It's a platform built at DALI Lab to make learning about large language models accessible to the Dartmouth community. The platform lets users play with the mechanics behind LLMs by adjusting temperature, top k, and other generation parameters and see how the output shifts. Each word in a response shows its probability of appearing, so instead of treating these models like black boxes, users develop an intuition for how language models are actually deciding what to say next.
Users can choose from a large selection of models and compare how different architectures respond to the same prompt under the same settings. The learning experience is gamified through progressive levels that unlock as users build understanding, keeping people engaged without oversimplifying the material. The content covers the full picture of LLMs ie. how they work, the probabilistic foundations underneath, and their environmental and social impacts, reviewed by domain experts at Dartmouth.
I was the product designer on the project and conducted user research, interaction design, content strategy, and user testing across a 10 week DALI sprint.