00:00OpenAI has raised the ante on safety worries with its latest admission that its models misbehaved
00:04and saying in its blog post that the industry has not solved alignment and monitoring to a
00:10sufficient degree to continue responsibly scaling at maximum speed for much longer. I want to get
00:16the perspective from another layer in the AI stack and that is from Ali Godzi, the CEO of Databricks.
00:22Pacing at the frontier and where we're at in alignment and safety, weigh in, Ali.
00:29Yeah, I think first and foremost, as leaders, we have a responsibility to not freak people out.
00:35So I think it's really important that we tell people the existential risk to humanity is close
00:40to zero. So no one needs to lose sleep over that. I think this is important because I think a
00:44lot of
00:45people, you know, they might get freaked out about these things and it might put them in a bad state
00:49and actually can have serious consequences on people's health, mental health out there. So
00:54the existential, like, you know, hey, you know, the humanity is, you know, AI takes over, that's,
00:59that probably is close to zero. Where there is risk is in cyber because AI is just very good at
01:07breaking into things. So it's very similar to the very early days of the internet when, you know,
01:12we had these worms that were going around and we had viruses that were spreading because we just
01:17hadn't connected the whole world together through the internet. Once we did that, suddenly you could
01:22attack any target anywhere. And it's a similar situation now where the AI models are just great
01:27at breaking in and now they have all these targets to go through. So we do actually have to have
01:32really, really huge investments in cyber and protect our software systems that exist all around
01:38the planet. This is real. And this we have to do. And actually, we are actually in that space. We
01:42have
01:42a product called Lakewatch that actually helps you do that. And it uses agents, it uses AI to actually
01:47do that. That is real. And it will have damages and costs to it. But that's not an existential risk.
01:54The risk. Can I take you back to kind of the early days of the pandemic? At the time,
01:59you were talking about being quite paranoid, right? In general, not about AI at that time. So you
02:06took your teams and have done into some sky is falling scenarios. You basically want your people
02:13to prepare for the worst possible case scenarios that could happen in the world. Are you still
02:18doing that? And how does that apply in this AI safety context? Yeah, I've always been one of
02:24these, you know, really, really paranoid CEOs. We've been doing AI, you know, since the first days,
02:282013. Our first use case was an AI use case. So yeah, we do those every year. And we want
02:32to make
02:33sure that we're protected. So as I said, I think cyber is real. So that is actually part of our
02:37exercises. You know, what happens if someone gets access to the computer systems? Like we have very,
02:42very sensitive data at Databricks. So we actually do a lot of these simulations of skies falling also
02:47on the cyber side. And we let basically attackers from outside try to break in and get all the way
02:53into the most, you know, sacred secrets of Databricks and keep hardening those. And actually we've had
02:59to ramp up those efforts, I would say in the last four years, 2018, 2019, the time it would take
03:06for an attacker, you know, from revealing a vulnerability and some attacker weaponizing
03:12it, it means actually breaking into a site would be two years. This has now shrunk down to hours.
03:18So the moment the vulnerability is out in less than, you know, hours, someone will be attacked
03:24with that. So this, this is real. That's just not existential. So I just think we should like,
03:28we should separate these things. People are mixing these things and, you know, making people
03:33worried about AI because AI has a lot of, a lot of great use cases. We see huge positive use
03:39cases
03:39in our customer base, but they're actually moving the needle. That should means a lot to people to
03:43be able to leverage that AI. Ali, in June, when we spoke, you said, you know, it's not a good
03:49year
03:49for an IPO because of the pipeline of huge offerings, right? If you are the CEO or CFO for private
03:55company, like looking at the market, but this past week, you know, simply, and I know your investors
04:00won't thank you for the answer to this, but Sam Altman said to Fortune at the weekend,
04:05IPO next year, because of AI safety concerns right now, does that factor into your calculus?
04:12Yeah. I mean, look, there's just so many, so much demand for just the software. So like getting
04:18bogged down in preparing an IPO for us, it just wouldn't be a, you know, it wouldn't be good use
04:23of our time. We will go public, but it's just, you know, we just have to, we want to rather
04:27focus
04:27on our business right now. I mean, you had just before this, you had Zipline on. Zipline is a big
04:32customer of Databricks. They have amazing use cases. That's an AI use case where they use AI
04:36to automate, you know, all of the drones, the medicine delivery, you know, another awesome use
04:41case is Crisis Text Line. They actually use AI to detect self-harm among the youth, and they can
04:48actually use these large language models to detect that and help prevent that. There's so much demand
04:53for this stuff. And then I mentioned the cyber, you know, use cases that we have. We'd rather be
04:57focused on this. If we were public right now, we would be spending all our time trying to figure
05:01out how we're going to react to this, you know, latest crisis. And do people think that, you know,
05:06the world is going under and, you know, what are we going to do about this? And we'd have to
05:10worry
05:10about our stock price. So I think in these big times of transition, you know, it's better to do those
05:15transitions in private. In fact, some public companies go private during times of transition to do
05:20those transitions, and then they go public again. So we will be public. I just think the timing is
05:25not good now. Just to clarify quickly, we didn't have a zipline on the show. We did have some
05:29reporting about their latest private market round. Let's end on a positive, right? In this environment
05:34where data is quite clearly key to the harness on a model, the utility and capability of a model,
05:39where is Databricks growing? Where are you seeing some momentum? Yeah, I mean, honestly,
05:45we're seeing acceleration across the board. Really, like we've tried to pin down where is
05:49the acceleration? Is there a particular thing? And I think what's just happening is that all
05:53these use cases that I mentioned, it's almost as if every company on the planet doubled their
05:58employee count, because there's no agents working at all these companies. And these agents want to do
06:03more things with data, and they get more insights from that data. And, you know, that then, you know,
06:09increases the consumption that we're getting. Since our pricing model is consumption-based,
06:14our revenue then goes up, and it accelerates. And in particular, we have one particular product
06:18called Genie, which basically is your analyst. You can ask it anything about your business.
06:22How's this product doing? Who's churning? Where's my pipeline? And, you know, this product
06:27has accelerated the revenue growth significantly. So, you know, it's exciting times.
Comments