The future of advanced AI is simple

A big question, one that I certainly hear a lot, is, “What does the future of AI actually look like?” We hear a lot of theory and hype about the power of large language models, generative AI and natural language interfaces, but how is the tech industry going to help the individual enterprise and the individual user realize all this potential?

In the few months since Snowflake acquired Neeva , we’ve been working on exactly that. At our Snowday event last week, we announced Snowflake Cortex, an intelligent, fully managed service that hosts and serves industry-leading AI models, LLMs and vector functions. The Snowday discussions about Snowflake Cortex, which is in private preview, addressed much of the technical capability, and underscored the work we’ve done over the past six months to integrate Neeva’s AI-driven semantic search functionality and language models into the heart of Snowflake. (Read more here.) At a higher level, if someone were to ask me why that’s a good idea, a necessary step, the answer would be all about realizing the full potential of artificial intelligence.

Advanced AI has long been the domain of highly technical scientists. The entire point of the excitement triggered a year ago by the debut of ChatGPT was that the complexity of large language models, the amazing technological achievement that they represent, had been rendered simple enough for a schoolkid to use. While I wouldn’t swear that Snowflake is suited for the average middle schooler, we are following a similar path: making the complex simpler and more accessible to a broader audience. 

And that’s not just one tech company’s product plan. Simplicity is the direction AI development has to go. From many conversations with our customers, I’ve learned that there’s a mental tax for using advanced AI tools. ChatGPT and Bing are pretty easy, but even a widely used image generation tool like Midjourney delivers a higher level of complexity and frustration to the average user. As we talk about bringing new AI tools to the enterprise, whether that be a business analyst interacting directly with data or sales and marketing generating images and copy, we have to cut that mental tax and make things easier for the level of user at hand—who most often will not be a data scientist with matching Ph.D.s in machine learning and statistics, or a developer fully trained in computer programming.

This focus on simplicity is meaningful. As one example: Currently, a relatively basic AI project like building a chat app for customer support is fairly complex. You want the chatbot to surface the answers most relevant to your customers’ problems, so first you have to figure out how to get all the right data into cloud storage, then load it into a vector database. Then you have to create the app that first runs a vector search based on the customer’s natural language prompts, then sends over a call to a hosted LLM service. This is a meaningful amount of development work, stitching together various technologies that an enterprise IT org has to do to deploy that basic chat application. And every step requires attention to (and quite often, reimplementation of) governance, security and privacy issues around the data as it moves along.

Snowflake Cortex lets our users do this in an easier, end-to-end way, in minutes. It applies all previous work around role-based access control and other governance and security requirements. And this is what the future of AI has to be: Easier to use, widely accessible beyond the most technical users, and with all the essential security and governance in place to alleviate the technical burden this can put on central IT and platform teams.

This future is rolling out now. Developers can use AI in seconds and build applications in minutes. Of course, there are more complex applications that would take longer than a few trivial minutes to put together, but those, too, are considerably easier to create than they would’ve been only a few months ago.  

Since Snowflake's beginning, there was a guiding principle that became something of a mantra: Make simple things immediately usable; make the complex things easier and definitely achievable. This squarely applies as we make AI an integral part of Snowflake. We're offering a great balance between usability and flexibility, and that is what we think the future must deliver, not just from Snowflake, but across the tech industry.

At Snowflake, we are proud of delivering a polished, rock-solid, tightly integrated product for data, including AI. It’s what every business and IT leader should expect and demand as they work to bring these new technologies into their own enterprise.

For more on Snowflake Cortex and where AI is headed, watch me with Snowflake Co-Founder Benoit Dageville and SVP of Product Christian Kleinerman in this short segment from Data Cloud Now.