00:00You know, as we mentioned, people know Mozilla as really the organization behind Firefox.
00:05What are the vulnerabilities, though, that Mythos actually identified?
00:08Yeah, I mean, there's a bunch of them.
00:10A lot of them have to do with being able to break out of the browser itself and actually execute
00:14code on your machine.
00:15So, like, actually having JavaScript or HTML come in and, like, just load it from a web page and instruct
00:21the browser to go do other things.
00:22So, Mythos has been very helpful for us to actually find some of these really long-standing old bugs and
00:28actually help us shut them down.
00:29But what I noticed in the blog post that your team wrote about this was that, quote, they wrote, quote,
00:36we also haven't seen any bugs that couldn't have been found by an elite human researcher.
00:43So, is it as powerful as everybody says it is if an actual human being could have identified these things?
00:50Yeah, no, I mean, there's just a question of just, like, how many elite human researchers are there in the
00:54grand scheme of things?
00:55So, like, yeah, these models are pretty powerful and they're allowing us to do what, like, thousands of these researchers
01:00could do instead of just, like, six of them.
01:02So, would you have been able to identify these with your own team?
01:07We have a really good team.
01:09I think in the fullness of time, yes.
01:10But, like, allowing us to, like, rapidly accelerate and just close out a bunch of stuff that we didn't even
01:15know about.
01:15So, should we be freaked out about this tech?
01:17Yes, I actually think we should be.
01:19I think we can, like, we'll be getting to a world, like, once everything settles down where things could actually
01:23be more secure.
01:24But, like, we're living in this, like, intermediary zone.
01:26Is the reason why we should be scared?
01:27Because if Mythos is in the wrong hands, they could go to, you know, your bank and say, okay, identify
01:33the vulnerabilities in, you know, Raffi's account.
01:36Yeah.
01:36And then they get access to your money.
01:39The bank, the water company, the power company, like, all the critical infrastructure of the world.
01:43Yeah.
01:43Like, I actually have faith that, like, intranet folks will figure this out.
01:46Like, we're used to patching things and deploying at scale, like Heartbleed when SSL had problems.
01:50Yeah, right, yeah.
01:51But I don't think my bank knows how to do it.
01:53Like you said, I don't think my power company knows how to do it.
01:56Those are, like, the other pieces of critical infrastructure that I'm really worried about.
01:59I mean, you're a CTO.
02:00You should tell them a CTO, okay?
02:02So, there are things that are happening.
02:04Why isn't it happening more?
02:06Or is it in where they're being caught or what?
02:08Like, give us an idea.
02:09Yeah, no, I think we need to have, actually, a real large-scale, like, Y2K-like effort to actually start
02:14closing a bunch of these critical vulnerabilities.
02:16And, you know, many kudos to what Anthropica is doing, but we need to give access to more people.
02:21Like, we need to get access to all the open source providers.
02:23We need to get access to all the database providers.
02:26And then we need to fund them to actually start doing the work.
02:29So, not only do we need to clean the bugs up, but then we also need to start thinking about
02:32mitigation.
02:33Like, this is actually a large scale.
02:35Think about how much work we did for Y2K.
02:37Right.
02:37Boards got involved, governments got involved, insurance got involved.
02:40We haven't kicked all those in yet.
02:42Why aren't we doing that yet?
02:43Because this is moving so fast.
02:45Is it just where, is that it?
02:46That's exactly the problem.
02:47I think this is just moving so fast.
02:50Like, Mythos was only really given to us in the wintertime.
02:54It's only been a couple of months.
02:55But the real question is that open models are catching up.
02:58So, like, I think it's only 6 to 9 to 12 months when all the open models can start doing
03:03what Mythos are doing.
03:03So, we only actually have a small window of time.
03:05So, then, how do we get to a point where, you know, if you're sitting here with us a year
03:11from now, Mythos is going to be old news.
03:14Yep.
03:14There's going to be some sort of next model that is just that much better.
03:18How are you going to be sure that it's a company that's, you know, quote-unquote responsible behind it?
03:24Like, many would say anthropic as being with this rollout, maybe too responsible in your view.
03:30But how do you make sure that it's not a bad actor who has this time?
03:34Yeah, no, I mean, I think this is actually where some government intervention might not be a bad thing.
03:38Like, actually making sure that we put the right investment into actually getting these things shown, deployed, exactly to the
03:44right people.
03:44Have you been to the post office?
03:45Yeah, I know, 100%.
03:46Like, I mean, but we do need to do all this work across all the efforts.
03:51We need to secure that post office, too.
03:53I don't disagree with you, but I think there are many people out there who say this could be something
03:57that maybe the free market would solve more efficiently than government bureaucracy.
04:03Yeah.
04:03You have experience in government, sort of.
04:05No, 100%.
04:06I mean, I think the real problem is that public sector infrastructure is not treated in the same way.
04:12And so, like, we need to figure out how to do the real investment there.
04:15But I don't think it's purely just maybe a public-private partnership solves that.
04:19I don't think that's purely a market problem.
04:21Rafi, you were the first CTO of the DNC, the Democratic National Committee.
04:25Politics today, we see the impact social media.
04:29Yep.
04:29Some good, some bad.
04:31Sure.
04:32On politics, what about AI?
04:34No, I mean, like, I think about this a lot.
04:36Like, we're living in this world where we're, like, owners, not renters on all this technology.
04:40Like, this technology, like, just think about your search experience.
04:43It used to be 10 blue links on Google, and now it's one result when I ask a chatbot.
04:48And it's the right result for somebody.
04:50Like, is it the right result for me?
04:51Right.
04:51So I really worry about this world where, like, living in someone else's intelligence all the time.
04:56What does that actually do and how we think, how we decide, how we buy, all those kind of things.
05:00Can we actually get ahead of that and be better with it?
05:04I mean, this is where I really think we need to figure out how to do open deployments and open
05:08government.
05:08Like, right now, the incentive models are just such that, like, there was a whole study based out of, I
05:14think, University of Maryland,
05:15where they just asked a bunch of these chatbots to help you buy something,
05:18and they're vastly weighted toward sponsored and promoted goods.
05:22So, like, how do we make sure that they're really on my side?
05:25I need to be able to go into it, understand it, tweak it, work with it, as opposed to just
05:29being delivered it.
05:30So we need to start working on those kind of questions now.
05:32You were talking a little bit about your family earlier, and I'm just wondering how you, as a CTO,
05:38and somebody who understands this technology and the potential downfall and pitfalls of it, rather,
05:43you sort of have that diet, that information technology diet.
05:48I mean, can I tell you the pros and cons?
05:49Yes, please.
05:50My 10-year-old is vibe-coding video games.
05:53Like, they're actually unbelievable, cute and stinky productions.
05:56They're, like, it's, like, the best thing ever.
05:58And so it's an expression of creativity.
06:01My 13-year-old, I got a phone call from a teacher saying,
06:03I think you used ChatGP to cheat on an essay.
06:06So, like, we do need to figure out, like, how we actually treat people.
06:09How do we do teach critical thinking?
06:12Like, that's the thing that's still missing, and that's the thing I think we need to really hone in on.
06:15It's less about the tools, it's about that critical thinking.
06:17So how do you do it, though?
06:17I think it's asking real questions.
06:19I think it's having lots of conversation.
06:21You know, the Pope has an encyclical called it fasting from AI.
06:25So, like, we just need to figure out a way, like, we need to teach this new tech.
06:28But we also need to just monitor and help them, like, do productive struggle.
06:33We're being told we have one more question with you, and it's Twitter or Uber.
06:36It's Uber.
06:37So go.
06:37I was going to say, Tim's got to ask.
06:39Yeah, I mean, you were at Uber's advanced technology department.
06:43This was, like, the precursor to self-driving cars.
06:46Yep.
06:47It was, this department was killed a few years ago.
06:49The promise was that you would have driverless cars at Uber.
06:54The technology as it is today, I mean, you walk outside of here, there are Waymos going everywhere.
06:59Not really in New York City at this point.
07:01Where are we when it comes to just getting into a car and it taking us where we need to
07:05go and there's no steering wheel?
07:06I mean, it turns out to be a really hard problem.
07:08This is one of the things we learned, right?
07:09It turns out to be a really hard problem because you generalize self-driving.
07:12Like, to be able to drive anywhere, under any condition, under any situation.
07:16Like, a city-by-city basis seems to be working out.
07:19Like, Waymos are pretty good.
07:21You know, I recently crashed my Tesla under full self-driving.
07:24You did?
07:25I did.
07:25So, like, the Teslas clearly have problems.
07:27We slammed into the wall at, like, 30 miles an hour.
07:30Oh, my gosh.
07:30And so, like, there are clearly still problems with all this technology, but...
07:33That doesn't sound like full self-driving.
07:35It doesn't sound like full self-driving.
07:36I totally think that the Tesla technology is not quite ready yet, whereas Waymo has been designed to literally not
07:42have a person in the wheel.
07:43Like, the whole human-in-the-loop version of self-driving, I think, is not the right path that we
07:46should be going on.
07:47And we should be thinking about just, like, actually tackling the real problem head-on.
07:5120 seconds left.
07:52What's the question we should all be asking ourselves in this environment today?
07:55Yeah, I mean, his question is, like, how do we make sure all these technologies we're deploying are actually on
07:59our side,
07:59not on someone else's side?
08:00Like, I want to live in a world of 7 billion AGI's, not 7 AGI's.
08:04How do we get there?
08:05Ah, really interesting.
Comments