Remote, San Francisco, Delhi
Responded to 100% candidates
OneBit AI is exploring ultra-low-bit AI models, with a focus on reducing model precision toward 1-bit / 1.58-bit / ternary representations while preserving model capability and making inference more efficient. We are looking for someone who enjoys understanding why things work, questioning assumptions, and experimenting with ideas from research papers.
Work on research and experiments around ultra-low-bit neural networks and LLMs.
Help the team understand why model capability is lost during extreme quantisation and how it can be recovered.
Turn ideas from papers into experiments, prototypes, and practical learning.
Explore and evaluate techniques for ultra-low-bit neural networks and LLMs.
Help us understand and reproduce relevant research papers.
Design experiments to test different quantisation and training approaches.
Analyse why model performance is lost during extreme quantisation and how it can be recovered.
Help identify gaps, hidden assumptions, and limitations in existing research.
Guide the team on model training, quantisation, and efficient inference.
Work closely with engineers to turn research ideas into experiments and prototypes.
Challenge our assumptions and propose better approaches when appropriate.
We care more about fundamental understanding and research ability than a specific degree or programming background.
You may come from computer science / AI / ML, physics, mathematics, electrical or electronics engineering, applied mathematics, computational science, or another quantitative field.
You should have a good understanding of basic machine learning and neural networks.
You should be curious about how modern AI models actually work.
You should be able to read, write and understand research papers.
You should have strong analytical and problem-solving ability.
You should be willing to experiment, fail, debug, and investigate.
You should be able to explain technical ideas clearly.
Experience with LLMs, quantisation, model compression, low-bit neural networks, or efficient AI is a strong advantage but is not mandatory.
You do not need to be an expert programmer or know dozens of ML libraries.
We are comfortable with candidates using AI coding tools to implement experiments. What matters most is that you understand what we are doing, why we are doing it, and whether the results actually make sense.
A PhD is not required yet preffered in AI related fields. Demonstrated research ability and strong fundamentals matter more than formal credentials.
A resume is not mandatory, but we would love to see it.
PLEASE DON'T USE AI TO FILL THIS FORM;