AI Emergent Capability Probe Engineer

Velvix Labs

Hybrid · $160k – $200k / year · Full-time · Posted 4 months ago

This role is no longer accepting applications

Don't miss the next role like this

Create a free profile with your skills. We match you to new AI roles as they open, and companies hiring can contact you directly.

Create free profile, 2 minutes

Similar open roles you can apply to now:

See all open AI jobs

This role is no longer accepting applications

Don't miss the next role like this

Create a free profile with your skills. We match you to new AI roles as they open, and companies hiring can contact you directly.

Create free profile, 2 minutes

Similar open roles you can apply to now:

See all open AI jobs

About the role

Emergent capabilities — behaviors that appear in large language models at scale without being explicitly trained — are one of the most poorly understood phenomena in AI development. They matter for safety because a capability that isn't known to exist can't be evaluated for misuse potential or monitored in deployment. The goal of capability probing is to systematically discover what a model can do before deployment, not after. We're an AI safety research organization that develops evaluation methodology for frontier models. Our capability probing work involves designing elicitation prompts that reliably surface latent capabilities, building automated discovery pipelines that search for unexpected model behaviors, and creating standardized assessments that can track capability emergence as a function of model scale. This is a senior research engineer role at the intersection of ML and AI safety evaluation. You'll design probing methodologies for capability categories we care about — deceptive alignment, situational awareness, strategic reasoning, novel tool use — build the automated probing infrastructure, and document your findings in research-quality reports that inform our policy recommendations and model evaluation partnerships. The role is hybrid, with 2-3 days per week in our SF office for research team collaboration.

Responsibilities

Design systematic elicitation methodologies for probing specific model capability categories
Build automated capability discovery pipelines searching for unexpected model behaviors at scale
Implement quantitative capability benchmarks enabling cross-model and cross-scale comparison
Develop probing evaluation methodology robust to prompt sensitivity and elicitation variance
Document capability findings in research reports informing safety evaluation frameworks
Collaborate with alignment researchers to prioritize capability categories based on risk relevance

Requirements

6+ years in ML research with model evaluation, AI safety, or interpretability experience
Deep understanding of LLM capabilities and their relationship to scale, training data, and RLHF
Strong Python and experimental design skills
Research track record with publications or significant contributions in AI evaluation or safety
Ability to design rigorous empirical studies with appropriate controls and statistical analysis