Snorkel AI is a data development company focused on improving AI systems. They are seeking experienced electrical engineers to contribute to an AI evaluation initiative by creating challenging graduate-level questions to benchmark AI models across electrical engineering disciplines.
Responsibilities:
- Create challenging, graduate-level multiple-choice questions in your area of electrical engineering expertise
- Design questions that a working engineer could answer correctly, but that reliably stump frontier AI models
- Write questions that demand real reasoning, conceptual precision, and expert judgment, not surface-level recall
- Develop questions around the hard problems in your field: circuit analysis edge cases, power systems design trade-offs, signal processing theory, control systems stability, semiconductor device physics, competing approaches to electromagnetic modeling, or subtle distinctions between similar components and their behaviors under real-world conditions
- Support each question with a verifiable citation from an authoritative source, such as a peer-reviewed journal, IEEE publication, engineering textbook, or technical standard
- Use the platform's evaluation tool to confirm that frontier AI models answer your question incorrectly at least 4 out of 5 times
- Depending on the task, contribute as a submitter, reviewer, or adjudicator, authoring original questions, independently evaluating peer submissions, or making final quality decisions
Requirements:
- Graduate-level expertise or strong professional experience in electrical engineering or a closely related discipline such as computer engineering, power systems, photonics, or RF engineering
- Deep command of your field's core concepts, methods, standards, and advanced technical reasoning
- Ability to identify questions that are genuinely hard for AI but fair and answerable for human experts
- Excellent written communication and strong attention to detail
- Experience writing, reviewing, teaching, researching, or evaluating advanced technical content
- Comfort using citations to substantiate domain-specific claims and correct answers