arXiv subjects

Carrie Cai

Publications and source records attributed to Carrie Cai.

5 recordsLinked to original sources

arXiv2024

LLM Agents Grounded in Self-Reports Enable General-Purpose Simulation of Individuals

Machine learning can predict human behavior well when substantial structured data are available for well-defined outcomes. Such models are typically outcome-specific, however, requiring training data for each target outcome, limiting their applicability to new domains. We test whether large language models (LLMs) can relax these requirements by using self-report data to build attitudinal and behavioral simulations, or "generative agents," that can predict responses across outcomes without outcome-specific training data. Using data from a diverse national sample of 1,052 Americans, we built agents from (i) two-hour, semi-structured interviews elicited using the American Voices Project interview schedule, (ii) structured surveys including General Social Survey items and the Big Five personality inventory, or (iii) both sources combined. On held-out General Social Survey items, interview-only, survey-only, and combined agents achieved accuracies equal to 83%, 82%, and 86% of participants' own two-week test-retest consistency benchmark, respectively, compared with 74% for demographics-only agents. Combining interviews and surveys produced the highest accuracy, though gains over either source alone were modest, suggesting that predictive benefits from data begin to asymptote once the model has observed sufficient evidence within a domain. We find that these agents also predict personality traits, economic-game behavior, and experimental responses, while reducing accuracy disparities across racial and ideological groups relative to demographics-only agents. Together, these results show that LLM agents grounded in qualitative or quantitative self-reports can support general-purpose simulation of individuals across outcomes, without requiring task-specific training data.

cs.AI↗

arXiv2023

The Prompt Artists

This paper examines the art practices, artwork, and motivations of prolific users of the latest generation of text-to-image models. Through interviews, observations, and a user survey, we present a sampling of the artistic styles and describe the developed community of practice around generative AI. We find that: 1) the text prompt and the resulting image can be considered collectively as an art piece prompts as art and 2) prompt templates (prompts with ``slots'' for others to fill in with their own words) are developed to create generative art styles. We discover that the value placed by this community on unique outputs leads to artists seeking specialized vocabulary to produce distinctive art pieces (e.g., by reading architectural blogs to find phrases to describe images). We also find that some artists use "glitches" in the model that can be turned into artistic styles of their own right. From these findings, we outline specific implications for design regarding future prompting and image editing options.

cs.HC↗

arXiv2023

A Word is Worth a Thousand Pictures: Prompts as AI Design Material

Recent advances in Machine-Learning have led to the development of models that generate images based on a text description.Such large prompt-based text to image models (TTIs), trained on a considerable amount of data, allow the creation of high-quality images by users with no graphics or design training. This paper examines the role such TTI models can playin collaborative, goal-oriented design. Through a within-subjects study with 14 non-professional designers, we find that such models can help participants explore a design space rapidly and allow for fluid collaboration. We also find that text inputs to such models ("prompts") act as reflective design material, facilitating exploration, iteration, and reflection in pair design. This work contributes to the future of collaborative design supported by generative AI by providing an account of how text-to-image models influence the design process and the social dynamics around design and suggesting implications for tool design

cs.HC↗

arXiv2021

Program Synthesis with Large Language Models

This paper explores the limits of the current generation of large language models for program synthesis in general purpose programming languages. We evaluate a collection of such models (with between 244M and 137B parameters) on two new benchmarks, MBPP and MathQA-Python, in both the few-shot and fine-tuning regimes. Our benchmarks are designed to measure the ability of these models to synthesize short Python programs from natural language descriptions. The Mostly Basic Programming Problems (MBPP) dataset contains 974 programming tasks, designed to be solvable by entry-level programmers. The MathQA-Python dataset, a Python version of the MathQA benchmark, contains 23914 problems that evaluate the ability of the models to synthesize code from more complex text. On both datasets, we find that synthesis performance scales log-linearly with model size. Our largest models, even without finetuning on a code dataset, can synthesize solutions to 59.6 percent of the problems from MBPP using few-shot learning with a well-designed prompt. Fine-tuning on a held-out portion of the dataset improves performance by about 10 percentage points across most model sizes. On the MathQA-Python dataset, the largest fine-tuned model achieves 83.8 percent accuracy. Going further, we study the model's ability to engage in dialog about code, incorporating human feedback to improve its solutions. We find that natural language feedback from a human halves the error rate compared to the model's initial prediction. Additionally, we conduct an error analysis to shed light on where these models fall short and what types of programs are most difficult to generate. Finally, we explore the semantic grounding of these models by fine-tuning them to predict the results of program execution. We find that even our best models are generally unable to predict the output of a program given a specific input.

cs.PL↗

arXiv2018

Interpretability Beyond Feature Attribution: Quantitative Testing with Concept Activation Vectors (TCAV)

The interpretation of deep learning models is a challenge due to their size, complexity, and often opaque internal state. In addition, many systems, such as image classifiers, operate on low-level features rather than high-level concepts. To address these challenges, we introduce Concept Activation Vectors (CAVs), which provide an interpretation of a neural net's internal state in terms of human-friendly concepts. The key idea is to view the high-dimensional internal state of a neural net as an aid, not an obstacle. We show how to use CAVs as part of a technique, Testing with CAVs (TCAV), that uses directional derivatives to quantify the degree to which a user-defined concept is important to a classification result--for example, how sensitive a prediction of "zebra" is to the presence of stripes. Using the domain of image classification as a testing ground, we describe how CAVs may be used to explore hypotheses and generate insights for a standard image classification network as well as a medical application.

stat.ML↗