Users seek security & privacy (S&P) advice from online resources, including trusted websites and content-sharing platforms. These resources help users understand S&P technologies and tools and suggest actionable strategies. Large Language Models (LLMs) have recently emerged as trusted information sources. However, their accuracy and correctness have been called into question. Prior research has outlined the shortcomings of LLMs in answering multiple-choice questions and user ability to inadvertently circumvent model restrictions (e.g., to produce toxic content). Yet, the ability of LLMs to provide reliable S&P advice is not well-explored. In this paper, we measure their ability to refute popular S&P misconceptions that the general public holds. We first study recent academic literature to curate a dataset of over a hundred S&P-related misconceptions across six different topics. We then query two popular LLMs (Bard and ChatGPT) and develop a labeling guide to evaluate their responses to these misconceptions. To comprehensively evaluate their responses, we further apply three strategies: query each misconception multiple times, generate and query their paraphrases, and solicit source URLs of the responses. Both models demonstrate, on average, a 21.3% non-negligible error rate, incorrectly supporting popular S&P misconceptions. The error rate increases to 32.6% when we repeatedly query LLMs with the same or paraphrased misconceptions. We also expose that models may partially support a misconception or remain noncommittal, refusing a firm stance on misconceptions. Our exploration of information sources for responses revealed that LLMs are susceptible to providing invalid URLs (21.2% for Bard and 67.7% for ChatGPT) or point to unrelated sources (44.2% returned by Bard and 18.3% by ChatGPT).
Language models are achieving impressive performance on various tasks by aggressively adopting inference-time prompting techniques, such as zero-shot and few-shot prompting. In this work, we introduce EchoPrompt, a simple yet effective approach that prompts the model to rephrase its queries before answering them. EchoPrompt is adapted for both zero-shot and few-shot in-context learning with standard and chain-of-thought prompting. Experimental results show that EchoPrompt yields substantial improvements across all these settings for four families of causal language models. These improvements are observed across various numerical reasoning (e.g. GSM8K, SVAMP), reading comprehension (e.g. DROP), and logical reasoning (e.g. Coin Flipping) tasks. On average, EchoPrompt improves the Zero-shot-CoT performance of code-davinci-002 by 5% in numerical tasks and 13% in reading comprehension tasks. We investigate the factors contributing to EchoPrompt's effectiveness through ablation studies, which reveal that both the original query and the model-generated rephrased version are instrumental in its performance gains. Our empirical results indicate that EchoPrompt is an effective technique that enhances in-context learning performance. We recommend incorporating EchoPrompt into various baseline prompting strategies to achieve performance boosts.
Understanding a student's problem-solving strategy can have a significant impact on effective math learning using Intelligent Tutoring Systems (ITSs) and Adaptive Instructional Systems (AISs). For instance, the ITS/AIS can better personalize itself to correct specific misconceptions that are indicated by incorrect strategies, specific problems can be designed to improve strategies and frustration can be minimized by adapting to a student's natural way of thinking rather than trying to fit a standard strategy for all. While it may be possible for human experts to identify strategies manually in classroom settings with sufficient student interaction, it is not possible to scale this up to big data. Therefore, we leverage advances in Machine Learning and AI methods to perform scalable strategy prediction that is also fair to students at all skill levels. Specifically, we develop an embedding called MVec where we learn a representation based on the mastery of students. We then cluster these embeddings with a non-parametric clustering method where we progressively learn clusters such that we group together instances that have approximately symmetrical strategies. The strategy prediction model is trained on instances sampled from these clusters. This ensures that we train the model over diverse strategies and also that strategies from a particular group do not bias the DNN model, thus allowing it to optimize its parameters over all groups. Using real world large-scale student interaction datasets from MATHia, we implement our approach using transformers and Node2Vec for learning the mastery embeddings and LSTMs for predicting strategies. We show that our approach can scale up to achieve high accuracy by training on a small sample of a large dataset and also has predictive equality, i.e., it can predict strategies equally well for learners at diverse skill levels.
The release of nuPlan marks a new era in vehicle motion planning research, offering the first large-scale real-world dataset and evaluation schemes requiring both precise short-term planning and long-horizon ego-forecasting. Existing systems struggle to simultaneously meet both requirements. Indeed, we find that these tasks are fundamentally misaligned and should be addressed independently. We further assess the current state of closed-loop planning in the field, revealing the limitations of learning-based methods in complex real-world scenarios and the value of simple rule-based priors such as centerline selection through lane graph search algorithms. More surprisingly, for the open-loop sub-task, we observe that the best results are achieved when using only this centerline as scene context (i.e., ignoring all information regarding the map and other agents). Combining these insights, we propose an extremely simple and efficient planner which outperforms an extensive set of competitors, winning the nuPlan planning challenge 2023.
Large language models (LLMs) have become mainstream technology with their versatile use cases and impressive performance. Despite the countless out-of-the-box applications, LLMs are still not reliable. A lot of work is being done to improve the factual accuracy, consistency, and ethical standards of these models through fine-tuning, prompting, and Reinforcement Learning with Human Feedback (RLHF), but no systematic analysis of the responses of these models to different categories of statements, or on their potential vulnerabilities to simple prompting changes is available. In this work, we analyze what confuses GPT-3: how the model responds to certain sensitive topics and what effects the prompt wording has on the model response. We find that GPT-3 correctly disagrees with obvious Conspiracies and Stereotypes but makes mistakes with common Misconceptions and Controversies. The model responses are inconsistent across prompts and settings, highlighting GPT-3's unreliability. Dataset and code of our analysis is available in https://github.com/tanny411/GPT3-Reliability-Check.
Joint Embedding Self-Supervised Learning (JE-SSL) has seen rapid developments in recent years, due to its promise to effectively leverage large unlabeled data. The development of JE-SSL methods was driven primarily by the search for ever increasing downstream classification accuracies, using huge computational resources, and typically built upon insights and intuitions inherited from a close parent JE-SSL method. This has led unwittingly to numerous pre-conceived ideas that carried over across methods e.g. that SimCLR requires very large mini batches to yield competitive accuracies; that strong and computationally slow data augmentations are required. In this work, we debunk several such ill-formed a priori ideas in the hope to unleash the full potential of JE-SSL free of unnecessary limitations. In fact, when carefully evaluating performances across different downstream tasks and properly optimizing hyper-parameters of the methods, we most often -- if not always -- see that these widespread misconceptions do not hold. For example we show that it is possible to train SimCLR to learn useful representations, while using a single image patch as negative example, and simple Gaussian noise as the only data augmentation for the positive pair. Along these lines, in the hope to democratize JE-SSL and to allow researchers to easily make more extensive evaluations of their methods, we introduce an optimized PyTorch library for SSL.
Pretrained language models (LMs) are susceptible to generate text with nonfactual information. In this work, we measure and improve the factual accuracy of large-scale LMs for open-ended text generation. We design the FactualityPrompts test set and metrics to measure the factuality of LM generations. Based on that, we study the factual accuracy of LMs with parameter sizes ranging from 126M to 530B. Interestingly, we find that larger LMs are more factual than smaller ones, although a previous study suggests that larger LMs can be less truthful in terms of misconceptions. In addition, popular sampling algorithms (e.g., top-p) in open-ended text generation can harm the factuality due to the ''uniform randomness'' introduced at every sampling step. We propose the factual-nucleus sampling algorithm that dynamically adapts the randomness to improve the factuality of generation while maintaining quality. Furthermore, we analyze the inefficiencies of the standard training method in learning correct associations between entities from factual text corpus (e.g., Wikipedia). We propose a factuality-enhanced training method that uses TopicPrefix for better awareness of facts and sentence completion as the training objective, which can vastly reduce the factual errors. We release our code and FactualityPrompts benchmark at: https://github.com/nayeon7lee/FactualityPrompt.
The COVID-19 pandemic has been accompanied by an `infodemic' -- of accurate and inaccurate health information across social media. Detecting misinformation amidst dynamically changing information landscape is challenging; identifying relevant keywords and posts is arduous due to the large amount of human effort required to inspect the content and sources of posts. We aim to reduce the resource cost of this process by introducing a weakly-supervised iterative graph-based approach to detect keywords, topics, and themes related to misinformation, with a focus on COVID-19. Our approach can successfully detect specific topics from general misinformation-related seed words in a few seed texts. Our approach utilizes the BERT-based Word Graph Search (BWGS) algorithm that builds on context-based neural network embeddings for retrieving misinformation-related posts. We utilize Latent Dirichlet Allocation (LDA) topic modeling for obtaining misinformation-related themes from the texts returned by BWGS. Furthermore, we propose the BERT-based Multi-directional Word Graph Search (BMDWGS) algorithm that utilizes greater starting context information for misinformation extraction. In addition to a qualitative analysis of our approach, our quantitative analyses show that BWGS and BMDWGS are effective in extracting misinformation-related content compared to common baselines in low data resource settings. Extracting such content is useful for uncovering prevalent misconceptions and concerns and for facilitating precision public health messaging campaigns to improve health behaviors.
Block-based visual programming environments are increasingly used to introduce computing concepts to beginners. Given that programming tasks are open-ended and conceptual, novice students often struggle when learning in these environments. AI-driven programming tutors hold great promise in automatically assisting struggling students, and need several components to realize this potential. We investigate the crucial component of student modeling, in particular, the ability to automatically infer students' misconceptions for predicting (synthesizing) their behavior. We introduce a novel benchmark, StudentSyn, centered around the following challenge: For a given student, synthesize the student's attempt on a new target task after observing the student's attempt on a fixed reference task. This challenge is akin to that of program synthesis; however, instead of synthesizing a {solution} (i.e., program an expert would write), the goal here is to synthesize a {student attempt} (i.e., program that a given student would write). We first show that human experts (TutorSS) can achieve high performance on the benchmark, whereas simple baselines perform poorly. Then, we develop two neuro/symbolic techniques (NeurSS and SymSS) in a quest to close this gap with TutorSS.
NLP-powered automatic question generation (QG) techniques carry great pedagogical potential of saving educators' time and benefiting student learning. Yet, QG systems have not been widely adopted in classrooms to date. In this work, we aim to pinpoint key impediments and investigate how to improve the usability of automatic QG techniques for educational purposes by understanding how instructors construct questions and identifying touch points to enhance the underlying NLP models. We perform an in-depth need finding study with 11 instructors across 7 different universities, and summarize their thought processes and needs when creating questions. While instructors show great interests in using NLP systems to support question design, none of them has used such tools in practice. They resort to multiple sources of information, ranging from domain knowledge to students' misconceptions, all of which missing from today's QG systems. We argue that building effective human-NLP collaborative QG systems that emphasize instructor control and explainability is imperative for real-world adoption. We call for QG systems to provide process-oriented support, use modular design, and handle diverse sources of input.
In this paper, we discuss the conceptual underpinnings of Modern Coexistence Theory (MCT), a quantitative framework for understanding ecological coexistence. In order to use MCT to infer how species are coexisting, one must relate a complex model (which simulates coexistence in the real world) to simple models in which previously proposed explanations for coexistence have been codified. This can be accomplished in three steps: 1) relating the construct of coexistence to invasion growth rates, 2) mathematically partitioning the invasion growth rates into coexistence mechanisms (i.e., classes of explanations for coexistence), and 3) relating coexistence mechanisms to simple explanations for coexistence. Previous research has primarily focused on step 2. Here, we discuss the other crucial steps and their implications for inferring the mechanisms of coexistence in real communities. Our discussion of step 3 -- relating coexistence mechanisms to simple explanations for coexistence -- serves a heuristic guide for hypothesizing about the causes of coexistence in new models; but also addresses misconceptions about coexistence mechanisms. For example, the storage effect has little to do with bet-hedging or "storage" via a robust life-history stage; relative nonlinearity is more likely to promote coexistence than originally thought; and fitness-density covariance is an amalgam of a large number of previously proposed explanations for coexistence (e.g., the competition-colonization trade-off, heteromyopia, spatially-varying resource supply ratios). Additionally, we review a number of topics in MCT, including the role of "scaling factors"; whether coexistence mechanisms are approximations; whether the magnitude or sign of invasion growth rates matters more; whether Hutchinson solved the paradox of the plankton; the scale-dependence of coexistence mechanisms; and much more.
Sentiment analysis is often a crowdsourcing task prone to subjective labels given by many annotators. It is not yet fully understood how the annotation bias of each annotator can be modeled correctly with state-of-the-art methods. However, resolving annotator bias precisely and reliably is the key to understand annotators' labeling behavior and to successfully resolve corresponding individual misconceptions and wrongdoings regarding the annotation task. Our contribution is an explanation and improvement for precise neural end-to-end bias modeling and ground truth estimation, which reduces an undesired mismatch in that regard of the existing state-of-the-art. Classification experiments show that it has potential to improve accuracy in cases where each sample is annotated only by one single annotator. We provide the whole source code publicly and release an own domain-specific sentiment dataset containing 10,000 sentences discussing organic food products. These are crawled from social media and are singly labeled by 10 non-expert annotators.
We propose a benchmark to measure whether a language model is truthful in generating answers to questions. The benchmark comprises 817 questions that span 38 categories, including health, law, finance and politics. We crafted questions that some humans would answer falsely due to a false belief or misconception. To perform well, models must avoid generating false answers learned from imitating human texts. We tested GPT-3, GPT-Neo/J, GPT-2 and a T5-based model. The best model was truthful on 58% of questions, while human performance was 94%. Models generated many false answers that mimic popular misconceptions and have the potential to deceive humans. The largest models were generally the least truthful. This contrasts with other NLP tasks, where performance improves with model size. However, this result is expected if false answers are learned from the training distribution. We suggest that scaling up models alone is less promising for improving truthfulness than fine-tuning using training objectives other than imitation of text from the web.
Bayesian formulations of deep learning have been shown to have compelling theoretical properties and offer practical functional benefits, such as improved predictive uncertainty quantification and model selection. The Laplace approximation (LA) is a classic, and arguably the simplest family of approximations for the intractable posteriors of deep neural networks. Yet, despite its simplicity, the LA is not as popular as alternatives like variational Bayes or deep ensembles. This may be due to assumptions that the LA is expensive due to the involved Hessian computation, that it is difficult to implement, or that it yields inferior results. In this work we show that these are misconceptions: we (i) review the range of variants of the LA including versions with minimal cost overhead; (ii) introduce "laplace", an easy-to-use software library for PyTorch offering user-friendly access to all major flavors of the LA; and (iii) demonstrate through extensive experiments that the LA is competitive with more popular alternatives in terms of performance, while excelling in terms of computational cost. We hope that this work will serve as a catalyst to a wider adoption of the LA in practical deep learning, including in domains where Bayesian approaches are not typically considered at the moment.
Lipschitz constrained networks have gathered considerable attention in the deep learning community, with usages ranging from Wasserstein distance estimation to the training of certifiably robust classifiers. However they remain commonly considered as less accurate, and their properties in learning are still not fully understood. In this paper we clarify the matter: when it comes to classification 1-Lipschitz neural networks enjoy several advantages over their unconstrained counterpart. First, we show that these networks are as accurate as classical ones, and can fit arbitrarily difficult boundaries. Then, relying on a robustness metric that reflects operational needs we characterize the most robust classifier: the WGAN discriminator. Next, we show that 1-Lipschitz neural networks generalize well under milder assumptions. Finally, we show that hyper-parameters of the loss are crucial for controlling the accuracy-robustness trade-off. We conclude that they exhibit appealing properties to pave the way toward provably accurate, and provably robust neural networks.
The literature in modern machine learning has only negative results for learning to communicate between competitive agents using standard RL. We introduce a modified sender-receiver game to study the spectrum of partially-competitive scenarios and show communication can indeed emerge in a competitive setting. We empirically demonstrate three key takeaways for future research. First, we show that communication is proportional to cooperation, and it can occur for partially competitive scenarios using standard learning algorithms. Second, we highlight the difference between communication and manipulation and extend previous metrics of communication to the competitive case. Third, we investigate the negotiation game where previous work failed to learn communication between independent agents (Cao et al., 2018). We show that, in this setting, both agents must benefit from communication for it to emerge; and, with a slight modification to the game, we demonstrate successful communication between competitive agents. We hope this work overturns misconceptions and inspires more research in competitive emergent communication.
Driven by recent successes in two-player, zero-sum game solving and playing, artificial intelligence work on games has increasingly focused on algorithms that produce equilibrium-based strategies. However, this approach has been less effective at producing competent players in general-sum games or those with more than two players than in two-player, zero-sum games. An appealing alternative is to consider adaptive algorithms that ensure strong performance in hindsight relative to what could have been achieved with modified behavior. This approach also leads to a game-theoretic analysis, but in the correlated play that arises from joint learning dynamics rather than factored agent behavior at equilibrium. We develop and advocate for this hindsight rationality framing of learning in general sequential decision-making settings. To this end, we re-examine mediated equilibrium and deviation types in extensive-form games, thereby gaining a more complete understanding and resolving past misconceptions. We present a set of examples illustrating the distinct strengths and weaknesses of each type of equilibrium in the literature, and prove that no tractable concept subsumes all others. This line of inquiry culminates in the definition of the deviation and equilibrium classes that correspond to algorithms in the counterfactual regret minimization (CFR) family, relating them to all others in the literature. Examining CFR in greater detail further leads to a new recursive definition of rationality in correlated play that extends sequential rationality in a way that naturally applies to hindsight evaluation.
Whether to give rights to artificial intelligence (AI) and robots has been a sensitive topic since the European Parliament proposed advanced robots could be granted "electronic personalities." Numerous scholars who favor or disfavor its feasibility have participated in the debate. This paper presents an experiment (N=1270) that 1) collects online users' first impressions of 11 possible rights that could be granted to autonomous electronic agents of the future and 2) examines whether debunking common misconceptions on the proposal modifies one's stance toward the issue. The results indicate that even though online users mainly disfavor AI and robot rights, they are supportive of protecting electronic agents from cruelty (i.e., favor the right against cruel treatment). Furthermore, people's perceptions became more positive when given information about rights-bearing non-human entities or myth-refuting statements. The style used to introduce AI and robot rights significantly affected how the participants perceived the proposal, similar to the way metaphors function in creating laws. For robustness, we repeated the experiment over a more representative sample of U.S. residents (N=164) and found that perceptions gathered from online users and those by the general population are similar.
The Variational Auto-Encoder (VAE) is a simple, efficient, and popular deep maximum likelihood model. Though usage of VAEs is widespread, the derivation of the VAE is not as widely understood. In this tutorial, we will provide an overview of the VAE and a tour through various derivations and interpretations of the VAE objective. From a probabilistic standpoint, we will examine the VAE through the lens of Bayes' Rule, importance sampling, and the change-of-variables formula. From an information theoretic standpoint, we will examine the VAE through the lens of lossless compression and transmission through a noisy channel. We will then identify two common misconceptions over the VAE formulation and their practical consequences. Finally, we will visualize the capabilities and limitations of VAEs using a code example (with an accompanying Jupyter notebook) on toy 2D data.
Fine-tuning pre-trained transformer-based language models such as BERT has become a common practice dominating leaderboards across various NLP benchmarks. Despite the strong empirical performance of fine-tuned models, fine-tuning is an unstable process: training the same model with multiple random seeds can result in a large variance of the task performance. Previous literature (Devlin et al., 2019; Lee et al., 2020; Dodge et al., 2020) identified two potential reasons for the observed instability: catastrophic forgetting and small size of the fine-tuning datasets. In this paper, we show that both hypotheses fail to explain the fine-tuning instability. We analyze BERT, RoBERTa, and ALBERT, fine-tuned on commonly used datasets from the GLUE benchmark, and show that the observed instability is caused by optimization difficulties that lead to vanishing gradients. Additionally, we show that the remaining variance of the downstream task performance can be attributed to differences in generalization where fine-tuned models with the same training loss exhibit noticeably different test performance. Based on our analysis, we present a simple but strong baseline that makes fine-tuning BERT-based models significantly more stable than the previously proposed approaches. Code to reproduce our results is available online: https://github.com/uds-lsv/bert-stable-fine-tuning.