Despite the importance of developing generative AI models that can effectively resist scams, current literature lacks a structured framework for evaluating their vulnerability to such threats. In this work, we address this gap by constructing a benchmark based on the FINRA taxonomy and systematically assessing Large Language Models' (LLMs') vulnerability to a variety of scam tactics. First, we incorporate 37 well-defined base scam scenarios reflecting the diverse scam categories identified by FINRA taxonomy, providing a focused evaluation of LLMs' scam detection capabilities. Second, we utilize representative proprietary (GPT-3.5, GPT-4) and open-source (Llama) models to analyze their performance in scam detection. Third, our research provides critical insights into which scam tactics are most effective against LLMs and how varying persona traits and persuasive techniques influence these vulnerabilities. We reveal distinct susceptibility patterns across different models and scenarios, underscoring the need for targeted enhancements in LLM design and deployment.

In this paper, we focus on estimating the causal effect of an intervention over time on a dynamical system. To that end, we formally define causal interventions and their effects over time on discrete-time stochastic processes (DSPs). Then, we show under which conditions the equilibrium states of a DSP, both before and after a causal intervention, can be captured by a structural causal model (SCM). With such an equivalence at hand, we provide an explicit mapping from vector autoregressive models (VARs), broadly applied in econometrics, to linear, but potentially cyclic and/or affected by unmeasured confounders, SCMs. The resulting causal VAR framework allows us to perform causal inference over time from observational time series data. Our experiments on synthetic and real-world datasets show that the proposed framework achieves strong performance in terms of observational forecasting while enabling accurate estimation of the causal effect of interventions on dynamical systems. We demonstrate, through a case study, the potential practical questions that can be addressed using the proposed causal VAR framework.

Estimating ideological positions of lawmakers has a long tradition in political science. Poole & Rosenthal (1985) proposed a "scaling procedure" to estimate ideological positions of lawmakers based on their voting behavior. Dynamic weighted nominal three-step estimation (McCarty et al. 1997), an extension of this procedure, results in the DW-Nominate scores that are widely accepted as benchmark ideological positions both on party level as well as on individual level (see, e.g., Poole et al. 2011, Lewis et al. 2022, Boche et al. 2018). Legislative votes, however, provide limited information on the latent ideological positions because voting behavior on individual level is often not documented and lawmakers rarely diverge from party-line voting due to robust party discipline (Hug 2010). Consequently, roll-call analysis for inferring the ideological positions adopted by legislators both within and across parties is of limited value (see, e.g., Lauderdale & Herzog 2016). Text-based scaling models are a promising alternative method to discern ideological stances based on political discussions.

Deep learning models have become crucial for tackling complex computational problems, owing to the rich representations they develop through their multi-layered structures and non-linear transformations [1, 2]. Despite their remarkable effectiveness, these models are often perceived as black boxes, raising concerns related to their robustness, reliability, and safety. As neural networks become increasingly integral to critical applications, ensuring that they are properly trained and perform as intended is paramount. To evaluate the training performance and a network's ability to store a model after training (i.e., achieve zero loss), one approach is to statistically analyze neural networks under certain assumptions. This has been done for networks with thresholding activation functions like ReLU, where researchers have determined the number of parameters needed to achieve full memory capacity [3]. It is well-known that for ReLU-based neural networks (NNs), once a sufficient number of weights is reached, the network can achieve full memory capacity or even zero loss in some cases. In [4], the authors theoretically demonstrate that in the over-parameterization regime, the stochastic gradient descent (SGD) algorithm can converge to the global minimum. However, these methods are statistical in nature and rely on specific assumptions about the input data and the model, which may limit their applicability. The first two authors contributed equally to this work.

In the absence of abundant reliable annotations for challenging tasks and contexts, how can we expand the frontier of LLM capabilities with potentially wrong answers? We focus on two research questions: (1) Can LLMs generate reliable preferences among wrong options? And if so, (2) Would alignment with such wrong-over-wrong preferences be helpful? We employ methods based on self-consistency, token probabilities, and LLM-as-a-judge to elicit wrong-over-wrong preferences, and fine-tune language models with preference optimization approaches using these synthesized preferences. Extensive experiments with seven LLMs and eight datasets demonstrate that (1) LLMs do have preliminary capability in distinguishing various shades of wrong, achieving up to 20.9% higher performance than random guess; (2) Alignment with wrong-over-wrong preferences helps LLMs to produce less wrong and sometimes even outright correct answers, while overall improving model calibration.

Designing a cognitive radar system capable of adapting its parameters is challenging, particularly when tasked with tracking a ballistic missile throughout its entire flight. In this work, we focus on proposing adaptive algorithms that select waveform parameters in an online fashion. Our novelty lies in formulating the learning problem using domain knowledge derived from the characteristics of ballistic trajectories. We propose three reinforcement learning algorithms: bandwidth scaling, Q-learning, and Q-learning lookahead. These algorithms dynamically choose the bandwidth for each transmission based on received feedback. Through experiments on synthetically generated ballistic trajectories, we demonstrate that our proposed algorithms achieve the dual objectives of minimizing range error and maintaining continuous tracking without losing the target.

Fine-tuning large language models (LLMs) on additional datasets is often necessary to optimize them for specific downstream tasks. However, existing safety alignment measures, which restrict harmful behavior during inference, are insufficient to mitigate safety risks during fine-tuning. Alarmingly, fine-tuning with just 10 toxic sentences can make models comply with harmful instructions. We introduce SafetyLock, a novel alignment intervention method that maintains robust safety post-fine-tuning through efficient and transferable mechanisms. SafetyLock leverages our discovery that fine-tuned models retain similar safety-related activation representations to their base models. This insight enables us to extract what we term the Meta-SafetyLock, a set of safety bias directions representing key activation patterns associated with safe responses in the original model. We can then apply these directions universally to fine-tuned models to enhance their safety. By searching for activation directions across multiple token dimensions, SafetyLock achieves enhanced robustness and transferability. Our experiments demonstrate that SafetyLock can reduce the harmful instruction response rate from 60% to below 1% in toxic fine-tuned models. It surpasses traditional methods in both performance and efficiency, offering a scalable, non-invasive solution for ensuring the safety of customized LLMs. Our analysis across various fine-tuning scenarios confirms SafetyLock's robustness, advocating its integration into safety protocols for aligned LLMs. Large language models (LLMs) have demonstrated increasing utility across various domains (Wei et al., 2022b;a; Weng et al., 2023; Hadar-Shoval et al., 2024), yet their potential to handle harmful queries has raised significant concerns (Carroll et al., 2023; Hendrycks et al., 2023). These measures are expected to teach models to refuse harmful queries during inference (Huang et al., 2024b; Wang et al., 2024b; Raza et al., 2024; Zou et al., 2024).

The training data for LLMs embeds societal values, increasing their familiarity with the language's culture. Our analysis found that 44% of the variance in the ability of GPT-4o to reflect the societal values of a country, as measured by the World Values Survey, correlates with the availability of digital resources in that language. Notably, the error rate was more than five times higher for the languages of the lowest resource compared to the languages of the highest resource. For GPT-4-turbo, this correlation rose to 72%, suggesting efforts to improve the familiarity with the non-English language beyond the web-scraped data. Our study developed one of the largest and most robust datasets in this topic area with 21 country-language pairs, each of which contain 94 survey questions verified by native speakers. Our results highlight the link between LLM performance and digital data availability in target languages. Weaker performance in low-resource languages, especially prominent in the Global South, may worsen digital divides. We discuss strategies proposed to address this, including developing multilingual LLMs from the ground up and enhancing fine-tuning on diverse linguistic datasets, as seen in African language initiatives.

Artificial Intelligence (AI), particularly large language models (LLMs), holds the potential to bridge language and information gaps, which can benefit the economies of developing nations. However, our analysis of FLORES-200, FLORES+, Ethnologue, and World Development Indicators data reveals that these benefits largely favor English speakers. Speakers of languages in low-income and lower-middle-income countries face higher costs when using OpenAI's GPT models via APIs because of how the system processes the input -- tokenization. Around 1.5 billion people, speaking languages primarily from lower-middle-income countries, could incur costs that are 4 to 6 times higher than those faced by English speakers. Disparities in LLM performance are significant, and tokenization in models priced per token amplifies inequalities in access, cost, and utility. Moreover, using the quality of translation tasks as a proxy measure, we show that LLMs perform poorly in low-resource languages, presenting a ``double jeopardy" of higher costs and poor performance for these users. We also discuss the direct impact of fragmentation in tokenizing low-resource languages on climate. This underscores the need for fairer algorithm development to benefit all linguistic groups.

This study exposes the safety vulnerabilities of Large Language Models (LLMs) in multi-turn interactions, where malicious users can obscure harmful intents across several queries. We introduce ActorAttack, a novel multi-turn attack method inspired by actor-network theory, which models a network of semantically linked actors as attack clues to generate diverse and effective attack paths toward harmful targets. ActorAttack addresses two main challenges in multi-turn attacks: (1) concealing harmful intents by creating an innocuous conversation topic about the actor, and (2) uncovering diverse attack paths towards the same harmful target by leveraging LLMs' knowledge to specify the correlated actors as various attack clues. In this way, ActorAttack outperforms existing single-turn and multi-turn attack methods across advanced aligned LLMs, even for GPT-o1. We will publish a dataset called SafeMTData, which includes multi-turn adversarial prompts and safety alignment data, generated by ActorAttack. We demonstrate that models safety-tuned using our safety dataset are more robust to multi-turn attacks. Code is available at https://github.com/renqibing/ActorAttack.