Young Investigator, FlexOlmo

The Allen Institute for Artificial Intelligence•Berkeley, CA

4d•Remote

About The Position

Ai2 is seeking a talented and motivated Postdoctoral Young Investigator to join the FlexOlmo team, which develops large language models for flexible data use, with a focus on Mixture-of-Experts (MoE), long-context language models (LCLMs), and retrieval. The role is based in Berkeley, California, with remote work welcome from Berkeley. The position offers an opportunity to conduct cutting-edge research in natural language processing and machine learning in a fast-paced research environment, define and lead a high-impact research project, train and release leading models, collaborate with Ai2 team members, build open-source software, and publish scientific papers in high-profile venues. The Young Investigator role lasts 1–3 years and has a flexible start date. Candidates must be within one year of completing a PhD or already hold a PhD. The program combines collaborative work on an Ai2 project with opportunities to mentor junior researchers. The role includes a dedicated Ai2 mentor who is also a University of California, Berkeley faculty member. Approximately 50% of the work is as an independent contributor leading and collaborating on an Ai2 project, and approximately 50% is mentoring junior researchers, with optional mentorship in academic activities such as grant writing and teaching. The FlexOlmo team designs architectures and training methods that help models use data more effectively through improved training, inference-time conditioning, and retrieval. It also develops scientific methods for evaluating and understanding these systems and produces open-source research tools. The team leads the FlexOlmo project, whose first release in July 2025 introduced a Mixture-of-Experts architecture. The team plans further research into architectures and training methods for using large and diverse data sources. The group is small and closely collaborative, welcomes open collaborations with external researchers, and offers opportunities to engage with UC Berkeley and the BAIR lab. Ai2 supports open science and publication, and visa sponsorship is available. Compensation is $159,650. The position requires the ability to remain stationary for long periods, communicate and exchange information accurately, observe details at close range, and work under deadlines. Reasonable accommodations may be made for individuals with disabilities.

Requirements

Be within one year of completing a PhD or already hold a PhD in Computer Science or a similar field.
Have research experience in machine learning, natural language processing, language and vision, or related areas.
Demonstrate outstanding individual-contributor skills, especially with deep-learning frameworks such as PyTorch.
Have an outstanding publication record at AI-related venues such as NeurIPS, ICLR, ICML, COLM, ACL, or EMNLP; publications are evaluated for rigor and impact rather than quantity.
Have extensive research experience in areas such as large language models, training dynamics, scaling laws, and data curation.
Be located in Berkeley, California, or be willing to relocate there.
Be able to remain stationary for long periods, communicate and exchange information accurately, observe details at close range, and work under deadlines.

Nice To Haves

Experience with Mixture-of-Experts.
Experience with long-context language models.
Experience with retrieval.

Responsibilities

Define and lead a high-impact research project.
Train and release leading models.
Lead and collaborate as an independent contributor on an Ai2 project.
Mentor junior researchers, including PhD students, interns, and predoctoral students or interns.
Build open-source software for the research community.
Author scientific papers for publication in high-profile conferences or journals.
Collaborate with Ai2 team members and, where appropriate, researchers at external institutions.
Conduct research into language-model architectures and training methods that improve flexible use of data.
Develop scientific methodologies for evaluating and understanding language-model systems.

Benefits

Medical, dental, and vision coverage.
Basic life insurance and basic accidental death and dismemberment insurance.
Short-term and long-term disability coverage.
Employee assistance program.
Voluntary life insurance and voluntary accidental death and dismemberment coverage.
Health savings account, healthcare reimbursement arrangement, and health care and dependent care flexible spending account plans.
Company 401(k) plan.
$125 per month to assist with commuting or internet expenses.
$200 per month for fitness and wellbeing expenses.
Up to ten sick days per year, up to seven personal days per year, up to 20 vacation days per year, and twelve paid holidays per year.
Annual bonuses.
Family leave.
Ongoing education and learning opportunities.