Snowflake Data Cloud Expands with Multimodal Large Language Models (LLMs)
Snowflake Data Cloud, led by Sridhar Ramaswamy, is advancing its capabilities by integrating multimodal large language models (LLMs). The company has partnered with Reka, an AI startup founded by former researchers from DeepMind, Google, and Meta, to incorporate its proprietary models into Snowflake’s data platform.
This collaboration follows Snowflake's recent partnership with Mistral and will allow businesses utilizing the data cloud to develop generative AI applications that can process text, images, and videos. This integration paves the way for teams to unlock new insights from their datasets.
Snowflake participated in Reka’s $60 million funding round last year but has not confirmed whether it will increase its investment through this partnership. Baris Gultekin, head of product management at Snowflake AI, expressed that the company continually seeks to support partners and enhance customer innovation, though no specifics regarding investment were disclosed.
Introducing Reka Flash and Core in Snowflake Cortex
Since its launch, Snowflake has aimed to establish itself as the preferred data infrastructure for customers. Initially providing a basic data warehouse, Snowflake evolved to support diverse data formats and capabilities, resulting in a comprehensive data cloud that accommodates various AI and analytics applications.
In response to the booming generative AI landscape, Snowflake introduced Snowflake Cortex, a fully managed service designed for developing LLM applications. Cortex offers enterprises a collection of AI building blocks, including open-source LLMs, enabling them to analyze data securely while developing targeted applications for distinct business needs. The company initially focused on specialized LLMs for tasks like sentiment analysis and is now expanding to include two LLMs from Reka: Flash and Core.
Reka Flash is a cutting-edge model featuring 21 billion parameters, optimized to deliver performance comparable to larger models across language and vision benchmarks. Conversely, Core is Reka's largest model, nearing the performance of advanced models such as GPT-4 and Gemini Ultra, but it is not publicly available at this time.
Snowflake plans to integrate the Flash model into Cortex immediately, while support for the Core model is under development for a future release. Although Gultekin did not provide a timeline, he indicated that it would be available soon and mentioned the possibility of adding other Reka models based on demand.
Benefits of Multimodal AI for Snowflake Users
With the integration of Cortex and Reka’s AI models, Snowflake users can create generative AI applications capable of processing text, images, and videos. This functionality supports diverse applications, such as video captioning, image tagging, generating product descriptions for e-commerce, and analyzing graphical data.
Gultekin highlighted several potential applications, including chatbots that interpret charts and marketing content generation for entertainment businesses using their video and image assets.
While Gultekin did not disclose the number of companies leveraging Reka models specifically, he shared that over 400 enterprises are utilizing Cortex and its hosted models to develop generative AI applications. These applications span various sectors, from identifying security vulnerabilities in service tickets to enhancing healthcare provider communications with insurer data.
The addition of Reka’s models will expand the total number of LLMs available in Cortex to a dozen, joining those from Mistral and Google introduced recently.
Gultekin described Snowflake’s AI innovation pipeline as being in "overdrive," with the goal of making AI accessible to all users to drive meaningful business outcomes quickly. He hinted at upcoming announcements of further AI advancements leading up to the annual summit in June.
“Our roadmap reflects the principle that effective AI strategies are grounded in robust data strategies—data is the fuel for AI. We are committed to enhancing productivity, collaboration, and overall efficiency in AI and ML workflows, all built on Snowflake's secure and trusted data foundation,” Gultekin stated.
Notably, Databricks, a competitor in the data ecosystem, is implementing a similar approach. Following its acquisition of MosaicML, the company has introduced open models and improved tools for developing generative AI applications. Recently, Databricks acquired Lilac, an AI startup specializing in analyzing and refining unstructured data for AI training.
Ranking
