technique for aligning large language models (LLMs) relies on acquiring human preferences. We propose a new axis that is based on eliciting preferences jointly over the instruction-response pairs. We find that joint preferences over instruction and response pairs can significantly enhance the alignment of LLMs.
  arXiv  Detail & Related papers  (2024-03-31T02:05:40Z)
• Dissecting Human and LLM Preferences [80.55271307662365]
  We find that humans are less sensitive to errors, favor responses that support their stances, and show clear dislike when models admit their limits. advanced LLMs like GPT-4-Turbo emphasize correctness, clarity, and harmlessness more. We show that preference-based evaluation can be intentionally manipulated.
  arXiv  Detail & Related papers  (2024-02-17T14:34:31Z)
• Personalized Soups: Personalized Large Language Model Alignment via Post-hoc Parameter Merging [148.77027765872006]
  We study Reinforcement Learning from Personalized Human Feedback (RLPHF) problem. LLMs are aligned to multiple preferences by modeling alignment as a Multi-Objective Reinforcement Learning (MORL) problem. We show that we can achieve personalized alignment by decomposing preferences into multiple dimensions.
  arXiv  Detail & Related papers  (2023-10-17T20:22:13Z)
• AlignDiff: Aligning Diverse Human Preferences via Behavior-Customisable Diffusion Model [69.12623428463573]
  AlignDiff is a novel framework to quantify human preferences, covering abstractness, and guide diffusion planning. It can accurately match user-customized behaviors and efficiently switch from one to another. We demonstrate its superior performance on preference matching, switching, and covering compared to other baselines.
  arXiv  Detail & Related papers  (2023-10-03T13:53:08Z)
• Fine-Grained Human Feedback Gives Better Rewards for Language Model Training [108.25635150124539]
  Language models (LMs) often exhibit undesirable text generation behaviors, including generating false, toxic, or irrelevant outputs. We introduce Fine-Grained RLHF, a framework that enables training and learning from reward functions that are fine-grained in two respects.
  arXiv  Detail & Related papers  (2023-06-02T17:11:37Z)
• Training Language Models with Language Feedback at Scale [50.70091340506957]
  We introduce learning from Language Feedback (ILF), a new approach that utilizes more informative language feedback. ILF consists of three steps that are applied iteratively: first, conditioning the language model on the input, an initial LM output, and feedback to generate refinements. We show theoretically that ILF can be viewed as Bayesian Inference, similar to Reinforcement Learning from human feedback.
  arXiv  Detail & Related papers  (2023-03-28T17:04:15Z)
• Meet in the Middle: A New Pre-training Paradigm [41.52858444519968]
  Most language models (LMs) are trained and applied in an autoregressive left-to-right fashion. We propose a new pre-training paradigm with techniques that jointly improve the training data efficiency. We show the effectiveness of our pre-training paradigm with extensive experiments on both programming and natural language models.
  arXiv  Detail & Related papers  (2023-03-13T17:17:11Z)
• Pretraining Language Models with Human Preferences [21.724817280998696]
  Language models (LMs) are pretrained to imitate internet text, including content that would violate human preferences if generated by an LM. Here, we explore alternative objectives for pretraining LMs in a way that also guides them to generate text aligned with human preferences.
  arXiv  Detail & Related papers  (2023-02-16T21:03:33Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.