Large Language Models (LLMs) possess the potential to exert substantial
influence on public perceptions and interactions with information. This raises
concerns about the societal impact that could arise if the ideologies within
these models can be easily manipulated. In this work, we investigate how
effectively LLMs can learn and generalize ideological biases from their
instruction-tuning data. Our findings reveal a concerning vulnerability:
exposure to only a small amount of ideologically driven samples significantly
alters the ideology of LLMs. Notably, LLMs demonstrate a startling ability to
absorb ideology from one topic and generalize it to even unrelated ones. The
ease with which LLMs' ideologies can be skewed underscores the risks associated
with intentionally poisoned training data by malicious actors or inadvertently
introduced biases by data annotators. It also emphasizes the imperative for
robust safeguards to mitigate the influence of ideological manipulations on
LLMs.

大型语言模型（LLMs）可能对公众对信息的认知和互动产生重大影响，因此需要对其内部思想意识是否可以轻易操纵引起关注。本研究探讨了 LLMs 在指导训练数据中学习和推广意识形态偏见的效果，并揭示了令人担忧的脆弱性：仅接触少量的思想倾向样本就会明显改变 LLMs 的思想意识，尤为显著的是，LLMs 展示了吸收一个主题的意识形态，并将其普遍应用于其他无关主题的惊人能力。LLMs 思想意识易受到恶意操作者故意操纵训练数据或数据标注者无意引入的偏见的风险，这强调了采取强有力的安全措施以减轻意识形态操纵对 LLMs 影响的重要性。

大语言模型对意识形态操控的易感性研究

How Susceptible are Large Language Models to Ideological Manipulation?

Impressive progress has been made on chat models based on Large Language
Models (LLMs) recently; however, there is a noticeable lag in multi-turn
conversations between open-source chat models (e.g., Alpaca and Vicuna) and the
leading chat models (e.g., ChatGPT and GPT-4). Through a series of analyses, we
attribute the lag to the lack of enough high-quality multi-turn
instruction-tuning data. The available instruction-tuning data for the
community are either single-turn conversations or multi-turn ones with certain
issues, such as non-human-like instructions, less detailed responses, or rare
topic shifts. In this paper, we address these challenges by introducing Parrot,
a highly scalable solution designed to automatically generate high-quality
instruction-tuning data, which are then used to enhance the effectiveness of
chat models in multi-turn conversations. Specifically, we start by training the
Parrot-Ask model, which is designed to emulate real users in generating
instructions. We then utilize Parrot-Ask to engage in multi-turn conversations
with ChatGPT across a diverse range of topics, resulting in a collection of 40K
high-quality multi-turn dialogues (Parrot-40K). These data are subsequently
employed to train a chat model that we have named Parrot-Chat. We demonstrate
that the dialogues gathered from Parrot-Ask markedly outperform existing
multi-turn instruction-following datasets in critical metrics, including topic
diversity, number of turns, and resemblance to human conversation. With only
40K training examples, Parrot-Chat achieves strong performance against other
13B open-source models across a range of instruction-following benchmarks, and
particularly excels in evaluations of multi-turn capabilities. We make all
codes, datasets, and two versions of the Parrot-Ask model based on LLaMA2-13B
and KuaiYii-13B available at this https URL

通过引入 Parrot，一种具有高可扩展性的解决方案，我们在多轮对话中自动生成高质量的指导语料，并利用这些语料来提高聊天模型在多轮对话中的效果。

Parrot: 通过学习提问增强多轮聊天模型

Parrot: Enhancing Multi-Turn Chat Models by Learning to Ask Questions

The integration of visual encoders and large language models (LLMs) has
driven recent progress in multimodal large language models (MLLMs). However,
the scarcity of high-quality instruction-tuning data for vision-language tasks
remains a challenge. The current leading paradigm, such as LLaVA, relies on
language-only GPT-4 to generate data, which requires pre-annotated image
captions and detection bounding boxes, suffering from understanding image
details. A practical solution to this problem would be to utilize the available
multimodal large language models (MLLMs) to generate instruction data for
vision-language tasks. However, it's worth noting that the currently accessible
MLLMs are not as powerful as their LLM counterparts, as they tend to produce
inadequate responses and generate false information. As a solution for
addressing the current issue, this paper proposes the Visual Instruction
Generation and Correction (VIGC) framework that enables multimodal large
language models to generate instruction-tuning data and progressively enhance
its quality on-the-fly. Specifically, Visual Instruction Generation (VIG)
guides the vision-language model to generate diverse instruction-tuning data.
To ensure generation quality, Visual Instruction Correction (VIC) adopts an
iterative update mechanism to correct any inaccuracies in data produced by VIG,
effectively reducing the risk of hallucination. Leveraging the diverse,
high-quality data generated by VIGC, we finetune mainstream models and validate
data quality based on various evaluations. Experimental results demonstrate
that VIGC not only compensates for the shortcomings of language-only data
generation methods, but also effectively enhances the benchmark performance.
The models, datasets, and code will be made publicly available.

本文提出了视觉指令生成和校正（VIGC）框架，使多模态大型语言模型能够生成指令调整数据并在生成过程中逐步提高其质量。

VIGC: 视觉指令生成和校正

VIGC: Visual Instruction Generation and Correction

Instruction tuning has emerged to enhance the capabilities of large language
models (LLMs) in providing appropriate outputs based on input instructions.
However, existing methods for collecting instruction-tuning data suffer from
limitations in scalability and affordability. In this paper, we propose
Dynosaur, a dynamic growth paradigm for instruction-tuning data curation. Built
upon the metadata of existing NLP datasets, we generate multiple task
instructions applicable to various NLP datasets and determine the relevant data
fields for constructing instruction-tuning data with LLMs. Dynosaur offers
several advantages: 1) lower generation costs (less than $12 for generating
800K instruction-tuning data), 2) good quality of instruction-tuning data
(better performance than Alpaca and Instruction GPT-4 on Super-NI with
comparable data sizes), and 3) the ability to grow dynamically by incorporating
new datasets from Huggingface Datasets Platform. We further investigate
continual learning as an approach to learning with the ever-growing
instruction-tuning dataset. We demonstrate that replay methods not only help
mitigate forgetting issues but help generalize to unseen tasks better. As a
novel continual learning scenario for instruction tuning, selecting tasks based
on instruction representations can be an effective replaying strategy. Code and
data are released at https://github.com/WadeYin9712/Dynosaur.

本文提出了 Dynosaur，一种基于现有 NLP 数据集元数据的动态增长范式，用于构建 LLM 的指令调整数据，同时研究将连续学习作为指令调整的一种新方法，并演示了在任务表示选择方面进行任务重演的有效性。