For each project, a Large Language Model (Llama 3.3-70B Instruct) was used to generate defined problem statements for each use case and to classify the project description as part of (12) individual AI activities. Activities come from the National Institute for Technology & Standards (NIST) AI Use Case Taxonomy and can be viewed as functions describing how an AI system contributes to a human’s overall task and intended outcomes. Each activity describes the manner in which the AI system augments or replaces human effort and maps to the goals of the user and their interaction with an AI system.

Results were audited by the dataPIG team to correct inaccuracies in the LLM classification process, yielding substantial revisions, as only approximately 40% of classifications were deemed accurate. This highlights the ongoing difficulties in using LLMs for classification in cases of extreme nuance, such as those contained in the NIST AI use cases taxonomy. In contrast, only approximately 10% of problem statements required revisions, as LLMs are well-equipped for reformatting of text into structured formats.

Model type classifications were made exclusively by members of the dataPIG team based on the project descriptions and there may be significant overlap between model types. Please consult model descriptions on the summary dashboard page for additional information on how each model was classified. 

Background images for the dashboard were generated using Midjourney v6.1 and icons sourced from the Noun Project.