Preethi Seshadri

Preethi Seshadri, Sameer Singh, Yanai Elazar

Bias amplification is a phenomenon in which models exacerbate biases or stereotypes present in the training data. In this paper, we study bias amplification in the text-to-image domain using Stable Diffusion by comparing gender ratios in training vs. generated images. We find that the model appears to amplify gender-occupation biases found in the training data (LAION) considerably. However, we discover that amplification can be largely attributed to discrepancies between training captions and model prompts. For example, an inherent difference is that captions from the training data often contain explicit gender information while our prompts do not, which leads to a distribution shift and consequently inflates bias measures. Once we account for distributional differences between texts used for training and generation when evaluating amplification, we observe that amplification decreases drastically. Our findings illustrate the challenges of comparing biases in models and their training data, and highlight confounding factors that impact analyses.

Preethi Seshadri, Pouya Pezeshkpour, Sameer Singh

Recently, there has been an increase in efforts to understand how large language models (LLMs) propagate and amplify social biases. Several works have utilized templates for fairness evaluation, which allow researchers to quantify social biases in the absence of test sets with protected attribute labels. While template evaluation can be a convenient and helpful diagnostic tool to understand model deficiencies, it often uses a simplistic and limited set of templates. In this paper, we study whether bias measurements are sensitive to the choice of templates used for benchmarking. Specifically, we investigate the instability of bias measurements by manually modifying templates proposed in previous works in a semantically-preserving manner and measuring bias across these modifications. We find that bias values and resulting conclusions vary considerably across template modifications on four tasks, ranging from an 81% reduction (NLI) to a 162% increase (MLM) in (task-specific) bias measurements. Our results indicate that quantifying fairness in LLMs, as done in current practice, can be brittle and needs to be approached with more care and caution.

Preethi Seshadri, Samuel Cahyawijaya, Ayomide Odumakinde et al.

Agentic benchmarks increasingly rely on LLM-simulated users to scalably evaluate agent performance, yet the robustness, validity, and fairness of this approach remain unexamined. Through a user study with participants across the United States, India, Kenya, and Nigeria, we investigate whether LLM-simulated users serve as reliable proxies for real human users in evaluating agents on τ-Bench retail tasks. We find that user simulation lacks robustness, with agent success rates varying up to 9 percentage points across different user LLMs. Furthermore, evaluations using simulated users exhibit systematic miscalibration, underestimating agent performance on challenging tasks and overestimating it on moderately difficult ones. African American Vernacular English (AAVE) speakers experience consistently worse success rates and calibration errors than Standard American English (SAE) speakers, with disparities compounding significantly with age. We also find simulated users to be a differentially effective proxy for different populations, performing worst for AAVE and Indian English speakers. Additionally, simulated users introduce conversational artifacts and surface different failure patterns than human users. These findings demonstrate that current evaluation practices risk misrepresenting agent capabilities across diverse user populations and may obscure real-world deployment challenges.

Abhipsa Basu, Yugam Bahl, Kirti Bhagat et al.

Recent studies show that text-to-image models often fail to generate geographically representative images, raising concerns about the representativeness of their training data and motivating the question: which parts of the world do these training examples come from? We geographically profile large-scale multimodal datasets by mapping image-caption pairs to countries based on location information extracted from captions using LLMs. Studying English captions from three widely used datasets (Re-LAION, DataComp1B, and Conceptual Captions) across $20$ common entities (e.g., house, flag), we find that the United States, the United Kingdom, and Canada account for $48.0\%$ of samples, while South American and African countries are severely under-represented with only $1.8\%$ and $3.8\%$ of images, respectively. We observe a strong correlation between a country's GDP and its representation in the data ($ρ= 0.82$). Examining non-English subsets for $4$ languages from the Re-LAION dataset, we find that representation skews heavily toward countries where these languages are predominantly spoken. Additionally, we find that higher representation does not necessarily translate to greater visual or semantic diversity. Finally, analyzing country-specific images generated by Stable Diffusion v1.3 trained on Re-LAION, we show that while generations appear realistic, they are severely limited in their coverage compared to real-world images.

Sahil Verma, Royi Rassin, Arnav Das et al. · uw

Text-to-image models are trained using large datasets collected by scraping image-text pairs from the internet. These datasets often include private, copyrighted, and licensed material. Training models on such datasets enables them to generate images with such content, which might violate copyright laws and individual privacy. This phenomenon is termed imitation -- generation of images with content that has recognizable similarity to its training images. In this work we study the relationship between a concept's frequency in the training dataset and the ability of a model to imitate it. We seek to determine the point at which a model was trained on enough instances to imitate a concept -- the imitation threshold. We posit this question as a new problem: Finding the Imitation Threshold (FIT) and propose an efficient approach that estimates the imitation threshold without incurring the colossal cost of training multiple models from scratch. We experiment with two domains -- human faces and art styles -- for which we create four datasets, and evaluate three text-to-image models which were trained on two pretraining datasets. Our results reveal that the imitation threshold of these models is in the range of 200-600 images, depending on the domain and the model. The imitation threshold can provide an empirical basis for copyright violation claims and acts as a guiding principle for text-to-image model developers that aim to comply with copyright and privacy laws. We release the code and data at \url{https://github.com/vsahil/MIMETIC-2.git} and the project's website is hosted at \url{https://how-many-van-goghs-does-it-take.github.io}.

Team Cohere, Aakanksha, Arash Ahmadian et al. · mila

In this report we describe the development of Command A, a powerful large language model purpose-built to excel at real-world enterprise use cases. Command A is an agent-optimised and multilingual-capable model, with support for 23 languages of global business, and a novel hybrid architecture balancing efficiency with top of the range performance. It offers best-in-class Retrieval Augmented Generation (RAG) capabilities with grounding and tool use to automate sophisticated business processes. These abilities are achieved through a decentralised training approach, including self-refinement algorithms and model merging techniques. We also include results for Command R7B which shares capability and architectural similarities to Command A. Weights for both models have been released for research purposes. This technical report details our original training pipeline and presents an extensive evaluation of our models across a suite of enterprise-relevant tasks and public benchmarks, demonstrating excellent performance and efficiency.

Catarina G Belém, Preethi Seshadri, Yasaman Razeghi et al.

Gender bias research has been pivotal in revealing undesirable behaviors in large language models, exposing serious gender stereotypes associated with occupations, and emotions. A key observation in prior work is that models reinforce stereotypes as a consequence of the gendered correlations that are present in the training data. In this paper, we focus on bias where the effect from training data is unclear, and instead address the question: Do language models still exhibit gender bias in non-stereotypical settings? To do so, we introduce UnStereoEval (USE), a novel framework tailored for investigating gender bias in stereotype-free scenarios. USE defines a sentence-level score based on pretraining data statistics to determine if the sentence contain minimal word-gender associations. To systematically benchmark the fairness of popular language models in stereotype-free scenarios, we utilize USE to automatically generate benchmarks without any gender-related language. By leveraging USE's sentence-level score, we also repurpose prior gender bias benchmarks (Winobias and Winogender) for non-stereotypical evaluation. Surprisingly, we find low fairness across all 28 tested models. Concretely, models demonstrate fair behavior in only 9%-41% of stereotype-free sentences, suggesting that bias does not solely stem from the presence of gender-related words. These results raise important questions about where underlying model biases come from and highlight the need for more systematic and comprehensive bias evaluation. We release the full dataset and code at https://ucinlp.github.io/unstereo-eval.

Preethi Seshadri, Hongyu Chen, Sameer Singh et al.

Large language models (LLMs) are increasingly being deployed in high-stakes applications like hiring, yet their potential for unfair decision-making remains understudied in generative and retrieval settings. In this work, we examine the allocational fairness of LLM-based hiring systems through two tasks that reflect actual HR usage: resume summarization and applicant ranking. By constructing a synthetic resume dataset with controlled perturbations and curating job postings, we investigate whether model behavior differs across demographic groups. Our findings reveal that generated summaries exhibit meaningful differences more frequently for race than for gender perturbations. Models also display non-uniform retrieval selection patterns across demographic groups and exhibit high ranking sensitivity to both gender and race perturbations. Surprisingly, retrieval models can show comparable sensitivity to both demographic and non-demographic changes, suggesting that fairness issues may stem from broader model brittleness. Overall, our results indicate that LLM-based hiring systems, especially in the retrieval stage, can exhibit notable biases that lead to discriminatory outcomes in real-world contexts.

Arjun Subramonian, Vagrant Gautam, Preethi Seshadri et al. · meta-ai

Numerous methods have been proposed to measure LLM misgendering, including probability-based evaluations (e.g., automatically with templatic sentences) and generation-based evaluations (e.g., with automatic heuristics or human validation). However, it has gone unexamined whether these evaluation methods have convergent validity, that is, whether their results align. Therefore, we conduct a systematic meta-evaluation of these methods across three existing datasets for LLM misgendering. We propose a method to transform each dataset to enable parallel probability- and generation-based evaluation. Then, by automatically evaluating a suite of 6 models from 3 families, we find that these methods can disagree with each other at the instance, dataset, and model levels, conflicting on 20.2% of evaluation instances. Finally, with a human evaluation of 2400 LLM generations, we show that misgendering behaviour is complex and goes far beyond pronouns, which automatic evaluations are not currently designed to capture, suggesting essential disagreement with human evaluations. Based on our findings, we provide recommendations for future evaluations of LLM misgendering. Our results are also more widely relevant, as they call into question broader methodological conventions in LLM evaluation, which often assume that different evaluation methods agree.

Shubham Paul, Sudharsan Senthil, Preethi Seshadri et al.

In the context of cellular networks, users located at the periphery of cells are particularly vulnerable to substantial interference from neighbouring cells, which can be represented as a two-user interference channel. This study introduces two highly effective methodologies, namely TwinNet and SiameseNet, using autoencoders, tailored for the design of encoders and decoders for block transmission and detection in interference-limited environments. The findings unambiguously illustrate that the developed models are capable of leveraging the interference structure to outperform traditional methods reliant on complete orthogonality. While it is recognized that systems employing coordinated transmissions and independent detection can offer greater capacity, the specific gains of data-driven models have not been thoroughly quantified or elucidated. This paper conducts an analysis to demonstrate the quantifiable advantages of such models in particular scenarios. Additionally, a comprehensive examination of the characteristics of codewords generated by these models is provided to offer a more intuitive comprehension of how these models achieve superior performance.

Vinay Uday Prabhu, Sanghyun Han, Dian Ang Yap et al.

In this paper, we propose a Seed-Augment-Train/Transfer (SAT) framework that contains a synthetic seed image dataset generation procedure for languages with different numeral systems using freely available open font file datasets. This seed dataset of images is then augmented to create a purely synthetic training dataset, which is in turn used to train a deep neural network and test on held-out real world handwritten digits dataset spanning five Indic scripts, Kannada, Tamil, Gujarati, Malayalam, and Devanagari. We showcase the efficacy of this approach both qualitatively, by training a Boundary-seeking GAN (BGAN) that generates realistic digit images in the five languages, and also quantitatively by testing a CNN trained on the synthetic data on the real-world datasets. This establishes not only an interesting nexus between the font-datasets-world and transfer learning but also provides a recipe for universal-digit classification in any script.