arXiv subjects

Zohar Elyoseph

Publications and source records attributed to Zohar Elyoseph.

4 recordsLinked to original sources

arXiv2025

State over Tokens: Characterizing the Role of Reasoning Tokens

Large Language Models (LLMs) can generate reasoning tokens before their final answer to boost performance on complex tasks. While these sequences seem like human thought processes, empirical evidence reveals that they are not a faithful explanation of the model's actual reasoning process. To address this gap between appearance and function, we introduce the State over Tokens (SoT) conceptual framework. SoT reframes reasoning tokens not as a linguistic narrative, but as an externalized computational state -- the sole persistent information carrier across the model's stateless generation cycles. This explains how the tokens can drive correct reasoning without being a faithful explanation when read as text and surfaces previously overlooked research questions on these tokens. We argue that to truly understand the process that LLMs do, research must move beyond reading the reasoning tokens as text and focus on decoding them as state.

cs.CL↗

arXiv2025

Do AI Voices Learn Social Nuances? A Case of Politeness and Speech Rate

Voice-based artificial intelligence is increasingly expected to adhere to human social conventions, but can it learn implicit cues that are not explicitly programmed? This study investigates whether state-of-the-art text-to-speech systems have internalized the human tendency to reduce speech rate to convey politeness - a non-obvious prosodic marker. We prompted 22 synthetic voices from two leading AI platforms (AI Studio and OpenAI) to read a fixed script under both "polite and formal" and "casual and informal" conditions and measured the resulting speech duration. Across both AI platforms, the polite prompt produced slower speech than the casual prompt with very large effect sizes, an effect that was statistically significant for all of AI Studio's voices and for a large majority of OpenAI's voices. These results demonstrate that AI can implicitly learn and replicate psychological nuances of human communication, highlighting its emerging role as a social actor capable of reinforcing human social norms.

cs.CL↗

arXiv2025

Inducing State Anxiety in LLM Agents Reproduces Human-Like Biases in Consumer Decision-Making

Large language models (LLMs) are rapidly evolving from text generators to autonomous agents, raising urgent questions about their reliability in real-world contexts. Stress and anxiety are well known to bias human decision-making, particularly in consumer choices. Here, we tested whether LLM agents exhibit analogous vulnerabilities. Three advanced models (ChatGPT-5, Gemini 2.5, Claude 3.5-Sonnet) performed a grocery shopping task under budget constraints (24, 54, 108 USD), before and after exposure to anxiety-inducing traumatic narratives. Across 2,250 runs, traumatic prompts consistently reduced the nutritional quality of shopping baskets (Change in Basket Health Scores of -0.081 to -0.126; all pFDR<0.001; Cohens d=-1.07 to -2.05), robust across models and budgets. These results show that psychological context can systematically alter not only what LLMs generate but also the actions they perform. By reproducing human-like emotional biases in consumer behavior, LLM agents reveal a new class of vulnerabilities with implications for digital health, consumer safety, and ethical AI deployment.

cs.HC↗

arXiv2025

Humans Perceive Wrong Narratives from AI Reasoning Texts

A new generation of AI models generates step-by-step reasoning text before producing an answer. This text appears to offer a human-readable window into their computation process, and is increasingly relied upon for transparency and interpretability. However, it is unclear whether human understanding of this text matches the model's actual computational process. In this paper, we investigate a necessary condition for correspondence: the ability of humans to identify which steps in a reasoning text causally influence later steps. We evaluated humans on this ability by composing questions based on counterfactual measurements and found a significant discrepancy: participant accuracy was only 29%, barely above chance (25%), and remained low (42%) even when evaluating the majority vote on questions with high agreement. Our results reveal a fundamental gap between how humans interpret reasoning texts and how models use it, challenging its utility as a simple interpretability tool. We argue that reasoning texts should be treated as an artifact to be investigated, not taken at face value, and that understanding the non-human ways these models use language is a critical research direction.

cs.HC↗