01
Teach
How can users communicate concepts and preferences to AI?
I build learning-centered interfaces that help people express knowledge and refine it alongside a model.
Human-AI Interaction
Simret Araya Gebreegziabher
ስምረት አርአያ ገብረእግዚአብሄር
I study how humans and AI learn together.
I am a Postdoctoral Fellow at the Carnegie Mellon University Human-Computer Interaction Institute. My research examines how humans interact with, evaluate, and shape increasingly autonomous AI systems. I design and study human-AI systems that support effective collaboration, alignment, and learning, with a particular focus on how people communicate their preferences and values, interpret AI behavior, and maintain meaningful oversight as AI systems take on longer-horizon and consequential tasks.
Before joining CMU, I earned my Ph.D. in Computer Science and Engineering from the University of Notre Dame, where I was advised by Prof. Toby Li. My doctoral research explored human-AI alignment as a bi-directional, co-adaptive process in which users and AI systems learn from and adapt to one another. Drawing on theories of human learning and cognition, I developed interactive systems that help AI better understand human intent while also helping users articulate, refine, and evaluate their own concepts, preferences, and values.
How can humans remain active partners in shaping AI behavior?
My research follows the relationship between humans and AI across three connected activities: teaching models, evaluating their behavior, and maintaining meaningful human oversight.
02
Evaluate
Who defines what good AI behavior means, and how?
I design workflows that bring domain experts, end users, and model-generated criteria into the evaluation process.
03
Oversee & Align
How can users recognize and correct agent misalignment?
I study how agent behavior reflects human values and how oversight can help users identify and address harmful behavior.
Selected work
Publications
Stayin'Aligned Over Time: Towards Longitudinal Human-LLM Alignment via Contextual Reflection and Privacy-Preserving Behavioral Data
Dark Patterns Meet GUI Agents: LLM Agent Susceptibility to Manipulative Interfaces and the Role of Human Oversight
The Obvious Invisible Threat: LLM-Powered GUI Agents’ Vulnerability to Fine-Print Injections
Designing Staged Evaluation Workflows for LLMs: Integrating Domain Experts, Lay Users, and Model-Generated Evaluation Criteria
Supporting Co-Adaptive Machine Teaching through Human Concept Learning and Cognitive Theories
MetricMate: An Interactive Tool for Generating Evaluation Criteria for LLM-as-a-Judge Workflow
News
I joined Carnegie Mellon University as a Postdoctoral Fellow.
I defended my dissertation and earned my Ph.D. from the University of Notre Dame.
Our CHI ’25 paper on co-adaptive machine teaching won a Best Paper Award.
I will be at UIST presenting our demo paper, MOCHA: Model Optimization through Collaborative Human-AI Alignment.
Finished my internship at Meta Reality Labs.
Started my internship at Meta Reality Labs.
Our paper PaTAT was accepted to CHI.
Started my Ph.D. at the University of Notre Dame.
Graduated from Addis Ababa University with a B.Sc. in Software Engineering, with great distinction.
Gave my graduation lecture as a Global Teaching Fellow at Delta Analytics. Watch it ↗
Started as a Global Teaching Fellow at Delta Analytics ↗
Started working as a research intern at the University of Michigan’s AURA 2020 program ↗
Organized the first AI in Ethiopia Conference. Read about it ↗