Reka, an AI startup founded by researchers from DeepMind, Google, Baidu, and Meta, has announced its latest innovation in the form of Yasa-1, a multimodal AI assistant. Yasa-1 goes beyond text understanding and can analyze images, short videos, and audio snippets.
Yasa-1, currently available in private preview, can be customized to work with private datasets of any modality, enabling enterprises to create new experiences for a wide range of use cases. The assistant supports 20 different languages and has the ability to provide answers with context from the internet, process long context documents, and execute code.
This development by Reka puts Yasa-1 in direct competition with OpenAI’s ChatGPT, which recently received its own multimodal upgrade. OpenAI’s ChatGPT now supports visual and audio prompts alongside text.
Yi Tay, the chief scientist and co-founder of Reka, expressed his pride in the team’s achievement in transforming an idea into a fully functional product in just six months.
Yasa-1 leverages a single unified model trained by Reka to deliver multimodal understanding. It can comprehend not only words and phrases but also images, audio, and short video clips. This unique capability allows users to combine traditional text-based prompts with multimedia files to obtain more specific answers from the assistant.
For example, Yasa-1 can generate a social media post promoting a product when prompted with an image of that product. It can also detect specific sounds and their sources. Moreover, Yasa-1 can analyze videos, identify the topics being discussed, and predict future actions. This feature is particularly useful for video analytics. However, Reka acknowledges that there are still some limitations and recommends that audio or video clips be no longer than one minute for the best experience.
In addition to its multimodal capabilities, Yasa-1 offers other features such as support for 20 different languages, processing of long context documents, and the ability to execute code. The code execution feature is exclusively available for on-premise deployments and enables users to perform arithmetic operations, analyze spreadsheets, and create visualizations for specific data points.
Furthermore, Yasa-1 provides the option to incorporate the latest web content into its answers. By connecting to various commercial search engines in real-time, the assistant can utilize up-to-date information without any cut-off date restrictions. However, Reka notes that there is no guarantee that the assistant will fetch the most relevant documents as citations for a particular query.
Despite being a new player in the AI race, Reka has a core team with researchers from reputable companies like Google and Meta. The company came out of stealth just three months ago and secured $58 million in funding from DST Global Partners, Radical Ventures, and other investors. However, Reka faces stiff competition from well-established players such as Microsoft-backed OpenAI and Amazon-backed Anthropic. Other notable competitors in the AI space include Inflection AI, which raised nearly $1.5 billion, and Adept, which has $415 million in funding.
In the coming weeks, Reka plans to expand Yasa-1’s availability to more enterprises and work on enhancing its capabilities while addressing its limitations. The company aims to create a future where superintelligent AI works alongside humans to solve major challenges.
While Reka has a promising foundation and a team of talented researchers, it will be interesting to see how the company competes against deep-pocketed competitors and navigates the rapidly evolving AI landscape.
VentureBeat is committed to providing a platform for technical decision-makers to gain knowledge about transformative enterprise technology and transact. To stay informed about the latest developments, you can subscribe to VentureBeat’s Briefings.