arXiv subjects

Zeynab Samei

Publications and source records attributed to Zeynab Samei.

6 recordsLinked to original sources

arXiv2026

Unsupervised Elicitation of Moral Values from Language Models

As AI systems become pervasive, grounding their behavior in human values is critical. Prior work suggests that language models (LMs) exhibit limited inherent moral reasoning, leading to calls for explicit moral teaching. However, constructing ground truth data for moral evaluation is difficult given plural frameworks and pervasive biases. We investigate unsupervised elicitation as an alternative, asking whether pretrained (base) LMs possess intrinsic moral reasoning capability that can be surfaced without human supervision. Using the Internal Coherence Maximization (ICM) algorithm across three benchmark datasets and four LMs, we test whether ICM can reliably label moral judgments, generalize across moral frameworks, and mitigate social bias. Results show that ICM outperforms all pre-trained and chatbot baselines on the Norm Bank and ETHICS benchmarks, while fine-tuning on ICM labels performs on par with or surpasses those of human labels. Across theoretically motivated moral frameworks, ICM yields its largest relative gains on Justice and Commonsense morality. Furthermore, although chatbot LMs exhibit social bias failure rates comparable to their pretrained ones, ICM reduces such errors by more than half, with the largest improvements in race, socioeconomic status, and politics. These findings suggest that pretrained LMs possess latent moral reasoning capacities that can be elicited through unsupervised methods like ICM, providing a scalable path for AI alignment.

cs.CL↗

arXiv2025

Web-Browsing LLMs Can Access Social Media Profiles and Infer User Demographics

Large language models (LLMs) have traditionally relied on static training data, limiting their knowledge to fixed snapshots. Recent advancements, however, have equipped LLMs with web browsing capabilities, enabling real time information retrieval and multi step reasoning over live web content. While prior studies have demonstrated LLMs ability to access and analyze websites, their capacity to directly retrieve and analyze social media data remains unexplored. Here, we evaluate whether web browsing LLMs can infer demographic attributes of social media users given only their usernames. Using a synthetic dataset of 48 X (Twitter) accounts and a survey dataset of 1,384 international participants, we show that these models can access social media content and predict user demographics with reasonable accuracy. Analysis of the synthetic dataset further reveals how LLMs parse and interpret social media profiles, which may introduce gender and political biases against accounts with minimal activity. While this capability holds promise for computational social science in the post API era, it also raises risks of misuse particularly in information operations and targeted advertising underscoring the need for safeguards. We recommend that LLM providers restrict this capability in public facing applications, while preserving controlled access for verified research purposes.

cs.CL↗

arXiv2025

Simple Prompt Injection Attacks Can Leak Personal Data Observed by LLM Agents During Task Execution

Previous benchmarks on prompt injection in large language models (LLMs) have primarily focused on generic tasks and attacks, offering limited insights into more complex threats like data exfiltration. This paper examines how prompt injection can cause tool-calling agents to leak personal data observed during task execution. Using a fictitious banking agent, we develop data flow-based attacks and integrate them into AgentDojo, a recent benchmark for agentic security. To enhance its scope, we also create a richer synthetic dataset of human-AI banking conversations. In 16 user tasks from AgentDojo, LLMs show a 15-50 percentage point drop in utility under attack, with average attack success rates (ASR) around 20 percent; some defenses reduce ASR to zero. Most LLMs, even when successfully tricked by the attack, avoid leaking highly sensitive data like passwords, likely due to safety alignments, but they remain vulnerable to disclosing other personal data. The likelihood of password leakage increases when a password is requested along with one or two additional personal details. In an extended evaluation across 48 tasks, the average ASR is around 15 percent, with no built-in AgentDojo defense fully preventing leakage. Tasks involving data extraction or authorization workflows, which closely resemble the structure of exfiltration attacks, exhibit the highest ASRs, highlighting the interaction between task type, agent performance, and defense efficacy.

cs.CR↗

arXiv2024

Exploring Relationships Between Cryptocurrency News Outlets and Influencers' Twitter Activity and Market Prices

Academics increasingly acknowledge the predictive power of social media for a wide variety of events and, more specifically, for financial markets. Anecdotal and empirical findings show that cryptocurrencies are among the financial assets that have been affected by news and influencers' activities on Twitter. However, the extent to which Twitter crypto influencer's posts about trading signals and their effect on market prices is mostly unexplored. In this paper, we use LLMs to uncover buy and not-buy signals from influencers and news outlets' Twitter posts and use a VAR analysis with Granger Causality tests and cross-correlation analysis to understand how these trading signals are temporally correlated with the top nine major cryptocurrencies' prices. Overall, the results show a mixed pattern across cryptocurrencies and temporal periods. However, we found that for the top three cryptocurrencies with the highest presence within news and influencer posts, their aggregated LLM-detected trading signal over the preceding 24 hours granger-causes fluctuations in their market prices, exhibiting a lag of at least 6 hours. In addition, the results reveal fundamental differences in how influencers and news outlets cover cryptocurrencies.

cs.SI↗

arXiv2024

Comparing Methods for Creating a National Random Sample of Twitter Users

Twitter data has been widely used by researchers across various social and computer science disciplines. A common aim when working with Twitter data is the construction of a random sample of users from a given country. However, while several methods have been proposed in the literature, their comparative performance is mostly unexplored. In this paper, we implement four common methods to collect a random sample of Twitter users in the US: 1% Stream, Bounding Box, Location Query, and Language Query. Then, we compare the methods according to their tweet- and user-level metrics as well as their accuracy in estimating US population with and without using inclusion probabilities of various demographics. Our results show that the 1% Stream method performs differently than others in tweet- and user-level metrics, and best for the construction of a population representative sample. We discuss the conditions under which the 1% Stream method may not be suitable and suggest the Bounding Box method as the second-best method to use.

cs.SI↗

arXiv2023

Open-Source LLMs for Text Annotation: A Practical Guide for Model Setting and Fine-Tuning

This paper studies the performance of open-source Large Language Models (LLMs) in text classification tasks typical for political science research. By examining tasks like stance, topic, and relevance classification, we aim to guide scholars in making informed decisions about their use of LLMs for text analysis. Specifically, we conduct an assessment of both zero-shot and fine-tuned LLMs across a range of text annotation tasks using news articles and tweets datasets. Our analysis shows that fine-tuning improves the performance of open-source LLMs, allowing them to match or even surpass zero-shot GPT-3.5 and GPT-4, though still lagging behind fine-tuned GPT-3.5. We further establish that fine-tuning is preferable to few-shot training with a relatively modest quantity of annotated text. Our findings show that fine-tuned open-source LLMs can be effectively deployed in a broad spectrum of text annotation applications. We provide a Python notebook facilitating the application of LLMs in text annotation for other researchers.

cs.CL↗