United States

Large Language Models (LLMs) have demonstrated impressive capabilities in complex reasoning tasks. However, they can be easily misled by unfaithful arguments during conversations, even when their original statements are correct. To this end, we investigate the problem of maintaining faithful integrity in LLMs. This involves ensuring that LLMs adhere to their faithful statements in the face of opposing arguments and are able to correct their incorrect statements when presented with faithful arguments. In this work, we propose a novel framework, named Alignment for Faithful Integrity with Confidence Estimation (AFICE), which aims to align the LLM responses with faithful integrity. Specifically, AFICE first designs a Bilateral Confidence Estimation (BCE) approach for calibrating each response generated by the LLM given a specific context, which simultaneously estimate the model&#39;s confidence to the question based on the internal states during decoding as well as to the answer based on cumulative probability ratios. With the BCE, we construct a preference dataset for aligning the model using Direct Preference Optimization (DPO). Extensive experimental results on a wide range of benchmarks demonstrate significant improvements in the LLM&#39;s ability to maintain faithful responses when encountering opposing arguments, ensuring both the practical utility and trustworthiness of LLMs in complex interactive settings.

AAAI 2025

Aligning Large Language Models for Faithful Integrity Against Opposing Argument

**Welcome to EMNLP 2023!**

On behalf of the EMNLP 2023 Organizing Committee, I extend a warm and heartfelt welcome to all of you to the 2023 Conference on Empirical Methods in Natural Language Processing (EMNLP). It is with immense pleasure and excitement that we gather here in Singapore, a vibrant hub of innovation and technological advancement.

The conference program is packed with insightful presentations, thought-provoking workshops, and engaging networking opportunities. In addition to the technical sessions, we have also planned several social events that will provide you with the opportunity to connect with your colleagues.

The conference is held in person at the Resorts World Convention Centre in Singapore, and available on-line with the help of Underline.

For those in Singapore, I hope that you will find time to explore the exciting Sentosa Island, Gardens by the Bay, Singapore Botanic Gardens and the many other attractions unique to Singapore.

Of course, we are grateful to our sponsors and partners for their generous support of EMNLP 2023, whose contributions make it possible for us to host this world-class event.

Yuji Matsumoto (RIKEN AIP) 
EMNLP 2023 General Chair

To access the **EMNLP 2023** event page on Underline, you need to register for the Conference. 
Please follow **[this link](https://2023.emnlp.org/registration/)** for more details.

EMNLP 2023

EMNLP 2023 took place in Singapore from Dec 6th to Dec 10th, 2023.

Large Language Models (LLMs) have demonstrated impressive capabilities in complex reasoning tasks. However, they can be easily misled by unfaithful arguments during conversations, even when their original statements are correct. To this end, we investigate the problem of maintaining faithful integrity in LLMs. This involves ensuring that LLMs adhere to their faithful statements in the face of opposing arguments and are able to correct their incorrect statements when presented with faithful arguments. In this work, we propose a novel framework, named Alignment for Faithful Integrity with Confidence Estimation (AFICE), which aims to align the LLM responses with faithful integrity. Specifically, AFICE first designs a Bilateral Confidence Estimation (BCE) approach for calibrating each response generated by the LLM given a specific context, which simultaneously estimate the model's confidence to the question based on the internal states during decoding as well as to the answer based on cumulative probability ratios. With the BCE, we construct a preference dataset for aligning the model using Direct Preference Optimization (DPO). Extensive experimental results on a wide range of benchmarks demonstrate significant improvements in the LLM's ability to maintain faithful responses when encountering opposing arguments, ensuring both the practical utility and trustworthiness of LLMs in complex interactive settings.

poster

We are pleased to announce the Thirty-Ninth AAAI Conference on Artificial Intelligence (AAAI-25), which will be held in Philadelphia, Pennsylvania at the Pennsylvania Convention Center from February 25 to March 4, 2025.

The purpose of the AAAI conference series is to promote research in Artificial Intelligence (AI) and foster scientific exchange between researchers, practitioners, scientists, students, and engineers across the entirety of AI and its affiliated disciplines. AAAI-25 will feature technical paper presentations, special tracks, invited speakers, workshops, tutorials, poster sessions, senior member presentations, competitions, and exhibit programs, and a range of other activities to be announced.

### [Invited Speakers](https://aaai.org/conference/aaai/aaai-25/aaai-25-invited-speakers/)

Register [here](https://aaai.org/conference/aaai/aaai-25/registration/)

The purpose of the AAAI conference series is to promote research in Artificial Intelligence (AI) and foster scientific exchange between researchers, practitioners, scientists, students, and engineers across the entirety of AI and its affiliated disciplines. AAAI-25 will feature technical paper presentations, special tracks, invited speakers, workshops, tutorials, poster sessions, senior member presentations, competitions, and exhibit programs, and a range of other activities to be announced.



When LLMs are deployed in sensitive, human-facing settings, it is crucial that they do not output unsafe, biased, or privacy-violating outputs. For this reason, models are both trained and instructed to refuse to answer unsafe prompts such as ``Tell me how to build a bomb." We find that, despite these safeguards, it is possible to break model defenses simply by appending a space to the end of a model's input. In a study of ten open-source models, we demonstrate that simple perturbation is able to cause the majority of models to generate harmful outputs with very high probability. We examine the causes of this behavior, finding that the contexts in which single spaces occur in tokenized training data encourage models to generate lists when prompted, overriding training signals to refuse to answer unsafe requests. Our findings underscore the fragile state of current model alignment and promote the importance of developing more robust alignment methods.

Single Character Perturbations Break LLM Alignment

Large language models (LLMs) are expected to follow instructions from users and engage in conversations.  Techniques to enhance LLMs' instruction-following capabilities typically fine-tune them using data structured according to a predefined chat template. Although chat templates are shown to be effective in optimizing LLM performance, their impact on safety alignment of LLMs has been less understood, which is crucial for deploying LLMs safely at scale.

In this paper, we investigate how chat templates affect safety alignment of LLMs. We identify a common vulnerability, named ChatBug, that is introduced by chat templates. Our key insight to identify ChatBug is that the chat templates provide a rigid format that need to be followed by LLMs, but not by users. Hence, a malicious user may not necessarily follow the chat template when prompting LLMs. Instead, malicious users could leverage their knowledge of the chat template and accordingly craft their prompts to bypass safety alignments of LLMs. We study two attacks to exploit the ChatBug vulnerability. Additionally, we demonstrate that the success of multiple existing attacks can be attributed to the ChatBug vulnerability. We show that a malicious user can exploit the ChatBug vulnerability of eight state-of-the-art (SOTA) LLMs and effectively elicit unintended responses from these models. Moreover, we show that ChatBug can be exploited by existing jailbreak attacks to enhance their attack success rates. We investigate potential countermeasures to ChatBug. Our results show that while adversarial training effectively mitigates the ChatBug vulnerability, the victim model incurs significant performance degradation. These results highlight the trade-off between safety alignment and helpfulness. Developing new methods for instruction tuning to balance this trade-off is an open and critical direction for future research.

ChatBug: A Common Vulnerability of Aligned LLMs Induced by Chat Templates

Despite the extent of recent advances in Machine Learning (ML) and Neural Networks, providing formal guarantees on the behavior of these systems is still an open problem, and a crucial requirement for their adoption in regulated or safety-critical scenarios. We consider the task of training differentiable ML models guaranteed to satisfy designer-chosen properties, stated as input-output implications. This is very challenging, due to the computational complexity of rigorously verifying and enforcing compliance in modern neural models. We provide an innovative approach based on three components: 1) a general, simple architecture enabling efficient verification with a conservative semantic; 2) a rigorous training algorithm based on the Projected Gradient Method; 3) a formulation of the problem of searching for strong counterexamples. The proposed framework, being only marginally affected by model complexity, scales well to practical applications, and produces models that provide full property satisfaction guarantees. We evaluate our approach on properties defined by linear inequalities in regression, and on mutually exclusive classes in multilabel classification. Our approach is competitive with a baseline that includes property enforcement during preprocessing, i.e. on the training data, as well as during postprocessing, i.e. on the model predictions. Finally, our contributions establish a framework that opens up multiple research directions and potential improvements.

SMLE: Safe Machine Learning via Embedded Overapproximation

The emergence of Vision-Language Models (VLMs) is significant advancement in integrating computer vision with Large Language Models (LLMs) to enhance multi-modal machine learning capabilities. However, this progress has made VLMs vulnerable to advanced adversarial attacks, raising concerns about reliability. Objective of this paper is to assess resilience of VLMs against jailbreak attacks that can compromise model safety compliance and result in harmful outputs. To evaluate VLM's ability to maintain robustness against adversarial input perturbations, we propose novel metric called \textbf{Retention Score}. Retention Score is multi-modal evaluation metric that includes Retention-I and Retention-T scores for quantifying jailbreak risks in visual and textual components of VLMs. Our process involves generating synthetic image-text pairs using conditional diffusion model. These pairs are then predicted for toxicity score by VLM alongside toxicity judgment classifier. By calculating margin in toxicity scores, we can quantify robustness of VLM in attack-agnostic manner. Our work has four main contributions. First, we prove that Retention Score can serve as certified robustness metric. Second, we demonstrate that most VLMs with visual components are less robust against jailbreak attacks than corresponding plain VLMs. Additionally, we evaluate black-box VLM APIs and find that security settings in Google Gemini significantly affect score and robustness. Moreover, robustness of GPT4V is similar to medium settings of Gemini. Finally, our approach offers time-efficient alternative to existing adversarial attack methods and provides consistent model robustness rankings when evaluated on VLMs including MiniGPT-4, InstructBLIP, and LLaVA.

Retention Score: Quantifying Jailbreak Risks for Vision Language Models

Recent work shows that LLMs are vulnerable to data poisoning, in which they are trained on partially corrupted or harmful data. Poisoned data is hard to detect, breaks guardrails, and leads to undesirable and harmful behavior. Given the intense efforts by leading labs to train and deploy increasingly larger and more capable LLMs, it is critical to ask if the risk of data poisoning will be naturally mitigated by scale, or if it is an increasing threat. We consider three threat models by which data poisoning can occur: malicious fine-tuning, imperfect data curation, and intentional data contamination. Our experiments evaluate the effects of data poisoning on 23 frontier LLMs ranging from 1.5-72 billion parameters on three datasets which speak to each of our threat models. We find that larger LLMs are increasingly vulnerable, learning harmful behavior -- including sleeper agent behavior -- significantly more quickly than smaller LLMs with even minimal data poisoning. These results underscore the need for robust safeguards against data poisoning in larger LLMs.

Scaling Trends for Data Poisoning in LLMs

Recent years have witnessed an increase in the parameter size of frontier AI models by multiple orders of magnitude. This trend is driven by empirical observations, known as scaling laws, which show that model performance scales with model size, dataset size, and computational power. Motivated by this, researchers are training ever-larger models in pursuit of unlocking new capabilities. However, the growing complexity of these models makes understanding their inner workings increasingly challenging. Interpretability is crucial not only in fields like medicine and biotechnology, where understanding the internals of these models could lead to new insights but also in super alignment, where it is the goal to ensure that AI is aligned and acts according to human values and interests. We present a generic, scalable first-of-its-kind method for automatically interpreting neural networks. In a proof-of-concept study we establish the viability of converting neural network activations - here for the first layer of a Convolutional Neural Network - into human-readable language. Additionally, we propose modifications to scale this method for understanding neural networks of any size. In anticipation of more capable large language models, this method could enable the monitoring of their internal mechanisms and decisions.

Neurons to Words: A Novel Method for Automated Neural Network Interpretability and Alignment

Farmers rely on in-field observations to make well-informed crop management decisions to maximize profit and minimize adverse environmental impact.
However, obtaining real-world crop state measurements is labor-intensive, time-consuming and expensive.
In most cases, it is not feasible to gather crop state measurements before every decision moment.
Moreover, in previous research pertaining to farm management optimization, these observations are often assumed to be readily available without any cost, which is unrealistic.
Hence, enabling optimization without the need to have *temporally complete* crop state observations is important.
An approach to that problem is to include measuring as part of decision making.
As a solution, we apply reinforcement learning (RL) to recommend opportune moments to simultaneously measure crop features and apply nitrogen fertilizer.
With realistic considerations, we design an RL environment with explicit crop feature measuring costs.
While balancing costs, we find that an RL agent, trained with recurrent PPO, discovers adaptive measuring policies that follow critical crop development stages, with results aligned by what domain experts would consider a sensible approach. 
Our results highlight the importance of measuring when crop feature measurements are not readily available.

To Measure or Not: A Cost-Sensitive, Selective Measuring Environment for Agricultural Management Decisions with Reinforcement Learning

In recent years, the rapid development of Large Language Models (LLMs) has highlighted the urgent need for large-scale, high-quality, and diverse data. We have launched an LLM data co-creation platform aimed at bringing together a wide range of participants to contribute data. Within six months, the platform has attracted over 10,000 participants who contributed more than 150,000 data entries across more than 200 tasks. An observable user cohort was constructed around the question, ``Who is the best data contributor?" along with sub-questions concerning user preferences, task competence, and more. Through a detailed analysis of data contributors, this paper reveals several data collection patterns related to human factors. It reveals that contributors who provide high-quality data often do not meet initial expectations, as their behavior exhibits typical characteristics of the Dunning-Kruger effect. This paper examined the cognitive bias between users' self-assessment and actual abilities, where individuals tend to overestimate their capabilities in certain tasks, leading to a decreased willingness to continue contributing and a consequent waste of human resources. To address this issue, we propose a task reassignment method based on multi-task fine-tuning of small language models (SLMs) to better align user groups with appropriate task types. After the reallocation, we observed a significant increase in user engagement and platform benefits, along with improved overall platform efficiency. The versatility of this method makes it applicable to broader data collection scenarios.

Cognitive Bias and Reassignment: Who Can Contribute High Quality LLM Data

Existing video fact-checking datasets often lack detailed evidence and explanations, compromising the reliability and interpretability of fact-checking methods. To address these gaps, we developed a novel dataset featuring comprehensive annotations for each news item, including veracity labels, the rationales behind these labels, and supporting evidence. This dataset significantly enhances models' ability to accurately identify and explain video content. 
We also present an explainable automatic framework 
$\textbf{3MFact}$, utilizing $\textbf{M}$ulti-role $\textbf{M}$ultimodal $\textbf{M}$odels for video $\textbf{Fact}$-checking. Our framework iteratively gathers and synthesizes online evidence to progressively determine the veracity label, generating three key outputs: veracity label, rationale, and supported evidence. We aim for this work to be a pioneering effort, providing robust support for the field of video fact-checking.

Pioneering Explainable Video Fact-Checking with a New Dataset and Multi-role Multimodal Model Approach

Extracting fine-grained OCR text from aged documents in diacritic languages remains challenging due to unexpected artifacts, time-induced degradation, and lack of datasets. While standalone spell correction approaches have been proposed, they show limited performance for historical documents due to numerous possible OCR error combinations and differences between modern and classical corpus distributions.
We propose a method utilizing available content-focused ebooks as a reference base to correct imperfect OCR-generated text, supported by large language models. This technique generates high-precision pseudo-page-to-page labels for diacritic languages, where small strokes pose significant challenges in historical conditions. The pipeline eliminates various types of noise from aged documents and addresses issues such as missing characters, words, and disordered sequences.
Our post-processing method, which generated a large OCR dataset of classical Vietnamese books, achieved a mean grading score of 8.72 on a 10-point scale. This outperformed the state-of-the-art transformer-based Vietnamese spell correction model, which scored 7.03, when evaluated on a sampled subset of the dataset. We also trained a baseline OCR model to assess and compare it with well-known engines. Experimental results demonstrate the strength of our baseline model compared to widely used open-source solutions. The resulting dataset will be released publicly to support future studies.

Downloads

Next from AAAI 2025

Single Character Perturbations Break LLM Alignment

Similar lecture

Pointwise Mutual Information Based Metric and Decoding Strategy for Faithful Generation in Document Grounded Dialogs

Stay up to date with the latest Underline news!

PRESENTATIONS

CONFERENCES

COMPANY

RESOURCES