Language models (LMs) are known to represent the perspectives of some social groups better than others, which may impact their performance, especially on subjective tasks such as content moderation and hate speech detection. To explore how LMs represent different perspectives, existing research focused on positional alignment, i.e., how closely the models mimic the opinions and stances of different groups, e.g., liberals or conservatives. However, human communication also encompasses emotional and moral dimensions. We define the problem of affective alignment, which measures how LMs' emotional and moral tone represents those of different groups. By comparing the affect of responses generated by 36 LMs to the affect of Twitter messages, we observe significant misalignment of LMs with both ideological groups. This misalignment is larger than the partisan divide in the U.S. Even after steering the LMs towards specific ideological perspectives, the misalignment and liberal tendencies of the model persist, suggesting a systemic bias within LMs.
相關內容
Yes. SE data can have "smoother" boundaries between classes (compared to traditional AI data sets). To be more precise, the magnitude of the second derivative of the loss function found in SE data is typically much smaller. A new hyper-parameter optimizer, called SMOOTHIE, can exploit this idiosyncrasy of SE data. We compare SMOOTHIE and a state-of-the-art AI hyper-parameter optimizer on three tasks: (a) GitHub issue lifetime prediction (b) detecting static code warnings false alarm; (c) defect prediction. For completeness, we also show experiments on some standard AI datasets. SMOOTHIE runs faster and predicts better on the SE data--but ties on non-SE data with the AI tool. Hence we conclude that SE data can be different to other kinds of data; and those differences mean that we should use different kinds of algorithms for our data. To support open science and other researchers working in this area, all our scripts and datasets are available on-line at //github.com/yrahul3910/smoothness-hpo/.
Knowledge editing techniques have been increasingly adopted to efficiently correct the false or outdated knowledge in Large Language Models (LLMs), due to the high cost of retraining from scratch. Meanwhile, one critical but under-explored question is: can knowledge editing be used to inject harm into LLMs? In this paper, we propose to reformulate knowledge editing as a new type of safety threat for LLMs, namely Editing Attack, and conduct a systematic investigation with a newly constructed dataset EditAttack. Specifically, we focus on two typical safety risks of Editing Attack including Misinformation Injection and Bias Injection. For the risk of misinformation injection, we first categorize it into commonsense misinformation injection and long-tail misinformation injection. Then, we find that editing attacks can inject both types of misinformation into LLMs, and the effectiveness is particularly high for commonsense misinformation injection. For the risk of bias injection, we discover that not only can biased sentences be injected into LLMs with high effectiveness, but also one single biased sentence injection can cause a high bias increase in general outputs of LLMs, which are even highly irrelevant to the injected sentence, indicating a catastrophic impact on the overall fairness of LLMs. Then, we further illustrate the high stealthiness of editing attacks, measured by their impact on the general knowledge and reasoning capacities of LLMs, and show the hardness of defending editing attacks with empirical evidence. Our discoveries demonstrate the emerging misuse risks of knowledge editing techniques on compromising the safety alignment of LLMs.
The Large Language Models (LLMs), such as GPT and BERT, were proposed for natural language processing (NLP) and have shown promising results as general-purpose language models. An increasing number of industry professionals and researchers are adopting LLMs for program analysis tasks. However, one significant difference between programming languages and natural languages is that a programmer has the flexibility to assign any names to variables, methods, and functions in the program, whereas a natural language writer does not. Intuitively, the quality of naming in a program affects the performance of LLMs in program analysis tasks. This paper investigates how naming affects LLMs on code analysis tasks. Specifically, we create a set of datasets with code containing nonsense or misleading names for variables, methods, and functions, respectively. We then use well-trained models (CodeBERT) to perform code analysis tasks on these datasets. The experimental results show that naming has a significant impact on the performance of code analysis tasks based on LLMs, indicating that code representation learning based on LLMs heavily relies on well-defined names in code. Additionally, we conduct a case study on some special code analysis tasks using GPT, providing further insights.
Humans appear to have a critical period (CP) for language acquisition: Second language (L2) acquisition becomes harder after early childhood, and ceasing exposure to a first language (L1) after this period (but not before) typically does not lead to substantial loss of L1 proficiency. It is unknown whether these CP effects result from innately determined brain maturation or as a stabilization of neural connections naturally induced by experience. In this study, we use language models (LMs) to test the extent to which these phenomena are peculiar to humans, or shared by a broader class of language learners. We vary the age of exposure by training LMs on language pairs in various experimental conditions, and find that LMs, which lack any direct analog to innate maturational stages, do not show CP effects when trained sequentially on L1 and L2. Our results contradict the claim that CP effects are an inevitable result of learning in statistical learners, and they are consistent with an innate mechanism for CP effects. We show that we can reverse-engineer the CP by introducing a regularizer partway through training to simulate a maturational decrease in plasticity. All in all, our results suggest that L1 learning on its own may not be enough to induce a CP, and additional engineering is necessary to make language models more cognitively plausible.
One of the major aspects contributing to the striking performance of large language models (LLMs) is the vast amount of factual knowledge accumulated during pre-training. Yet, many LLMs suffer from self-inconsistency, which raises doubts about their trustworthiness and reliability. In this paper, we focus on entity type ambiguity and analyze current state-of-the-art LLMs for their proficiency and consistency in applying their factual knowledge when prompted for entities under ambiguity. To do so, we propose an evaluation protocol that disentangles knowing from applying knowledge, and test state-of-the-art LLMs on 49 entities. Our experiments reveal that LLMs perform poorly with ambiguous prompts, achieving only 80% accuracy. Our results further demonstrate systematic discrepancies in LLM behavior and their failure to consistently apply information, indicating that the models can exhibit knowledge without being able to utilize it, significant biases for preferred readings, as well as self inconsistencies. Our study highlights the importance of handling entity ambiguity in future for more trustworthy LLMs
Language agents, built on top of language models (LMs), are systems that can interact with complex environments, such as the open web. In this work, we examine whether such agents can perform realistic and time-consuming tasks on the web, e.g., monitoring real-estate markets or locating relevant nearby businesses. We introduce AssistantBench, a challenging new benchmark consisting of 214 realistic tasks that can be automatically evaluated, covering different scenarios and domains. We find that AssistantBench exposes the limitations of current systems, including language models and retrieval-augmented language models, as no model reaches an accuracy of more than 25 points. While closed-book LMs perform well, they exhibit low precision since they tend to hallucinate facts. State-of-the-art web agents reach a score of near zero. Additionally, we introduce SeePlanAct (SPA), a new web agent that significantly outperforms previous agents, and an ensemble of SPA and closed-book models reaches the best overall performance. Moreover, we analyze failures of current systems and highlight that web navigation remains a major challenge.
Property-based testing (PBT), while an established technique in the software testing research community, is still relatively underused in real-world software. Pain points in writing property-based tests include implementing diverse random input generators and thinking of meaningful properties to test. Developers, however, are more amenable to writing documentation; plenty of library API documentation is available and can be used as natural language specifications for PBTs. As large language models (LLMs) have recently shown promise in a variety of coding tasks, we investigate using modern LLMs to automatically synthesize PBTs using two prompting techniques. A key challenge is to rigorously evaluate the LLM-synthesized PBTs. We propose a methodology to do so considering several properties of the generated tests: (1) validity, (2) soundness, and (3) property coverage, a novel metric that measures the ability of the PBT to detect property violations through generation of property mutants. In our evaluation on 40 Python library API methods across three models (GPT-4, Gemini-1.5-Pro, Claude-3-Opus), we find that with the best model and prompting approach, a valid and sound PBT can be synthesized in 2.4 samples on average. We additionally find that our metric for determining soundness of a PBT is aligned with human judgment of property assertions, achieving a precision of 100% and recall of 97%. Finally, we evaluate the property coverage of LLMs across all API methods and find that the best model (GPT-4) is able to automatically synthesize correct PBTs for 21% of properties extractable from API documentation.
The integration of new modalities into frontier AI systems offers exciting capabilities, but also increases the possibility such systems can be adversarially manipulated in undesirable ways. In this work, we focus on a popular class of vision-language models (VLMs) that generate text outputs conditioned on visual and textual inputs. We conducted a large-scale empirical study to assess the transferability of gradient-based universal image "jailbreaks" using a diverse set of over 40 open-parameter VLMs, including 18 new VLMs that we publicly release. Overall, we find that transferable gradient-based image jailbreaks are extremely difficult to obtain. When an image jailbreak is optimized against a single VLM or against an ensemble of VLMs, the jailbreak successfully jailbreaks the attacked VLM(s), but exhibits little-to-no transfer to any other VLMs; transfer is not affected by whether the attacked and target VLMs possess matching vision backbones or language models, whether the language model underwent instruction-following and/or safety-alignment training, or many other factors. Only two settings display partially successful transfer: between identically-pretrained and identically-initialized VLMs with slightly different VLM training data, and between different training checkpoints of a single VLM. Leveraging these results, we then demonstrate that transfer can be significantly improved against a specific target VLM by attacking larger ensembles of "highly-similar" VLMs. These results stand in stark contrast to existing evidence of universal and transferable text jailbreaks against language models and transferable adversarial attacks against image classifiers, suggesting that VLMs may be more robust to gradient-based transfer attacks.
Values are a basic driving force underlying human behavior. Large Language Models (LLM) technology is constantly improving towards human-like dialogue. However, little research has been done to study the values exhibited in text generated by LLMs. Here we study this question by turning to the rich literature on value structure in psychology. We ask whether LLMs exhibit the same value structure that has been demonstrated in humans, including the ranking of values, and correlation between values. We show that the results of this analysis strongly depend on how the LLM is prompted, and that under a particular prompting strategy (referred to as 'Value Anchoring') the agreement with human data is quite compelling. Our results serve both to improve our understanding of values in LLMs, as well as introduce novel methods for assessing consistency in LLM responses.
Graph Neural Networks (GNNs) for representation learning of graphs broadly follow a neighborhood aggregation framework, where the representation vector of a node is computed by recursively aggregating and transforming feature vectors of its neighboring nodes. Many GNN variants have been proposed and have achieved state-of-the-art results on both node and graph classification tasks. However, despite GNNs revolutionizing graph representation learning, there is limited understanding of their representational properties and limitations. Here, we present a theoretical framework for analyzing the expressive power of GNNs in capturing different graph structures. Our results characterize the discriminative power of popular GNN variants, such as Graph Convolutional Networks and GraphSAGE, and show that they cannot learn to distinguish certain simple graph structures. We then develop a simple architecture that is provably the most expressive among the class of GNNs and is as powerful as the Weisfeiler-Lehman graph isomorphism test. We empirically validate our theoretical findings on a number of graph classification benchmarks, and demonstrate that our model achieves state-of-the-art performance.
小貼士
登錄享
相關主題
注冊地址： 北京市海淀區羊坊店路18號2幢3層301-191