Introducing EJBench: engineering judgement for AI
We’ve introduced EJBench, a benchmark for evaluating AI models’ engineering judgement. Using evidence from battery cold-plate simulations, it asks models to reason about designs and make engineering decisions without running new simulations.
Read the EJBench paper on arXiv for the research, or visit the EJBench page to explore the leaderboard and our work on engineering RL environments and model evaluation.
