Malia Morgan

Malia Morgan is an AI researcher whose work spans computer science education, multi-agent planning, computer vision and open language models. At Harvey Mudd, they coauthored a biology-based CS1 study and a paper on improvisational teamwork, studied attention in action-recognition neural networks, and received a 2023 CRA undergraduate research honorable mention. In 2025, they coauthored Olmo 3; in 2026, they coauthored two papers on language-model evaluation and safety. They are a Predoctoral Young Investigator at Ai2; they began the role in 2025 with Maarten Sap as co-mentor.

Now at: Allen Institute for AI (Ai2)

Timeline

2021

A biology-based CS1

In 2021, they coauthor “A Biology-based CS1: Results and Reflections, Ten Years In,” a study that tracks a decade of student outcomes from Harvey Mudd’s biology-based introductory course alongside its breadth-first introduction to computer science.
2022

Planning improvised teamwork

In 2022, they coauthor “Simple Temporal Networks for Improvisational Teamwork,” a model that captures temporal dependencies and uncertainty among agents who coordinate with limited communication.
2023

Attention in action recognition

In 2023, during their senior year at Harvey Mudd, they work in Professor Calden Wloka’s Lab for CATS on attention in action-recognition neural networks and compare model attention with human attentional behavior.
2023

A CRA research honor

Harvey Mudd computer science senior Malia Morgan ’23 receives an honorable mention in the CRA’s 2023 Outstanding Undergraduate Researcher Awards.
2025

Olmo 3 open models

In 2025, they coauthor Olmo 3, a fully open family of language models at the 7B and 32B parameter scales that targets long-context reasoning, function calling, coding, instruction following, general chat and knowledge recall.
2026

BenchMIRT evaluation

In 2026, they coauthor the preprint “BenchMIRT: Disentangling Safety and General Capabilities in LLM Evaluation” with Vasudha Varadarajan, Faeze Brahman and Maarten Sap.
2026

CarryOnBench and LLM safety

In 2026, they coauthor “Useless but Safe? Benchmarking Utility Recovery with User Intent Clarification in Multi-Turn Conversations,” which introduces CarryOnBench, an interactive benchmark for whether LLMs can recover helpfulness after benign intent clarification while remaining safe.
Since 2025

Predoctoral Young Investigator at Ai2

They join Ai2 in 2025 as a Predoctoral Young Investigator, with Maarten Sap as a co-mentor. As of October 2026, they are a Pre-doctoral Young Investigator at Ai2.