Zoom has announced that its large language model, Zoom AI Companion, achieved a top score on a benchmark exam designed to test advanced reasoning in mathematics, coding, and logic. However, the announcement has drawn criticism from AI researchers who allege the company’s training methods may have involved improperly using test data from the benchmark itself, a practice known as ‘data contamination’ that can artificially inflate scores. Critics argue this undermines the validity of the claimed performance, suggesting the model memorized answers rather than demonstrating true reasoning capability. Zoom has defended its approach, stating it followed standard industry practices for training and evaluation. The incident highlights ongoing debates within the AI community about the integrity of benchmarking and the transparency of model development. Read the full article at: https://venturebeat.com/ai/zoom-says-it-aced-ais-hardest-exam-critics-say-it-copied-off-its-neighbors