As catalogs for digital marketplaces grow in size, customers have greater difficulty in finding products that meet their needs. Marketplace owners might improve product discovery by adding new categories, tags, or other searchable and filterable metadata to product descriptions. As it is often difficult to get product owners to update old product descriptions, a marketplace owner may use automated methods to “backfill” product data. We investigate using multi-modal computer models to extract data from digital products, specifically, digital maps for tabletop role playing games. Across five tasks, we find that models perform comparatively with their ranking on the Open VLM Leaderboard. State-of-the-commercial art models perform well with zero-shot feature extraction and image association/captioning tasks, but all models perform poorly with reasoning and quantitative tasks.
Connect 4 is a solved game in the m,n,k family. On a 7 column, 6 row board, players alternately drop a token into a column, attempting to establish four tokens of the same color along any row, column, or diagonal. John Tromp created a labeled dataset of all valid eight ply positions with their perfect play win results. We present a dataset that presents all sequences of plies that lead to each position along with their associated win result class.
Early post-war computing can often seem alien as terminology mixed metaphors and the pioneers brought their own distinct toolsets to the field. Thus, it seems fitting that a powerful influence on computing design would come from a “Martian”: John von Neumann. During the war, von Neumann surveyed computing capabilities within the United States, which introduced him to the ENIAC and the Harvard Mark I. Von Neumann joined as a consultant to the EDVAC project, a successor to the ENIAC, where he and the team worked out the concept of a stored-program computer.
Mark Priestley’s Routines of Substitution: John von Neumann’s Work on Software Development, 1945-1948 is a technical history of von Neumann’s programming work, with special focus on the “meshing” algorithm he wrote as part of a merge sort, his diagrammatic programming language, and the integration and execution of subroutines within a program. While the “Von Neumann Architecture,” as documented in First Draft of a Report on the EDVAC, is the most famous outcome of his work at this time, this study illuminates the invention process and the practical aspects of implementation.
Ludii is a general game system for modeling games and puzzles, although it focuses on historical and traditional games. Ludii was recently used to provide the training data for a Kaggle competition. I noticed their wishlist contained Wumpus World, a common puzzle used for AI training and education. Intrigued, I implemented and submitted a version of Wumpus World.