Inside Snowflake’s Vision for the “AI Data Cloud”: A Conversation with Head of AI, Baris Gultekin

At the Snowflake Summit 2024 in San Francisco, AIM Media House got the opportunity to personally learn about Snowflake’s artificial intelligence (AI) strategy and products from Baris Gultekin, the company’s Head of AI. The discussion focused on Snowflake’s AI principles, new product releases, ethical AI practices, and the company’s vision for the “AI Data Cloud.”

Snowflake’s AI Principles: Easy, Efficient, and Trusted

Gultekin stated three key concepts that define Snowflake’s AI strategy: simplicity, efficiency, and trust. Trust is essential, since businesses entrust Snowflake with their most sensitive data. Snowflake wants to operate AI workloads safely inside its platform’s security limits, while adhering to access rules and governance regulations.

“For us, having governance at the core of it is really important, and that’s what differentiates us,” Baris stated. Snowflake provides tools to enable customers to build responsible AI systems with guardrails against issues like hate speech.

Ease of use is achieved by bringing AI capabilities directly to where the data resides, eliminating complex pipelines and infrastructure management. “Everything works well together, and bringing AI next to the data achieves that,” Gultekin explained.

Efficiency encompasses cost-effectiveness and high accuracy. Snowflake has developed research models like Arctic Tilt for document understanding and Arctic Embed for search, which are smaller but achieve higher benchmarks than industry giants.

New AI Product Announcements at Summit 2024

During the Summit, Snowflake announced the general release of Cortex, its massive language model service, which would allow clients to utilise and fine-tune these models. Cortex Search, currently available, offers cutting-edge document search capabilities critical for developing chatbots and decreasing hallucinations.

Cortex Analyst enables business users to ask natural language inquiries about structured data and obtain queryable responses. Snowflake also unveiled Studio, a no-code experience for developing bespoke AI apps, and announced the public release of Copilot (coding assistance) and Document AI (structured data extraction from PDFs and documents).

Responsible AI Across Geographies and Cultures

Addressing the responsible adoption of AI across geographies and cultures, Baris outlined Snowflake’s efforts to ensure availability, governance, and respect for cultural nuances.

While Generative AI solutions are currently available in six regions due to cloud provider GPU limitations, Snowflake is working to enable secure cross-region usage. The company is developing a “guardrails” product to allow customers to set base guidelines on AI responses, such as prohibiting hate speech or violence.

“Over time, we are going to allow our customers to provide their input and fine-tune and customize that experience as well, which then talks to the third part, which is, you know, there are cultural nuances, and those AI respect those cultural nuances,” Gultekin explained.

Snowflake carefully curates the models it selects, considering bias, and will enable customers to provide further guardrails to address cultural nuances.

Partnering for Multimodal AI and GPU Acceleration

In response to a question about multimodal AI capabilities, Gultekin revealed that Snowflake has partnered with Reka, an AI company with highly-performing multimodal models. Customers will soon be able to use these models with text, images, and video inputs within Snowflake.

Snowflake has also deepened its partnership with NVIDIA, collaborating on research and integrating NVIDIA’s embedding models and inference software to accelerate Snowflake’s Arctic offerings.

The Vision: Snowflake as the “AI Data Cloud”

Gultekin emphasised Snowflake’s new positioning as the “AI Data Cloud,” underlining AI’s essential role in its products. “We believe there is no AI strategy without a data strategy,” he said, emphasising Snowflake’s unique position as a trusted data repository that can deliver high-value AI capabilities.

Addressing AI weariness and the hype surrounding generative AI, Gultekin believes AI’s influence is genuine and transformative. He listed instances such as Siemens making 7,00,000 pages of research available via AI and Vimeo utilising AI to comprehend video content and enhance conversion rates.

“For all of our customers who are both tired of hearing about AI as well as intimidated by AI, I think the answer is we’re making it simple, and the impact is real,” said Baris.

Snowflake’s objective is to make AI dependable, easy to use, and well-governed for businesses, presenting the firm as a facilitator of AI adoption and proving its concrete impact on cost savings and revenue growth.

Snowflake’s “AI Data Cloud” concept is to simplify and accelerate corporate AI adoption by delivering AI capabilities directly to the data itself, assuring trust, simplicity of use, and productivity.