Munachiso Samuel (Sam) Nwadike
•
•
•
Research Engineer (II) @ MBZUAI, RIKEN AIP
IRL Photo: Sam
I am a research-engineer, with dual affiliation to MBZUAI and RIKEN Center for Advanced Intelligence Project, Japan.
I earned my degrees at New York University (BSc) and MBZUAI (MSc).
For full details, please see my "introduction" section below.
Awards
Introduction
I am a Research Engineer II at Mohamed bin Zayed University of Artificial Intelligence (MBZUAI), a highly ranked AI university according to CSRankings. I earned my M.Sc. in Machine Learning from MBZUAI and my B.Sc. in Computer Science, with a minor in Mathematics, from New York University Abu Dhabi.
My research interests, within machine learning, center on deep learning, reasoning, and neural network architectural design. I am interested in understanding what principles may enable increasingly competent, reliable, and human-aligned deep learning models.
I currently work with Dr. Kentaro Inui on both language models and multimodal models at the MBZUAI Interpretability Team.
I have been privileged to receive world-class research training from brilliant researchers in AI. I received formative training from Dr. Kun Zhang and Dr. Martin Takáč, and later co-authored RECALL with Dr. Takáč, my first first-author paper as a research engineer at MBZUAI, which appeared in ACL 2025.
I also worked with Dr. Qirong Ho of Petuum on Bayesian optimization for hyperparameter tuning, helping researchers tune AI systems under strict compute and budget constraints. At NYU, I worked with Dr. Farah Shamout at the Clinical Artificial Intelligence Lab.
Selected Publications
For a full list, please see my
Google Scholar 
.
A guide to evaluating reasoning as an adaptive, multi-step search, placing emphasis on process evidence rather than final-answer accuracy alone. It sought to bridge the gap between theoretical proofs regarding the computational expressivity of chain-of-thought, and research practice.
Introduces self-referencing causal cycles as a mechanism that helps autoregressive language models retrieve information in reverse or non-standard directions. The paper connects this mechanism to the reversal curse and proposes ReCall as a two-step retrieval process.
Finds that sycophancy-related signals are especially linearly accessible in attention-head activations. The work uses probing and steering to show that targeted attention-head interventions can reduce deference to incorrect user beliefs.
Finds that Mamba models preserve recent inputs in short-lived memory and selected early inputs in sparse long-memory channels. Middle items often fade from short-lived memory without entering the long-memory channels, creating a bottleneck that helps explain “lost-in-the-middle” behavior.
News
June 2026 · Call for datasets
My paper, Einstein World Models, calls for datasets to train LLMs that can interleave chain-of-thought reasoning with counterfactual visualisation. Inspired by Einstein’s famous visual thought experiments, we propose a greater role for imagination in advancing AI reasoning.
May 2026 · ICML Seoul
I was selected as a Gold Reviewer for ICML 2026 and will be attending the conference in Seoul, Korea. I am grateful for the opportunity to contribute to the review process and to participate in the broader machine learning research community.
September 2025 · MBZUAI News
I was featured in MBZUAI News for work on using AI to help businesses query raw data and obtain instant analytics.
September 2025 · Khaleej Times
I was featured in Khaleej Times in coverage of UAE-based AI work for turning raw data into business insights. The article photos look better on mobile.
July 2025 · MBZUAI News
MBZUAI News covered our work on ReCall, a method aimed at addressing the reversal curse in large language models.
August 2023 · Geeniuc Games
Congratulations to my 8 interns for successfully completing their summer 2023 internship with Geeniuc Games. It was a pleasure to support their growth and see the projects they built over the summer.
Paperbox: Click To Read