Ep 41: Head of AI at Snowflake Baris Gultekin on Why They Built Their Own LLM, Governance as a Moat, and the Most Common Enterprise Use-Cases

"I think any of the main players who are waiting to build platforms are super interesting, especially from a large company perspective. They understand that they are not yet startups but have the opportunity to create significant solutions in AI."

"The application space is just forming, and I think over the next two years when I talk to customers, we're now transitioning from here are a bunch of building blocks to go build whatever you want."

"There are some awesome videos on YouTube that talk about all the things that we do on Snowflake, which make the conversation even more fascinating as we'll see a lot of flowers bloom in the next two years."

"I think people are becoming more comfortable with the technology, but I think there is product innovation needed as well to get the end user to be comfortable with the answer."

"There are still risks for hallucinations, which in some cases, especially in regulated industries, can be quite catastrophic."

"That effort was relatively small compared to some of the others in the market. We started in December or so time frame and we were fortunate to get amazing researchers from DeepSpeed as well as VLLM."

"Meta and Mistral have been hugely influential in the AI sector, particularly for their open source initiatives, which are changing how new technologies are developed."

"When we wanted to have our own model, it was because a lot of our customers really want to build BI-type experiences using AI. And there, of course, text to SQL, natural language from natural language to SQL to data really matters."

"Identifying deep use cases is really important. Startups that can do this will position themselves well in the AI landscape. It's about delivering real solutions to problems that customers have."

"You cannot start as a platform. It requires a solid foundation first, then you can build up from there to create something that users will number among their favorites."

"It's crucial to recognize that when it comes to SQL, there's a lot of complexities involved. We were focused on making the model efficient in terms of training and inference."

"I think the market is massive and it's only going to grow, and the opportunities for startup at various aspects of the stack will be there."

"We're seeing a lot of innovation in the whole stack, so I'm super excited about it because it immediately translates to cost."

""Hallucinations are still a concern. And measurement is still not mature, you know, for many companies. Quality when it comes to more complex scenarios like text-to-SQL is still a concern.""

"What are the categories of like AI infrastructure that you find interesting, or looking at the ecosystem and kind of excited about?"

"When somebody asks a question, how do you know that that can actually be answered? This underscores the necessity for AI systems to understand user context to provide meaningful responses."

""There's a lot of potential, but it's just getting started. We'd love to talk a little bit about how you think about acquiring companies in the AI world versus partnering with them?""

"We should know when not to answer. This highlights the complexity of AI interactions, where users must engage with systems that do not provide misleading or incorrect information."

""You know, when it comes to quality of these systems, either text-to-SQL system or Cortex Search, we have very high quality, you know, systems that our customers can use very easily.""

"You know, you can explore. It’s okay if you’re not right all the time because the analyst knows if you’ve got 80, 90% of it right, I’ll correct the rest. This emphasizes the collaborative aspect between AI and human oversight."

"We see cost as a big factor, and with a decline in market prices, companies are looking for ways to innovate and drive efficiency without compromising effectiveness. Distilling these large models into smaller versions is not just a trend, it's our reality now."

"As organizations grapple with the complexities of AI governance, the focus is shifting to ensuring that data security, governance, and policies are respected, which ultimately facilitates smoother AI implementation and builds trust within companies."

"Having built Arctic in a very efficient way, now it doesn't have to be outrageous to build your own specific custom model for your needs."

"ChatGPT is doing a phenomenal job there. But as all these assistants are becoming more prevalent, they are starting to show signs of utility."

"Sure, sounds good. But then when I'm actually doing something specific at work, I want this AI to be checking whatever."

"So for some small number of customers who are interested in building their own models, pre-trained models can give significant advantages, allowing them to satisfy unique requirements."

"We have very capable models that aren't used as much because the brand recognition on those models isn't there, even though they are super capable."

"When running large models, the straightforward approach can hide complexities underneath. Flexibility in adapting these models for various use cases is crucial for adoption."

"As organizations chase AI talent, the balance between innovation and cost-effectiveness becomes more challenging. Finding the right use case that justifies the investment is key."

"I think evaluation is something that a lot of people are missing. It's often considered underhyped in the AI sphere."

"The landscape of open-source tools and models is rapidly evolving. Companies need to ensure they implement effective guardrails to govern how these models operate and integrate."

What is "Ep 41: Head of AI at Snowflake Baris Gultekin on Why They Built Their Own LLM, Governance as a Moat, and the Most Common Enterprise Use-Cases" about?

In this episode, Baris Gultekin, the Head of AI at Snowflake, discusses the company's unique position in the AI landscape, particularly regarding their proprietary data access and the development of their own large language model (LLM), Arctic. Snowflake is focusing on enhancing AI capabilities for their customers, allowing organizations to leverage AI without having to build extensive infrastructure from scratch. Gultekin shares insights on the challenges of ensuring data quality, the complexities of integrating AI into existing systems, and the importance of governance. The conversation also touches on the transition from proof-of-concept projects to full-scale AI deployments and the vital role of product innovation to improve user experience in this rapidly evolving field.

What are the key takeaways from "Ep 41: Head of AI at Snowflake Baris Gultekin on Why They Built Their Own LLM, Governance as a Moat, and the Most Common Enterprise Use-Cases"?

Snowflake's development of Arctic reflects a strategic decision to enhance AI capabilities directly within their platform, facilitating better business intelligence solutions. The importance of data governance and quality is emphasized, especially as companies transition to AI-driven systems. Organizations are increasingly moving from proofs of concept to full production AI systems, marking a significant evolution in their deployment strategies.

Which podcast is this episode from?

"Ep 41: Head of AI at Snowflake Baris Gultekin on Why They Built Their Own LLM, Governance as a Moat, and the Most Common Enterprise Use-Cases" is an episode from Unsupervised Learning.