arXiv subjects

Sebastián Vallejo Vera

Publications and source records attributed to Sebastián Vallejo Vera.

3 recordsLinked to original sources

arXiv2026

Invisibilization as a Source of Probability Bias in Language Models

Language models are aligned to avoid using protected attributes such as race, ethnicity, and gender to inform their responses. Therefore, they increasingly refuse or hedge when such attributes appear in a prompt. However, if a system must decide how much to learn from what people tell it, it is costly to refuse valuable information because of its source. In this article, we measure the cost of declining information in a two-urn game with an exact Bayesian answer. We present leading LLM models from two developers (OpenAI and Anthropic) with a prompt that requires them to decide how far to trust an unverifiable report from a participant whose description we randomize across ten configurations (6,600 sessions). We show that leading models often decline to assess the reporter, treating information as worthless at rates set by the description rather than by the evidence. At its widest, models decline a Black-described reporter 46% of the time and a white-described reporter 9%; yet, when they do assess the information, the Black-described reporter receives slightly more trust. The bias is invisible to audits that compare average trust or look only at assessed cases. It grows with reasoning effort, follows words rather than categories (models invoke race for "Black" but not for its Spanish translation negro, and never for "Indigenous"), and worsens under warning instructions a practitioner would try first. We also present a scoring option that reduces the refusal rate bias. When a model's trust is scored against revealed truth over repeated rounds, it credits identical records equally. In four of five configurations from Anthropic and OpenAI, the racial gap in trust falls below 0.02 by the last rounds.

cs.CL↗

arXiv2025

Memory Is All You Need: Testing How Model Memory Affects LLM Performance in Annotation Tasks

Generative Large Language Models (LLMs) have shown promising results in text annotation using zero-shot and few-shot learning. Yet these approaches do not allow the model to retain information from previous annotations, making each response independent from the preceding ones. This raises the question of whether model memory -- the LLM having knowledge about its own previous annotations in the same task -- affects performance. In this article, using OpenAI's GPT-4o and Meta's Llama 3.1 on two political science datasets, we demonstrate that allowing the model to retain information about its own previous classifications yields significant performance improvements: between 5 and 25\% when compared to zero-shot and few-shot learning. Moreover, memory reinforcement, a novel approach we propose that combines model memory and reinforcement learning, yields additional performance gains in three out of our four tests. These findings have important implications for applied researchers looking to improve performance and efficiency in LLM annotation tasks.

cs.CL↗

arXiv2022

Depolarization of echo chambers by random dynamical nudge

In social networks, users often engage with like-minded peers. This selective exposure to opinions might result in echo chambers, i.e., political fragmentation and social polarization of user interactions. When echo chambers form, opinions have a bimodal distribution with two peaks on opposite sides. In certain issues, where either extreme positions contain a degree of misinformation, neutral consensus is preferable for promoting discourse. In this paper, we use an opinion dynamics model that naturally forms echo chambers in order to find a feedback mechanism that bridges these communities and leads to a neutral consensus. We introduce random dynamical nudge (RDN), which presents each agent with input from a random selection of other agents' opinions and does not require surveillance of every person's opinions. Our computational results in two different models suggest that the RDN leads to a unimodal distribution of opinions centered around the neutral consensus. Furthermore, the RDN is effective both for preventing the formation of echo chambers and also for depolarizing existing echo chambers. Due to the simple and robust nature of the RDN, social media networks might be able to implement a version of this self-feedback mechanism, when appropriate, to prevent the segregation of online communities on complex social issues.

physics.soc-ph↗