PaCE: Parsimonious Concept Engineering for Large Language Models

• URL: http://arxiv.org/abs/2406.04331v2
• Date: Tue, 05 Nov 2024 15:43:18 GMT
• Title: PaCE: Parsimonious Concept Engineering for Large Language Models
• Authors: Jinqi Luo, Tianjiao Ding, Kwan Ho Ryan Chan, Darshan Thaker, Aditya Chattopadhyay, Chris Callison-Burch, René Vidal,
• Abstract summary: Parsimonious Concept Engineering (PaCE) is a novel activation engineering framework for alignment. We construct a large-scale concept dictionary in the activation space, in which each atom corresponds to a semantic concept. We show that PaCE achieves state-of-the-art alignment performance while maintaining linguistic capabilities.
• Score: 57.740055563035256
• License:
• Abstract: Large Language Models (LLMs) are being used for a wide variety of tasks. While they are capable of generating human-like responses, they can also produce undesirable output including potentially harmful information, racist or sexist language, and hallucinations. Alignment methods are designed to reduce such undesirable outputs via techniques such as fine-tuning, prompt engineering, and representation engineering. However, existing methods face several challenges: some require costly fine-tuning for every alignment task; some do not adequately remove undesirable concepts, failing alignment; some remove benign concepts, lowering the linguistic capabilities of LLMs. To address these issues, we propose Parsimonious Concept Engineering (PaCE), a novel activation engineering framework for alignment. First, to sufficiently model the concepts, we construct a large-scale concept dictionary in the activation space, in which each atom corresponds to a semantic concept. Given any alignment task, we instruct a concept partitioner to efficiently annotate the concepts as benign or undesirable. Then, at inference time, we decompose the LLM activations along the concept dictionary via sparse coding, to accurately represent the activations as linear combinations of benign and undesirable components. By removing the latter ones from the activations, we reorient the behavior of the LLM towards the alignment goal. We conduct experiments on tasks such as response detoxification, faithfulness enhancement, and sentiment revising, and show that PaCE achieves state-of-the-art alignment performance while maintaining linguistic capabilities.

Related papers

• Supervised Chain of Thought [5.389461633686935]
  Chain of Thought (CoT) prompting offers a promising approach to solving complex reasoning tasks. One-prompt-for-all approach poses significant challenges for models to generate the correct reasoning steps. We show how task-specific supervision is essential for navigating the prompt space accurately and achieving optimal performance.
  arXiv  Detail & Related papers  (2024-10-18T06:25:27Z)
• Grounding Language Plans in Demonstrations Through Counterfactual Perturbations [25.19071357445557]
  Grounding the common-sense reasoning of Large Language Models (LLMs) in physical domains remains a pivotal yet unsolved problem for embodied AI. We show our approach improves the interpretability and reactivity of imitation learning through 2D navigation and simulated and real robot manipulation tasks.
  arXiv  Detail & Related papers  (2024-03-25T19:04:59Z)
• InfoCon: Concept Discovery with Generative and Discriminative Informativeness [7.160037417413006]
  We focus on the self-supervised discovery of manipulation concepts that can be adapted and reassembled to address various robotic tasks. We model manipulation concepts as generative and discriminative goals and derive metrics that can autonomously link them to meaningful sub-trajectories.
  arXiv  Detail & Related papers  (2024-03-14T14:14:04Z)
• Tuning-Free Accountable Intervention for LLM Deployment -- A Metacognitive Approach [55.613461060997004]
  Large Language Models (LLMs) have catalyzed transformative advances across a spectrum of natural language processing tasks. We propose an innovative textitmetacognitive approach, dubbed textbfCLEAR, to equip LLMs with capabilities for self-aware error identification and correction.
  arXiv  Detail & Related papers  (2024-03-08T19:18:53Z)
• DeAL: Decoding-time Alignment for Large Language Models [59.63643988872571]
  Large Language Models (LLMs) are nowadays expected to generate content aligned with human preferences. We propose DeAL, a framework that allows the user to customize reward functions and enables Detime Alignment of LLMs. Our experiments show that we can DeAL with fine-grained trade-offs, improve adherence to alignment objectives, and address residual gaps in LLMs.
  arXiv  Detail & Related papers  (2024-02-05T06:12:29Z)
• Can Large Language Models Follow Concept Annotation Guidelines? A Case Study on Scientific and Financial Domains [19.814974042343028]
  We examine the capacity of instruction-tuned language models to follow in-context concept guidelines for sentence labeling tasks. Our results show that although concept definitions consistently help in task performance, only the larger models have limited ability to work under counterfactual contexts.
  arXiv  Detail & Related papers  (2023-11-15T05:11:26Z)
• Towards Concept-Aware Large Language Models [56.48016300758356]
  Concepts play a pivotal role in various human cognitive functions, including learning, reasoning and communication. There is very little work on endowing machines with the ability to form and reason with concepts. In this work, we analyze how well contemporary large language models (LLMs) capture human concepts and their structure.
  arXiv  Detail & Related papers  (2023-11-03T12:19:22Z)
• Improving Language Models Meaning Understanding and Consistency by Learning Conceptual Roles from Dictionary [65.268245109828]
  Non-human-like behaviour of contemporary pre-trained language models (PLMs) is a leading cause undermining their trustworthiness. A striking phenomenon is the generation of inconsistent predictions, which produces contradictory results. We propose a practical approach that alleviates the inconsistent behaviour issue by improving PLM awareness.
  arXiv  Detail & Related papers  (2023-10-24T06:15:15Z)
• Text-Video Retrieval with Disentangled Conceptualization and Set-to-Set Alignment [17.423361070781876]
  We propose the Disentangled Conceptualization and Set-to-set Alignment (DiCoSA) to simulate the conceptualizing and reasoning process of human beings. For disentangled conceptualization, we divide the coarse feature into multiple latent factors related to semantic concepts. For set-to-set alignment, where a set of visual concepts correspond to a set of textual concepts, we propose an adaptive pooling method to aggregate semantic concepts.
  arXiv  Detail & Related papers  (2023-05-20T15:48:47Z)
• Pre-Trained Language Models for Interactive Decision-Making [72.77825666035203]
  We describe a framework for imitation learning in which goals and observations are represented as a sequence of embeddings. We demonstrate that this framework enables effective generalization across different environments. For test tasks involving novel goals or novel scenes, initializing policies with language models improves task completion rates by 43.6%.
  arXiv  Detail & Related papers  (2022-02-03T18:55:52Z)

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.