All Segments
Office Hours with Baris Gultekin: The Era of Enterprise AI is Here
22 segments available
Join us in discussion with Baris Gultekin, Head of AI at Snowflake, about the rapidly evolving AI stack. 01:26 Journey to Snowflake 02:32 Snowflake and AI 06:43 Choosing your model 07:44 Snowflake & OS 09:43 Innovations to reduce training data size 10:59 From large to small models 13:14 Snowflake and agentic systems 15:50 AI & data security 17:17 Access control layer 18:14 Embedded applications 19:55 Data sharing 21:37 Snowflake training & inference 23:12 Data reshaping 24:40 Structured versus unstructured model inputs 25:24 Models providing the mean v. exceptions 27:19 Vector databases 30:33 Summary
Segments Timeline
Baris Gultekin's Journey to Snowflake
Baris Gultekin shares his transition from Google to Snowflake, detailing how his startup, Nxy Z, was acquired. He reflects on his long tenure at Google, particularly his role in leading the Google Assistant product team, and discusses the rapid evolution of AI technologies over the past few years.
"having me it's great to talk to you it's been a while I joined snowflake about a year ago through an acquisition I had a startup called nxy Z and snowflake acquired our company and I came in and now I..."
Snowflake's AI Strategy
Gultekin explains Snowflake's approach to integrating AI capabilities directly where data resides. He highlights the importance of large language models (LLMs) and the Cortex product, which enables customers to leverage AI for various data analyses and applications, enhancing their data utilization.
"trust snowflake with uh their crown jewels their data and what we've been working on is how do we bring uh the world's best AI capabilities onto where the data is and and and the moment we do that eve..."
Choosing the Right AI Model
In this segment, Gultekin discusses the importance of selecting the appropriate AI model for enterprise needs. He emphasizes the focus on SQL instruction following and complex tasks, which are critical for Snowflake's customers, and introduces the Arctic model designed for enterprise intelligence.
"sense so touching on Arctic for a second when is the right time for a customer to use Arctic compared to another model yeah so we've built Arctic with a goal of focusing on what we're calling Enterpri..."
Snowflake's Open Source Evolution
Gultekin elaborates on Snowflake's strategic shift towards open source, particularly with the Arctic model. He discusses how contributions from the open source community have influenced their AI developments and the benefits of transparency for customer trust and business growth.
"so one thing that that we haven't historically or I haven't historically thought about snowflake is open source right and Arctic is open source Polaris I think is will be open source and you mentioned..."
Innovations in AI Training
This segment focuses on the innovations that have allowed Snowflake to reduce training costs for their AI models. Gultekin explains the architectural advancements that enable efficient training and the significance of these innovations in the competitive landscape of AI.
"sense and it seems the the emphasis has really picked up there and it it seems like the advances that you've made you mentioned it briefly but your ability Arctic I think was trained on significantly ..."
Choosing Between Large and Small Models
Gultekin discusses the decision-making process for customers when selecting between large language models and smaller models. He highlights the trend of starting with large models for proof of concept and then optimizing to smaller models for production, emphasizing the importance of cost-effectiveness in AI deployment.
"inference cost is a really big deal right a lot of customers I think are a lot of users are starting to see some of these bills and that is driven we have these Mega models like the one we were just t..."
The Role of Agentic Systems
This segment focuses on the emergence of agentic systems in AI, where multiple models work together to perform complex tasks. Gultekin introduces Snowflake's Cortex Analyst product, which utilizes these systems to answer intricate business questions, showcasing the evolution of AI capabilities in data analysis.
"effective llms and that's what doing yeah it makes a lot my mental model for it has been these really large Mo the large Mega models they're basically compressed representations of a big search index ..."
Snowflake's Role in Financial Applications
Gultekin explains Snowflake's pivotal role in the financial industry, where it serves as a trusted data cloud. He discusses how agentic systems leverage Snowflake's data to enhance financial applications, emphasizing the importance of easy access to trusted data for effective orchestration in complex workflows.
"SL for another application yeah exactly saves time to to start with the large one without having to worry about optimization but then of course you R rote of all evil right isn't there that computer s..."
Data Security and Governance in AI
In this segment, Gultekin addresses the critical aspects of data security and governance as companies transition from proof of concepts to production. He highlights Snowflake's approach to ensuring data security within its ecosystem, allowing customers to manage access controls effectively while integrating AI solutions.
"way I see uh snowflakes role here is of course we use identic systems ourselves as I said with things like cortex analyst but we're also seeing more and more our customers are starting to use these ag..."
Access Control in AI Systems
Gultekin discusses how Snowflake's access control layer works in conjunction with AI models like Cortex. He explains the importance of respecting granular access controls and how organizations can govern their AI models to ensure that only authorized users can access sensitive data, enhancing security and compliance.
"there's data risk right that like one of the key parts of any data security infrastructure is data loss preventing data loss and data leaders have been thrust into this position now where the board an..."
The Power of Embedded Applications
Gultekin shares insights on the advantages of building software companies on top of Snowflake. He discusses how embedded applications benefit from faster sales cycles and higher growth margins, as customers leverage existing Snowflake infrastructure. This partnership fosters innovation and efficiency, allowing companies to drive value while ensuring data security.
"offering for anes ex sense yeah yeah that's awesome that's cool then you have a lot of confidence that you're doing things the right way right as long as you're within the ecosystem let's talk a littl..."
Data Sharing: A Core Component of Snowflake
This segment delves into the significance of data sharing within Snowflake's ecosystem. Gultekin explains how data sharing enhances collaboration between companies and within organizations, making data more accessible and valuable. He emphasizes that as companies share data, they can leverage AI to extract deeper insights, thus enhancing their overall data strategy.
"for customers can we talk a little bit about how that's evolved is that still a key priority for Snowflake and then how does that interface with AI is there an overlap there that you're seeing today i..."
Transforming Data Sharing for AI Insights
Gultekin discusses the evolution of data sharing in the context of AI, highlighting how Snowflake facilitates secure access to diverse datasets. He illustrates the challenges of traditional data sharing methods and how Snowflake's cloud-based approach simplifies governance and access, making it easier for organizations to utilize AI for training and inference.
"sharing right if I think about like the story of the IPO the Salesforce investment if we talk to and talking to different customers the data sharing component of a financial institution wanting to sha..."
Simplifying AI Training and Inference
In this segment, Gultekin outlines how Snowflake streamlines the processes of AI training and inference. He emphasizes the ease of fine-tuning models and the integration of search capabilities, allowing users to query data effectively. This functionality empowers organizations to harness the full potential of their data, driving innovation and efficiency.
"inference side right because and so what I have to imagine is there are companies that that will develop or fine-tune their own models maybe they'll do it on Snowflake and so they'll want access to di..."
The Role of Data Reshaping in AI Workloads
Gultekin explains the importance of data reshaping for AI workloads, both for training and inference. He discusses how Snowflake enables users to prepare data efficiently, allowing for the extraction of valuable insights from unstructured data. This capability enhances the effectiveness of AI models, making it easier for organizations to leverage their data.
"application directly that makes sense do you find within the world of snowflake is there a lot of data reshaping that's required for AI workloads either on the inference or the training side to be use..."
Harnessing Structured and Unstructured Data
In this segment, Gultekin highlights the balance between structured and unstructured data in AI applications. He shares examples of how organizations can utilize both types of data to enhance productivity and decision-making. The integration of diverse data sources allows for richer insights and more effective AI solutions.
"yeah exactly if you think about the fraction of structured versus unstructured data coming into some of these machine learning models do you have a sense is it 5050 6040 75 25 in terms of usage in ter..."
Seeking Exceptions in Data Analysis
Gultekin raises an intriguing point about the limitations of large language models, which often average outputs. He expresses a desire for models that can identify exceptions and surprising insights, emphasizing the importance of understanding unique customer behaviors in data analysis.
"them you see both of them yeah I think the the core like knowledge based search applic blows my mind we deployed an internal chapot and the summarization of large amounts of content and synthesizing i..."
Seamless Analysis with Snowflake
This segment highlights how Snowflake enables seamless analysis of both structured and unstructured data. Gultekin explains the integration of SQL queries with large language models, allowing users to extract specific information efficiently and effectively.
"us it's customer references we're interviewing a startup okay or we're trying to understand like why does somebody use Salesforce why does somebody use HubSpot what are those exceptions have you seen ..."
The Role of Vector Databases
Gultekin discusses the significance of vector databases within the Snowflake ecosystem. He explains how Snowflake supports vector operations and partners with other database solutions, emphasizing the importance of combining vector and traditional search methods for enhanced data retrieval.
"you see customers actually using streamlet where you've got the data is coming in you're running large language models on top there's some output that's coming out and then they're building applicatio..."
Building Applications with Data Gravity
In this segment, Gultekin elaborates on Snowflake's role as a central hub for data, enabling customers to build applications easily. He discusses the importance of data gravity and the tools available for users to perform data analysis and create applications efficiently.
"so that you can do you can do rag very easily with high quality combining vector and keyword based Tech search and then ranking afterwards that ability to do this inside snowflake where data is very m..."
Navigating the Divide in Data Analytics
Gultekin addresses the ongoing debate among data leaders regarding centralized versus decentralized data analytics. He highlights Snowflake's approach to empowering users while ensuring data integrity through verified queries, balancing flexibility and control.
"Snowflake and then another big push is and we've heard this also from data leaders there's this kind of divide within the world of data leaders about pushing out analytics to the edge and some data le..."
The Future of Data and AI
In the closing segment, Gultekin shares his excitement about the rapid advancements in AI and data analytics. He reflects on the transformative potential of these technologies and the evolving landscape of data interaction, emphasizing the importance of staying ahead in this fast-paced environment.
"valid by the data team interpreted the right way normalized for whatever it needs to be normalized for yeah that's really smart it's wonderful to hear the vision from you as at the center of AI I thin..."