Case Study: Move.ai improves multimodal STEM reasoning with Labelbox

Move.ai builds expert STEM multimodal training with Labelbox's 150 vetted specialists

The customer, a leading AI lab, faced the challenge of identifying where its large language model (LLM) was failing on K-12 STEM questions. It needed a scalable source of original, domain-specific multimodal prompts and expert-graded answers to feed its real-time training loop, which lacked consistent, qualified human feedback. To address this, it worked with Labelbox and utilized its Alignerr network.

Labelbox's solution leveraged its platform and a network of 150 vetted STEM experts to produce unique multimodal prompts and accurate answers that exposed the model's limitations. This high-quality signal was fed into the lab's real-time training workflow, allowing it to target key weaknesses. As a result, Labelbox delivered a new dataset that significantly improved the lab's model performance on complex STEM questions and became a fully integrated part of its continuous improvement loop.

Move.ai