AI startup Reka founded by researchers from Google, DeepMind, Baidu, and Meta, has announced its new multimodal AI assistant, Yasa-1. The AI assistant is capable of understanding not just text but also images, short videos, and audio snippets. Yasa-1, currently in private preview, can be customized on private datasets of different types, providing businesses with an opportunity to create new experiences for various use cases.
Yasa-1 Features and Capabilities
The assistant can provide answers with context from the internet, process long context documents, and execute code while supporting 20 different languages. These features directly compete with OpenAI's ChatGPT, which recently received its multimodal upgrade supporting visual and audio prompts.
Yasa-1 uses a unified model developed by Reka for its multimodal comprehension. Users can use Yasa-1 to combine text-based prompts with multimedia files to get more specific responses. For instance, it can be prompted with an image to produce a social media post promoting it or used to detect a particular sound or its origin – an instrument, a machine, or an organism. The AI can even understand videos, its topics, and predict the next action of the subject, which makes it very useful for video analytics.
Yasa-1 supports continued conversations by appending the model turn to the conversation history. For instance, after asking about an image's content, users can further inquire about specific details, like what the subject in the image is doing.
The current version of Yasa-1 is only recommended for audio or video clips no longer than one minute, needing further customization for intricate details discernment in multimodal media. Reka has warned users about Yasa-1's tendency to hallucinate, advising against sole reliance on it.
Future Plans and Competitors
Among other things, Yasa-1 can process long context documents, support 20 different languages, actively execute code, incorporate the latest web content into its answers, and connect to commercial search engines in real-time. However, Reka has cautioned that there is no assurance that the assistant will find the most related documents for a specific query.
Despite these current challenges, Reka has laid out future plans to provide more businesses access to Yasa-1 and improve upon its capabilities. A goal underscored by the startup's long-term mission to create a world where superintelligent AI works alongside humans to solve significant challenges.
Reka is competing in the AI space with companies like ChatGPT-maker OpenAI, backed by Microsoft, and Anthropic, backed by Amazon. Other competition includes Inflection AI, with almost $1.5 billion in funding, and Adept, with $415 million.