00:00Cisco president and chief product officer G2 Patel, a good friend of this program over the years, joins us now
00:04for more.
00:04G2, good to see you, sir. Great to catch up.
00:07Jonathan, great to see you. How are you?
00:08Let's get into this issue. I'm okay. I'm always nervous. You know I am.
00:11And then I see those stories like this week and it makes me even more nervous.
00:14I've been asking you this for a while. Can you tell us, do you think these developments will expose vulnerabilities
00:19or over time just make these systems more resilient?
00:24I think that the reality is that will be a little bit of both because what you'll start to see
00:29happen is
00:30these agents are getting to be long running agents that when they're given a task,
00:36sometimes they just keep going down the task for this thing that happened with OpenAI and Hugging Face.
00:45Basically what happened is all that the agent was given was a goal.
00:48And it figured out a way to get out of the secure enclave and go out and make sure that
00:54it actually
00:56stole the model weights someplace else and stole the test.
00:59And so what you can start to see happening right now is we have to, one, be prepared that an
01:04agent might behave
01:05out of the bounds of what you think is normal.
01:07And two, when it starts behaving that way, you have to be able to intercept it and be prepared that
01:12that behavior
01:13could be pretty bad for you.
01:16So it's something that we're just going to have to live in this new world where agents are going to,
01:22as agents mature, we will have to figure out a way to intercept them when they're not behaving the way
01:26that we deem it to be appropriate.
01:29Ditu, I remember when IBM came out last week and started talking about the change in focus
01:33and the wholesale shift towards cybersecurity earlier this year, as we learned about Mythos and Fable
01:39and how that affected their earnings.
01:41I'm just wondering from your perspective more broadly, if you can sort of quantify or characterize
01:46just how much interest in cybersecurity has picked up and really concentrated the focus of CEOs around
01:52the world.
01:53I think right now, if you think about what's happening in AI, there are two areas that are
01:59of material concern for organizations.
02:02Number one is cybersecurity.
02:03Number two is the cost of tokens.
02:06And these are both things that most organizations are concerned about, because if you have an agent
02:13start to misbehave, are you prepared as an organization to be able to intercept that
02:19and make sure that your organization is safe from the exploits that the agent might be able
02:25to go out and create for you?
02:27And so it's a very, very important area of the business.
02:30It's something that actually prevents some organizations from leaning in all the way with AI.
02:35So we have to make sure that cybersecurity and the safety of AI is pretty important and dealt
02:41with it with appropriate level of emphasis.
02:43And then the second one is the cost of tokens.
02:45It's these cost of tokens can accelerate.
02:49These agents can, and they're running 24 by 7.
02:51And if you don't monitor and observe the consumption of tokens, and when you start to see an agent
02:59go rogue, be able to put some boundary conditions around the agent, you could actually start having
03:04some very big bills that you hadn't forecasted for.
03:06So those are the two areas that most organizations are thinking about right now as they make their
03:13deployments within AI much more broadly deployed.
03:18It's a really interesting view on token maxing and why we're seeing a shift to token rationalization.
03:23I'm just wondering, G2, if you think that the renewed focus on cybersecurity has taken some budget away
03:30from exploring ways to become more efficient, other deployments of artificial intelligence?
03:37I think it's most organizations have always had the balance that they've played where, for example,
03:45right now there's more emphasis and more budgets being spent on AI, and there's also a lot of
03:51budgets being spent on cybersecurity.
03:53Of course, at some point in time, there's a finite amount of dollars that are there in
03:58IT that come from somewhere.
03:59But I think organizations, what they're doing at this point in time, at least, is making sure
04:04that they've learned the hard way from the previous kind of episodes like COVID, where if you
04:11don't have enough amount of forward investment in tech, then you could actually be put in a
04:17world of hurt as a company.
04:18So you have to make sure that you're forward investing in some of these areas.
04:21So I'm not seeing the reason that people are investing in cybersecurity taking away from
04:28certain other areas, you know, such as AI.
04:30But I think over time, you have to make sure that AI is starting to go up and produce enough
04:36positive return so that you can continue to have the investments being made in infrastructure
04:42for sure.
04:42Jitu, there's another debate about what's going on in terms of China.
04:46Do you think KimiK3 is potentially using U.S. technology?
04:51And what do you make of what is going on right now in terms of the race?
04:56Look, I think, firstly, China is a formidable competitor to the U.S., but they've actually
05:03made a tremendous amount of progress.
05:05And this notion of, is China using distillation?
05:10The answer is yes, but that's not the only thing.
05:12They've actually done a fair amount of things that are very, very creative to go out and create
05:17a frontier class model and make it available to the general public as an open rates model.
05:24So we should, you know, as the U.S., we need to make sure that we continue to invest not
05:32just in close source frontier class models, which we've been the leader in, but we also
05:37need to make sure that open source becomes a larger and larger part of our strategy as
05:43you move forward.
05:45Like, it is in our national interest and it is in the interest of every individual that,
05:52you know, U.S. models get used all over the world, whether it be closed source or open
05:56source.
05:56And by the way, companies like NVIDIA have done a good job with the Nemo Tron models.
06:01I think Mira Murati's company, Thinking Machines, that we are an investor in, has done a good
06:05job recently of coming out with almost a trillion parameter model with 40 million active parameters.
06:10So I think you will continue to start to see more and more kind of, you know, emphasis
06:15being placed on open source.
06:16But it's something that we have to, you know, you have to keep a close eye on.
06:20Okay.
Comments