Desiree Cho
Desiree Cho
Interests
AI alignment, AI safety.
Biography
I'm a DPhil (PhD) student in Computer Science and a Doctoral Scholar at the Institute for Ethics in AI, supervised by Professor Sir Nigel Shadbolt (Chair of AI@Oxford Research). My research is in technical AI safety and alignment. My current work focuses on constitutional midtraining (generously funded by BlueDot Impact), where I've been particularly interested in values-based reasoning. Another recent project of mine was on diverse reward model ensembles, which won first place in Research Impact Oxford.
My background is in human-centred computing and natural language processing. Before this PhD, I completed my Master of Science at the University of Oxford, graduating with Distinction in my degree and thesis. My thesis was on recommender system alignment, where I built an NLP pipeline to measure the alignment between recommendations and user behaviour. I've published at leading human-centred computing venues, namely ACM CHI and the International Journal of Human-Computer Studies, the latter of which published my Bachelor's dissertation.
Selected Publications
-
Constitutional Midtraining: Content Presence Drives Alignment Gains
Desiree Cho; Cameron Tice; Bernie Hogan; Hunar Batra; Puria Radmard; Jun Zhao; Nigel Shadbolt
2026.
Details about Constitutional Midtraining: Content Presence Drives Alignment Gains | BibTeX data for Constitutional Midtraining: Content Presence Drives Alignment Gains | Link to Constitutional Midtraining: Content Presence Drives Alignment Gains