Artificial Intelligence Researcher

Cerebro; San Francisco (CA); On-site; USD 160,000 - 230,000; Full time

Job summary

Cerebro, a research-focused AI company in San Francisco, seeks a Research Scientist / Member of Technical Staff for LLM Evaluation & Post-Training. You will tackle the limitations of current evaluation methods as models improve at multi-step reasoning, tool use, and autonomous task execution.

You will work with a small, technically strong team, owning the problems you investigate and the methods you develop.

Qualifications

• Original research in LLMs, reinforcement learning, post-training, reasoning or evaluation.
• Experience bridging theory and hands-on experimentation in AI research.

Responsibilities

• Develop and evaluate robust methods for LLM evaluation and post-training.
• Design experiments to measure reasoning, tool use and autonomous task execution.
• Build scalable research infrastructure for experimental evaluation at scale.
• Contribute to publications and share findings with the team and broader community.

Job description

Research Scientist / Member of Technical Staff — LLM Evaluation & Post-Training

Most frontier AI labs are focused on making models more capable. Far fewer are tackling the equally difficult question of how we actually measure those capabilities.

I'm working with an AI research company building at precisely that intersection.

They're looking for researchers interested in the limitations of current LLM evaluation methods, particularly as models become better at multi-step reasoning, tool use and autonomous task execution.

The research challenges are interesting:

• Evaluating reasoning: How do you meaningfully assess complex, multi-step reasoning when existing benchmarks are increasingly saturated?
• Reward modelling: How do you build reliable signals for tasks where correctness is difficult to verify?
• Agent evaluation: How do you measure whether an agent can complete complex tasks reliably, rather than simply produce plausible outputs?
• Post-training: How can evaluation and feedback be used to systematically improve model behaviour?
• Research infrastructure: How do you turn experimental evaluation methods into robust systems that operate at scale?

This is a research position, not an LLM application engineering role.

The team is particularly interested in people who have conducted original research in LLMs, reinforcement learning, post-training, reasoning or evaluation, and who are comfortable moving between theoretical questions and hands‑on experimentation.

Publications at NeurIPS, ICML or ICLR are valuable, but so is evidence of having independently identified and solved difficult research problems.

You’ll work directly with a small, technically strong team, with significant ownership over both the problems you investigate and the methods you develop.