Roi Cohen
Senior Product Manager and AI researcher
Tel Aviv-Yafo, Israel
Joined January 2026
Summary
Bridges product leadership and applied AI research: Roi has held senior product management roles at multiple AI and technology companies while also authoring research on language-model behaviour and evaluation. rocketreach+3
Active contributor to LLM evaluation and knowledge-editing research: Roi is a co-author of papers addressing factuality detection, the effects of knowledge edits, and uncertainty in pretrained LLMs. arxiv+2
Hands-on product experience in foundation models and AI platform work: public directory entries and company listings identify Roi as a product manager focused on foundation-model products (AI21 Labs) and as a Senior Product Manager at Immunai. startuphub+1
Combines technical background with formal business training: Roi's background includes a technical undergraduate degree in computer science & economics and an MBA from Stanford GSB, reflecting both technical and business education. rocketreach+1
Work
Education
Projects
Writing
Pretrained LLMs Learn Multiple Types of Uncertainty
January 1, 2025Studies how pretrained large language models capture several distinct types of uncertainty and how these signals can predict correctness and abstention behavior.
Jamba-1.5: Hybrid Transformer-Mamba Models at Scale
January 1, 2024Describes the Jamba-1.5 models: instruction-tuned hybrid Transformer-Mamba models with large context windows and techniques for efficient inference.
LM vs LM: Detecting Factual Errors via Cross Examination
January 1, 2023Proposes a factuality evaluation framework for language models based on cross-examination between models to find inconsistencies and detect factual errors.
Evaluating the Ripple Effects of Knowledge Editing in Language Models
January 1, 2023Introduces evaluation criteria and a diagnostic benchmark (RippleEdits) to assess how editing one fact in a model affects related facts (the 'ripple effect').