 Hello, and welcome back to theCUBE's coverage of Databricks' Data and AI Summit. We're in the lake house again, and I'm Rob Streche this time, so I got that in. I'm doing better than I was doing yesterday, and we're going to rock and roll through here. But I'm so excited to be joined by some esteemed guests. We were in listening to the keynotes over the last two days, and been walking the floors and hearing things. I'm here with Doug Henshin, who's the VP and principal analyst at Constellation Research, and also Tony Baer, who's the principal at DB Insight. Thanks for joining me. I think this is just a great timing to get together. Great to be here. Yeah, thanks for having us. So let's start out. I think kind of give me your impression. John and I just did kind of our overview of the keynote from today, and kind of how we felt the energy level has been, how you've thought, just overview of what you think. I've been knocked out. I think this has been a great event. Kudos to them. Feels like a pre-pandemic level audience here. A lot of enthusiasm, a lot of spontaneous applause for the announcements and the demos. Databricks is back in their wheelhouse. I think Generative AI, for the last three years, they've been building up the warehouse side of their lake house and making a case. All this time, Data Science has been their wheelhouse and their strength and their customers are here, where others are making announcements of previews that'll help eventually down the road on AI. The customers here are doing it. This is where it's really happening and they're building generative models today. I definitely think that, I mean for Generative AI, I think caused us all by surprise this year. When we basically did our Cube session, our outlook for 2023 didn't really come up, but that was back in January. And I think it's forced all of us, including Databricks, to kind of rip up their playbooks. Now it's not to say that Databricks is, they're not stopping with the data warehouse. There's some very interesting announcements on the lake house side, with uniform, lake ice cube and all that stuff. But the thing is right now what they really need to do is that their core audience is demanding that they need to get up to speed on large language models in generative AI. And they're pretty much, that's where they're going at the moment. Yeah, and I think that when you start to look at how they've leaned into that, leaned into, I guess you could say the nerdiness of where their roots have come from and kind of got back to their roots and the research, I think that actually plays to that LLM strategy, that Elf Generative AI. And even more kind of when you start to look at it more, hey, how do I have more specific large language models versus generic or general language models as well? That was very much a theme today. The day two keynote, Matai Zaharia had a panel of folks including one of the executives from Mosaic ML, the company that just acquired. And he was very much talking about this idea of building on big models or building small focus models. Do you really want to pay for a really big model where it has knowledge of Taylor Swift and LeBron James, or do you want to develop something specific to your company that's focused on your business need? I thought that was a great point. For this one, I think we're going to come up with a new term in about another six months, which is small language models or medium language models. Because basically, I think in most, I mean, just from the logistics, and you and I were talking about this during the keynote, is that when you look at the compute, and you were showing me some stats on like the carbon footprint of all this, it just gets, it makes basically Bitcoin, basically mining look almost trivial in the long run. So I think in the long run, what we're going to be seeing, first off, it's going to be very purpose-built models that are very domain-specific on one hand. And in areas where you need broader coverage, you're going to have a portfolio of models, and you'll be kind of like sort of mixing and matching to basically get the piece the answer together. Yeah. I think one of the professors mentioned that even a small model over its lifetime has the carbon footprint of five automobiles over their lifetime. Pretty scary thought, also some concerns raised about the security of models. So some sobering insights here today, but certainly the theme that companies are want to build their own models, whether they have hundreds or thousands, that's another yet to be seen. No, and I think what we were talking about was that really bloom, the bloom model, which is one of the smaller models out there, had 25 tons of CO2 to train it, which if you look at that, I think it's 41 tons, is like the average lifespan of an average U.S. house produces 41 tons. So it's like half the average life cycle of an entire house, U.S. house that is. I'm sure we do more than some other parts of the world. So I think, again, your mileage varies with that, but you start to look at this. I think to me, that was almost an underrated discussion that happened with MIT and Berkeley folk. Again, looking at this, sustainability of this, because I think that, I kind of said, called them instead of smaller or mediums, I called it specialized language models, because they're going to be focused on, hey, I'm an insurance company. I'm not going to put, and having come out of that industry a long time ago, where I built a grid that did actuarial models, when you start to look at that, I'm not going to put that into a large language or a foundational model. I'm going to have a specialized model for going and doing that, where I can really tune it to my parameters for what I think is my IP, right? And I think that's what they were getting at today, which is why I thought it was, I think again, it was definitely a community day today, I think it was a very different open source and community and here's our partners, here's how we're going to go to market. One of the things I'd love to get your reaction is, the Delta sharing being picked up by Twilio and Oracle. Twilio has been so closed with their data, in their customer data platform, their CDP, which is segment, Oracle, I mean that just, I don't know if it's mind blowing or not, I just found it really interesting that the Oracle isn't getting a lot of press around that. No, no, the problem is that Oracle just does not, and it's kind of, I mean, they don't really have the street credit in this community, they have to kind of build it up, and I give them credit that they are making that effort. I mean, I follow Oracle very closely and that they really have, at this point, have embraced the lake house, but what's interesting, and they actually kind of surprised me on this one, they initially announced support for Delta sharing, so I thought that, well, Databricks is hungry for a partner, and last I checked, Databricks and Oracle were not necessarily enemies, but yet Oracle chose Iceberg. I think that basically, and it's the same thing, same strategy that Oracle's doing with OCI, which is that Oracle's basically been turned up to come to the conclusion that they're not going to be AWS or Azure or even Google Cloud, and so they're basically kind of specializing into more, they come up with a Starlink metaphor, I would say they want to replicate a bunch of AWS locals on it, so I know we're kind of digressing here, but it's the same type of strategy. They don't get the street credit, but the fact is they are learning from the market. I also want to add that I think, Databricks shows how you're just on lake house, and this is something which impressed me, which is that they've realized we're not going to fight religious wars here. Right, yeah. And they know they have to be open, and you know, Iceberg, Delta, I think it was clear today with their uniform announcement that it doesn't really matter, it's all parquet underneath. We can get them out of there together, translate, read and write from each one, so not a shock to see Oracle, but a shocking in some ways, but they know they have to be part of this open world with data. But to that exact point, they had DuckDB CEO on this morning too, which to me, it also says, we're not fighting a database war. It's all about the data's going to be everywhere, how do you bring it together versus, to your point, not a religious war. This is not going down, which AT&T versus BSD versus how UNIX, Windows, Linux, wars that we've seen. It'll be the control plane, it's the control plane that's the war. I think it is. Yes, no question about it. When I did my market landscape research on the lake houses, I was saying, look, right now they each have various different capabilities, but given the amount of community involvement in each of these projects, they're all essentially going to be technically basically, it's going to be a level playing field. What's really going to matter in the long run is the ecosystem. And so I think the uniform was a very, I think it was a very mature move on Databricks' part, realizing that they have their huge, basically ecosystem, but the iceberg ecosystem is even larger. They've realized we need to connect, and there was also the Lake House Federation. Yes. Yeah, I mean, they're very much in their wheelhouse here and I think that their AI and ML strengths are serving them well. We didn't hear a ton about their warehouse push. Bingo. But... A little bit yesterday from JPMC giving that plug, but other than that. I think it was more like we interrupt this program for a little bit of LLM. Well, they did have a session about how they're going to be applying AI to kind of try to leapfrog the performance and cost of current technology data warehouse. But you know, in these competing Lake Houses, you're still coming with your original dance partner for Databricks that dance partner is the data scientist. We were at event earlier this week where Data Warehouse is very much their core base and Snowflake. Absolutely. So before I let you off the stage, grading the keynotes and the session overall, what would you give them? I'm between a B and an A minus. In terms of that, you and I actually differing opinions on the keynotes which is that I really, I felt that the keynotes yesterday were very on point. Today, obviously, by the nature of it need to be more I think sort of meandering. But today was all about the community. But I do think as I said, so I would, as I said, I'm really in between a B plus and an A minus, basically for the presentation, but also for the fact that I think they've realized that they are part of a broader community that even goes beyond the Databricks community. Yeah, I'd give them an A minus. I think there's some differentiators they have that they could have highlighted even more. Their path from private preview to public preview to GA is much shorter than many of its competitors. So many of the announcements they made here we're going to see them in 2023. So they could have highlighted that. There are areas where they're playing catch up a little bit, like on Marketplace, they don't have a way to monetize yet to pay for things on that platform, which is where Snowflake has had a lead. But this has been their greatest data and AI summit I've been to and I've been to many of them. Yeah, I gave yesterday's keynote a B because I thought it kind of got lost at the end, but I thought the demos yesterday were killer. Like I gave those A plus. I mean, the demos showed the business value got technical enough, kept both sides of the audience engaged, really played to their base. I thought today's keynote was really a B plus to A minus from that perspective. And overall, I loved how they talked about the open source and really brought it back to their roots today and back to the community. They had another good demo, the English language SDK, that was first spark. Yes, that was, yeah. Which another big spontaneous plus from the audience. There was a number of those today. And I think that really was, again, they were playing to their base and doing a really good job of it today with the open source aspects of it. And I think that, again, I give the whole thing probably in the A minus for the entire event. I think it's just, like you said, I think it's back to pre-pandemic 12,000 people and 75,000 plus watching online. That's pretty impressive. And the feedback that we've been getting even online has been pretty amazing for people who are seeing it and watching it remotely. So I think, again, that's been fantastic. So, well, I want to thank you both for being on here and really helping me break this down. I think it's always fun. I mean, it was fun sharing notes while we were sitting there and going back and forth. You know, again, I really thank everybody for watching and stay tuned. The cube will be back from the lake house in a few minutes with our next guest. Thanks very much and stay tuned.