Zoom has filed a patent for a system called the “context similarity detector” to improve the training of its AI models. The goal is to address the issue of “data context mismatch,” where models are trained on irrelevant data, causing delays in development. The detector compares two data sets — one being tested and another known to be relevant — and assigns a “context similarity score” to determine their suitability for training. If the score exceeds a certain threshold, the datasets can be used for AI model training. This eliminates manual labor-intensive tasks such as visually inspecting qualitative data or listening to hours of audio samples.
Efficiently training AI models, particularly those handling audio and video processing tasks, is crucial for Zoom. They recently partnered with OpenAI to introduce AI features like message composition, email generation, and meeting summaries through their platform’s “smart companion,” Zoom IQ. In addition, Zoom aims to enhance video quality in hybrid meetings with a feature called Intelligent Director powered by AI. The company also invested in Anthropic, an AI startup whose chatbot will be integrated into Zoom’s Contact Center product.
However, Zoom has faced challenges alongside its growth. Like many companies during the pandemic, they had layoffs in February 2023, resulting in a 15% reduction in staff. Although revenue is still increasing, it grew by only 7% year-over-year in 2022 compared to 55% in 2021. The latest financial quarter showed further difficulties with revenue growing just 3%, while costs increased significantly by 33%. Net income dropped from $113.6 million to $15.4 million year-on-year.
Despite these challenges, Zoom remains optimistic about maintaining its momentum through high-profile partnerships and innovative AI productivity tools like the context similarity detector described in their patent application.