Vino Duraisamy, Snowflake, on Reliable AI — Interview with Alexy

I am Vinina Disami. I work as a I am Vinina Disami. I work as a developer advocate at Snowflake. Not developer advocate at Snowflake. Not sure if I would call this a professional sure if I would call this a professional highlight, but just last week I was the highlight, but just last week I was the presenter and builder of the keynote presenter and builder of the keynote demos for Snowflake's build developer demos for Snowflake's build developer conference. And honestly, it was so conference. And honestly, it was so refreshing as a data person to be using refreshing as a data person to be using AI to build the data foundation that AI to build the data foundation that powers the AI. You know what I'm saying? powers the AI. You know what I'm saying? Like data for AI, AI for data. And Like data for AI, AI for data. And Snowflake has recently announced this Snowflake has recently announced this product called Cortex code, which is product called Cortex code, which is essentially a natural language CLI mode essentially a natural language CLI mode for folks to really build anything they for folks to really build anything they want with it optimized for data want with it optimized for data workflows really. So getting to use that workflows really. So getting to use that to build the data foundation which in to build the data foundation which in turn would help the same AI tool or like turn would help the same AI tool or like better the same AI tool was very cool. better the same AI tool was very cool. Yes. I'm thinking from my point of view Yes. I'm thinking from my point of view unreliable anything is when I'm unreliable anything is when I'm expecting a certain outcome or like expecting a certain outcome or like certain something from you and then what certain something from you and then what you're giving me is not in alignment you're giving me is not in alignment with what I'm expecting. So with an AI with what I'm expecting. So with an AI agent, I expect a certain outcome or a agent, I expect a certain outcome or a certain result and if I don't get that, certain result and if I don't get that, that disconnect is what I would call that disconnect is what I would call unreliable AI. unreliable AI. I feel like now we are in the age of or I feel like now we are in the age of or the era of agents. So for an agent, when the era of agents. So for an agent, when I see that something is unreliable, I I see that something is unreliable, I really want to inspect the different really want to inspect the different components to it. Right? So is my ask components to it. Right? So is my ask not clear? Is that why it's being not clear? Is that why it's being unreliable? or is it not able to unreliable? or is it not able to communicate the results back to me in a communicate the results back to me in a way that is you know playable for me or way that is you know playable for me or is there something in the thinking or is there something in the thinking or the the whole process that it did the the whole process that it did internally that's making it unreliable. internally that's making it unreliable. So really having that ability to dive So really having that ability to dive deep into the thinking process reasoning deep into the thinking process reasoning and whatnot would make you know more and whatnot would make you know more reliable AI systems. I think this is a reliable AI systems. I think this is a really hard one because if I asked you really hard one because if I asked you five years ago, would you have thought five years ago, would you have thought about Chad GPT and all the craze that about Chad GPT and all the craze that happened right so it's it's very hard happened right so it's it's very hard but I think selfishly how I want the AI but I think selfishly how I want the AI stack to be in the five years is I want stack to be in the five years is I want a lot of voice mode operations like for a lot of voice mode operations like for CLI I'm thinking out loud in my head CLI I'm thinking out loud in my head when I ask hey can you do D for me why when I ask hey can you do D for me why can I just say that instead of having to can I just say that instead of having to do that extra level of processing in my do that extra level of processing in my mind so I've started using this open mind so I've started using this open source package called open super whisper source package called open super whisper I literally like talk to instead of I literally like talk to instead of typing out the prompts. Just love it and typing out the prompts. Just love it and I wish more and more of that happens. So I wish more and more of that happens. So like voice becomes the primary mode of like voice becomes the primary mode of communication for anything AI. Hope.