Kamal Premaratne

Large language models (LLMs) exhibit strong generative capabilities but remain vulnerable to confabulations, fluent yet unreliable outputs that vary arbitrarily even under identical prompts. Leveraging a quantum tensor network based pipeline, we propose a quantum physics inspired uncertainty quantification framework that accounts for aleatoric uncertainty in token sequence probability for semantic equivalence based clustering of LLM generations. This offers a principled and interpretable scheme for hallucination detection. We further introduce an entropy maximization strategy that prioritizes high certainty, semantically coherent outputs and highlights entropy regions where LLM decisions are likely to be unreliable, offering practical guidelines for when human oversight is warranted. We evaluate the robustness of our scheme under different generation lengths and quantization levels, dimensions overlooked in prior studies, demonstrating that our approach remains reliable even in resource constrained deployments. A total of 116 experiments on TriviaQA, NQ, SVAMP, and SQuAD across multiple architectures including Mistral-7B, Mistral-7B-instruct, Falcon-rw-1b, LLaMA-3.2-1b, LLaMA-2-13b-chat, LLaMA-2-7b-chat, LLaMA-2-13b, and LLaMA-2-7b show consistent improvements in AUROC and AURAC over state of the art baselines.

During the 2020 U.S. presidential election cycle, a combination of public statements and social media posts cast doubt on the legitimacy of the election. These sentiments flowed through various social networks and eventually sparked the January 6th insurrection at the Capitol. Here, we analyze both the network-level and content-level data that made the #StopTheSteal movement so effective online. We use Louvain clustering and a novel homogeneity metric to identify the most ideologically homogeneous groups within the discussion on the mainstream social network Twitter and alternative social network Parler. We show that these ideologically homogeneous groups spread messages further than their ideologically diverse counterparts. Our results also differentiate between ideologically homogeneous left- and right-leaning groups by measuring the characteristics of their texts, finding that right-leaning texts are stylistically similar to worldbuilding language that can be found in conspiracy theory texts.

The public is convinced that beliefs in conspiracy theories are increasing, and many scholars, journalists, and policymakers agree. Given the associations between conspiracy theories and many non-normative tendencies, lawmakers have called for policies to address these increases. However, little evidence has been provided to demonstrate that beliefs in conspiracy theories have, in fact, increased over time. We address this evidentiary gap. Study 1 investigates change in the proportion of Americans believing 46 conspiracy theories; our observations in some instances span half a century. Study 2 examines change in the proportion of individuals across six European countries believing six conspiracy theories. Study 3 traces beliefs about which groups are conspiring against “us,” while Study 4 tracks generalized conspiracy thinking in the U.S. from 2012 to 2021. In no instance do we observe systematic evidence for an increase in conspiracism, however operationalized. We discuss the theoretical and policy implications of our findings.

Contemporary political ills at the mass behavior level (e.g., outgroup aggression, conspiracy theories) are often attributed to increasing polarization and partisan tribalism. We theorize that many such problems are less the product of left‐right orientations than an orthogonal “anti‐establishment” dimension of opinion dominated by conspiracy, populist, and Manichean orientations. Using two national surveys from 2019 and 2020, we find that this dimension of opinion is correlated with several antisocial psychological traits, the acceptance of political violence, and time spent on extremist social media platforms. It is also related to support for populist candidates, such as Trump and Sanders, and beliefs in misinformation and conspiracy theories. While many inherently view politics as a conflict between left and right, others see it as a battle between “the people” and a corrupt establishment. Our findings demonstrate an urgent need to expand the traditional conceptualization of mass opinion beyond familiar left‐right identities and affective orientations.

Numerous studies find associations between social media use and beliefs in conspiracy theories and misinformation. While such findings are often interpreted as evidence that social media causally promotes conspiracy beliefs, we theorize that this relationship is conditional on other individual-level predispositions. Across two studies, we examine the relationship between beliefs in conspiracy theories and media use, finding that individuals who get their news from social media and use social media frequently express more beliefs in some types of conspiracy theories and misinformation. However, we also find that these relationships are conditional on conspiracy thinking––the predisposition to interpret salient events as products of conspiracies––such that social media use becomes more strongly associated with conspiracy beliefs as conspiracy thinking intensifies. This pattern, which we observe across many beliefs from two studies, clarifies the relationship between social media use and beliefs in dubious ideas.

The Dempster-Shafer (DS) belief theory constitutes a powerful framework for modeling and reasoning with a wide variety of uncertainties due to its greater expressiveness and flexibility. As in the Bayesian probability theory, the DS theoretic (DST) conditional plays a pivotal role in DST strategies for evidence updating and fusion. However, a major limitation in employing the DST framework in practical implementations is the absence of an efficient and feasible computational framework to overcome the prohibitive computational burden DST operations entail. The work in this article addresses the pressing need for efficient DST conditional computation via the novel computational model DS-Conditional-All. It requires significantly less time and space complexity for computing the Dempster's conditional and the Fagin-Halpern conditional, the two most widely utilized DST conditional strategies. It also provides deeper insight into the DST conditional itself, and thus acts as a valuable tool for visualizing and analyzing the conditional computation. We provide a thorough analysis and experimental validation of the utility, efficiency, and implementation of the proposed data structure and algorithms. A new computational library, which we refer to as DS-Conditional-One and DS-Conditional-All (DS-COCA), is developed and harnessed in the simulations.

The problem of inferring the hidden state of individual nodes in social/sensor networks in which node activities affect their neighbors is growing in importance. We present an undirected generative model, a type of probabilistic model that has so far not been used for modeling latent variables influenced by neighbors in a network. We also propose an efficient inference method based on variational inference principles which, in contrast to sampling methods used in most existing models, is scalable to larger networks. While training is intractable in general, by using stochastic methods to approximate the intractable derivative, we show that our model can be trained using the maximum likelihood method by formulating the model as an exponential family distribution. The results demonstrate that the proposed undirected model can accurately infer latent states compared to baseline methods.