Skip to main content
eScholarship
Open Access Publications from the University of California

Scholarly Works (2 results)

  • Thesis
  • Peer Reviewed
UC Davis Electronic Theses and Dissertations (2023)

Large Language Models (LLMs) have demonstrated impressive capability in generating fluent text. LLMs have also shown a tendency to reproduce social biases such as stereotypical associations between gender and occupation. Like race and gender, morality is an important social variable. This work investigates whether LLMs reproduce the moral biases associated with political groups in the United States, an instance of a broader capability I refer to as "moral mimicry". I explore this hypothesis in the GPT-3/3.5 and OPT families of Transformer-based LLMs. Using tools from Moral Foundations Theory, I show that these LLMs are indeed "moral mimics". When prompted with a "liberal" or "conservative" political identity, the models generate text reflecting the moral biases associated with these groups. I investigate how moral mimicry relates to model scale. I hope that this work encourages further investigation of the moral mimicry capability, including how to leverage it for social good and minimize its risks.

    • Article
    • Peer Reviewed
    UC Davis Previously Published Works (2019)
    The Automated Self-Administered 24-Hour Dietary Assessment Tool (ASA24) is a free dietary recall system that outputs fewer nutrients than the Nutrition Data System for Research (NDSR). NDSR uses the Nutrition Coordinating Center (NCC) Food and Nutrient Database, both of which require a license. Manual lookup of ASA24 foods into NDSR is time-consuming but currently the only way to acquire NCC-exclusive nutrients. Using lactose as an example, we evaluated machine learning and database matching methods to estimate this NCC-exclusive nutrient from ASA24 reports. ASA24-reported foods were manually looked up into NDSR to obtain lactose estimates and split into training (n = 378) and test (n = 189) datasets. Nine machine learning models were developed to predict lactose from the nutrients common between ASA24 and the NCC database. Database matching algorithms were developed to match NCC foods to an ASA24 food using only nutrients ("Nutrient-Only") or the nutrient and food descriptions ("Nutrient + Text"). For both methods, the lactose values were compared to the manual curation. Among machine learning models, the XGB-Regressor model performed best on held-out test data (R2 = 0.33). For the database matching method, Nutrient + Text matching yielded the best lactose estimates (R2 = 0.76), a vast improvement over the status quo of no estimate. These results suggest that computational methods can successfully estimate an NCC-exclusive nutrient for foods reported in ASA24.
      Cover page: Nutrient Estimation from 24-Hour Food Recalls Using Machine Learning and Database Mapping: A Case Study with Lactose
      Top