6 min Analytics

Snowflake makes AI Data Cloud the brain of any business

Snowflake makes AI Data Cloud the brain of any business

After over a decade of building, Snowflake faces a new era. The company shook up the traditional data warehousing market with the Data Cloud. This data foundation proved to be the ideal environment for many organizations to run analytics workloads. With the developments around AI over the past 16 months, it is also increasingly important to offer a prime environment for artificial intelligence. What is Snowflake doing to integrate generative AI and keep up with the times? We talked about it with the Head of AI, Baris Gultekin.

Snowflake hints at current and future realities with a small addition and change in the name of its core product. Nowadays, Snowflake speaks of the AI Data Cloud. Gultekin indicates that this is simply driven by AI strategy being inextricably linked to data strategy. “AI is driven by data.”

“There are already huge amounts of data, and that amount will only increase. So we believe the best way to take advantage of this great revolution is to bring AI close to the data. The other way around would be very complicated – bringing huge amounts of data out toward the AI,” Gultekin said.

The AI Data Cloud brings the left and right hemispheres of the brain together

Artificial intelligence, then, hinges on a solid data foundation. This means Snowflake must also reinvent itself to keep up with the market’s enthusiasm, Gultekin acknowledges. He outlines that Snowflake is committed to growing as a platform into a complete brain, comparable to how people’s left and right brain hemispheres work. On the left side of that brain, Snowflake is traditionally very strong. That is where the abilities such as fact storage, logic and analytical ability are located.

In the AI Data Cloud, this is the part where the mountain of data resides that the data teams have access to. For example, they can use the data from their role as a data scientist or engineer to build BI dashboards and models.

Ultimately, what you want as a company regarding data is to involve business users. Sales, marketing and HR should also be able to use data and AI. This is where the right brain hemisphere comes in. “It’s where the rest of the business is. That’s where they want to take action based on data, to get insights and information from it,” Gultekin explained. According to Snowflake’s Head of AI, until recently there was still too big a visible gap between these two hemispheres of the brain, but that has now been bridged with the rise of generative AI. As a result, the brain has more or less become one.

Superpowers in the hands of all employees

As far as Gultekin is concerned, generative AI is really the game-changer for perfecting company platforms and strategies. He points to the advances in AI by comparing current developments with earlier instances of machine learning and deep learning. Machine learning was and is a good technology, but it mainly focused on a specific task. With deep learning, we saw AI being taken to the next level by addressing previously unsolvable problems, such as image recognition and speech.

Yet even with deep learning, only highly specific tasks are addressed. That is exactly why generative AI is so transformative: it understands human language and can reason, plan, and take action, among other things.

Basis for a new technological era

“These are fantastic technological advances,” Gultekin observes. “You can do many different tasks with generative AI. It doesn’t mean everything is fully automated now. But I believe everyone will have these superpowers that allow them to solve problems that formerly took a lot of time. A task that required multiple steps before will now have a shortcut. This changes industries completely.”

Especially its general purpose nature, where AI has a broad use case, is the basis for a new technological era, in Gultekin’s view. For example, he mentions a chatbot optimized for a company’s specific domain. Such a chatbot does not need to know everything like LLMs from existing well-known chat services do. As a company, you want to build a chatbot that is cheaper to run and processes prompts based on just your own data. Fine-tuning and making such a chatbot available ultimately ensures that business users really become more involved in data-driven work.

We spoke to Baris Gultekin at the Data Cloud Summit 2024 in San Francisco. A significant update from Snowflake at the conference that further advances the mission described in this article is to lower the threshold of AI app development. In addition, the company unveiled new data catalog Polaris and a renewed partnership with Nvidia.

How can Snowflake help with the new challenges?

While the opportunities Gultekin sees seem realistic, it is also true that new developments often come with challenges. Gultekin acknowledges that companies are also concerned about working with AI. He sees that trust and governance are two typical issues that keep managers awake at night. Basically, the AI Data Cloud is built in such a way that it provides hard guarantees about data handling. “Everything runs only in Snowflake. No data is shared with any third party. Customer data will not be used for training,” Gultekin assures.

In addition to this guarantee, companies have additional governance capabilities within the AI Data Cloud. This allows an organization to control which user can access the model and its underlying data. For example, you can give the entire HR team access to a specific human resources LLM, with only a select group of HR professionals having access to certain data. With such components, additional trust can be built into data and AI.

Risk of hallucination

Gultekin then points us to a second major challenge companies face. With LLMs, there is the risk of hallucination. A chatbot gives the wrong answer, which it just makes up. Many LLMs are made to want to respond to every prompt.

Therefore, in Gultekin’s view, it makes sense to work toward a situation where a system can only give sensible responses, and only when there is reason to do so. “This can be achieved by building a search system that goes beyond semantic search, one that understands that a question cannot be answered without proper documentation. When that is unavailable, it doesn’t give an answer. This avoids the risk of incurring hallucinations,” Gultekin concludes.

The Head of AI indicates that although companies like this way of working, they want additional evaluations regarding these decisions. He is referring to what Snowflake may do in the future with its recently announced acquisition of TruEra.

TruEra’s technology allows it to evaluate models’ input and output and, based on that evaluation, make decisions about perfecting the model. Such an addition could help companies’ data and AI brains get a little further ahead in the age of artificial intelligence. We are curious to see how this will play out.