Netflix Trials AI-Powered Search Functionality

Streaming giant Netflix is currently experimenting with a new AI-driven search feature designed to enhance user experience by incorporating a variety of new criteria.

According to reports, the innovative search functionality will allow users to receive content recommendations based on their personal mood. This feature, powered by artificial intelligence from OpenAI, is presently available as an opt-in for select users in Australia and New Zealand, with plans for further testing in additional countries likely forthcoming.

While neither Netflix nor OpenAI has disclosed specific details regarding the mechanics of the AI search, Sudarshan Lamkhede, a researcher at Netflix, has publicly acknowledged the project through a LinkedIn post. The new search feature appears to rely on a large language model (LLM), or a closely related technology, which would enable it to handle more specific search inquiries than previous iterations.

This development aligns with a recent article authored by Lamkhede and colleagues on the Netflix Tech Blog, where they discussed the creation of a foundational model for Netflix recommendations. The authors highlighted challenges such as high maintenance costs, difficulties in transferring innovations between models, and limited temporal tracking of user interactions.

Inspired by large language models, Netflix is adopting a data-centric approach that employs semi-supervised learning. This model processes extensive user interactions through a specialized tokenization method, allowing for the identification of significant events enriched with relevant contextual and content information. However, it remains unclear whether the AI will assess user mood autonomously or if users will need to input their mood directly via prompts.

Discussions surrounding the new AI search feature have emerged on platforms like Reddit, where initial reactions have been mixed. Some users express skepticism regarding previous search limitations, such as inadequate filtering options. For instance, one user noted challenges in searching for films from the 1980s, while another expressed a desire for filters based on release years or IMDb ratings. Whether the new foundational model will address these concerns remains to be seen.