00:00:00
Heat. [music] [music] Heat. [music] Hello and welcome to TechTV. Today we have a special debate on AI where we have three guests who are going to be discussing some of the really deep issues around what AI is, how well it is coping at this moment in time and whether there’s a path to AGI using the current LLM models. Now Pete, tell us who the guests are that we have here today. >> I’ve got three esteemed guests in the studio. We have Leslie Wilcox. Professor Leslie Wilcox of the London School of
00:01:09
Economics, author of 80 books on technology in the workplace. He’s an emeritus professor of work technology and globalization. I could actually say so much about the uh things that Leslie uh has written about and it would take me ever such a long time. But his major research interests are service automation, robotic process automation, cognitive automation of knowledge work, artificial intelligence, and so much more. We’ve also got Neil Barrett. Professor Neil Barrett studied mathematics and computer
00:01:43
science at Nottingham University, where he also completed his PhD. He became the UK’s youngest lecturer following his appointment at York University in 1985. He’s been a professor of computer criminology at the Royal Military College of Science, Cranfield University. [snorts] He’s appeared as an expert witness in a number of court cases. But I suppose probably, you know, most fundamentally, Neil was a UK government adviser instrumental in the creation of the UK’s National High-Tech
00:02:14
Crime Unit, which is now part of the National Crime Agency, and he was responsible for a€ 1.2 2 billion euro fine that was levied against Microsoft by the um by the EU. Last but not least is the sponsor of the debate, Tariq Mustafa, founder, CEO, CTO, and the inventor behind Curologyy’s AI paint patented deep AI based compliance technology platform. He’s recognized by the industry as a leading visionary and expert in data compliance and knowledge, data security, posture enforcement, and
00:02:51
Tariq’s groundbreaking innovations in intelligently automated data discovery, classification, risk assessment. Again, as with all of our experts, I could go on and on and on, but that will take us time. We’ll take time away from what we’re going to talk about. So diving straight into the debate, we wanted to kick off with a bit of a level setting around understanding AI. Um, so Pete, I just wondered whether you wanted to kick off with some questions in terms of where we are at today.
00:03:28
Well, I think where we are at today is almost extremely puzzling. There are so many people who are sending press releases to me that are saying that AI is about to take over the world and that the world will never ever be the same again. At the same time, I get a lot of press releases from a lot of other people on the other side of the fence who are saying it doesn’t work and that there are some real fundamental issues with the technology. And I think that that’s the thing that we need to
00:04:00
challenge now. What we have to do is we have to look at what those problems are. I mean many people are likening this to the internet bubble and the internet bubble in with a little more fairness the internet technology seems to have been a lot more settled than the AI technology. So [clears throat] I mean without any more ado uh I’m going to ask a panel what they think. Where do you think we are now Leslie? Um, we’re in the midst of a massive hype on this and we need to bring it down to
00:04:36
earth what AI is and what it can do. It’s very impressive stuff, but it’s not what um everyone is is exactly saying these days. Um, you know, I think the hype is more intense than I’ve ever seen it. I’ve been in the industry since 1980 and the hype is so intense now that it’s not just the pigs that are flying. I like to say it’s the whole farm. Um, and you know, a simple phrase brings it down to earth really, which is if it’s artificial, it’s not intelligent. And if
00:05:11
it’s intelligence, it’s not artificial. Um, and I think we be trapped into a we’re into a language trap that uh pos misleads us. It’s based on a false brain computer metaphor. What we’re finding out in neuroscience these days takes us further and further away from the idea that the brain is anything like a computer and vice versa. Um AI is is it’s not it doesn’t think it doesn’t feel it doesn’t imagine. It doesn’t create it doesn’t empathize no learn like humans. It is it is not a brain or
00:05:55
a mind. It is never going to be in quite that way. It has other characteristics. But what we have here is Rob probably the illusion of of intelligent behavior at best. I think I I’ll stop there because the other two >> are very knowledgeable. >> I was thank you for for for doing my job for me. Uh Neil, uh [laughter] what’s your take on this? Um, I think if we [laughter] define AI as an artificial thing that’s doing something that looks intelligent, we’ve kind of got there. It
00:06:31
does look intelligent. It does do things which would require intelligence were it to be done by a human. So, it qualifies on that basis. It does do some surprising things. It does some clever things. Um, it also does some stupid things and some simple-minded things. And it truly isn’t yet anything near to where it could develop to. I mean, we’ve baby steps just at present. This this really is the beginning of it rather than the middle or close to the end. Um, baby steps, but it’s producing things which are
00:07:08
surprising, useful, scary. Um, I think I would say at this point it’s full of potential. That would be my take on it. >> Rick, you’re sitting there Silicon Valley. You talk to these people all of the time. What’s your take on this? And it’s an it’s a topic that you’re particularly concerned about. >> Yeah. Yes. Sure. Absolutely. I do agree with both Neil and my esteemed colleagues here and I agree a little bit more with Neil. >> [snorts] >> Um the definition see what AI really
00:07:45
lacks right now is a sound scientific definition what AI should mean first of all. Okay. And um the current hype cycle has created hype beyond hypes. This is the third hype cycle that I am experiencing myself. The previous one second hype cycle which collapsed completely during the early 1990s. I was in my PhD program at University of Southern California in artificial intelligence. So I witnessed that collapse firsthand being a student at that time. And now we are seeing the third uh hype cycle which by the way is
00:08:22
also going to collapse. Okay, this is a prediction uh just like the previous two. So why do I say it will collapse? collapse in the sense that there will be some very good results coming out of this hype cycle like we see you know this LM models and those things but keep in mind LLM model was invented in the previous hype cycle machine learning neural networks which was one of the two success stories of the second hype cycle before it collapsed is the main driver engine behind all these LLMs and learning
00:08:57
models and whatnot right so the this also is going to collapse. It will yield some good results. The order of magnitude is going to be proportionately much bigger. Previous life cycle total you know investment must have been like a few billion. This time it is many hundred billion dollars already invested. So yeah it is going to collapse again because of the the the the uh hype which has been has gone beyond the the definition of hype. So what I mean you’ve all mentioned that there are limitations Leslie. Why do you
00:09:35
think so many people are pouring so much money into this? I mean that you know the figures are astounding. I I got an email even this afternoon saying that somebody had had even more billions poured into the particular AI project that they were working with. Well, I think they’ve got more money than uh than they know what to do with a lot of the people pouring money into this and then there are a lot of people following on from the momentum that they develop. But, you know, it’s noticeable that it’s
00:10:05
it’s mainly the technology companies that are are doing this and uh they’ve don’t been known to have done this before. I mean um I can think of IBM and Watson as an example an infamous example of this. I think they got about 5% of the returns they expected from their massive investment in that particular cognitive automation initiative. I think that’s that’s one reason. um it it’s too much money available and not knowing where to to put it and I think there are a lack of other things around for people
00:10:40
to invest in other than technology. No other obvious targets at this stage then now yeah I can remember a long way back you used to be the head of technology for group bull in France. Why are big companies getting so interested in these this technology? Is it an arms race? >> It is an arms race. Yes. And you you got to look at the amount of money, the amount of resources that have been poured into computer centers to support it right now. It’s kind of the only game in town as far as a lot of places are
00:11:19
concerned. um a huge percentage of the overall compute capability, if you want to map it like that, the compute capability, a sizable proportion of that is now aimed at LLMs or you know some some sort of auxiliary processing around that. Um so why are they sending the money there? It is the only game in town. That’s That’s I’m afraid where we’re at. >> Derek, why is this the only game in town? I mean, there you are. You’re in Silicon Valley. You you must be in the fulcrum of all of this. You must be
00:12:00
seeing all of these venture capitalists stalking around with wallets uh ready to rip out and dump loads of cash around. >> Sure. [clears throat] So, very honest response to this one. And um you know, I’m not known for my diplomacy. >> [laughter] >> uh brazen greed naked brazen greed on part of two three set of people venture capitalists Wall Street and entrepreneurs in this these are the three hallmarks of this uh third hype cycle see the previous two hype cycles they were
00:12:39
created by academicians people who really thought genuinely thought that they can do the research and address some tough problems to enable AI. They succeeded in some easier ones and they failed in the remaining ones. First of all, in R&D there is no failure. It was only learning. >> Mhm. >> This one is very different because this hype cycle has been created by people who are not exactly very wellversed in AI technology. You know, on the technical level, they are basically entrepreneurs who have no business being
00:13:14
in the AI business. I’ll be very blunt here. Okay. Very few of them have any knowledge of the fact that AI and ML you know they will do their handwaving uh sitting on the high resen you know chairs in the on the stage waving their hands. Oh we are going to invest in AI and ML. AI and ML for heaven’s sake. It’s like saying algebra and mathematics. Algebra algebra is mathematics. ML is one of the branches learning of AI. Okay, they are that ignorant but they see opportunity because they
00:13:52
have created a new hype cycle over the last seven eight years and now they have to cash on it right. So basically a lot of those uh investors who are investing in this this is a terminology in Silicon Valley called FOMO F M O fear of missing out they don’t even know what they’re investing in it’s the same thing that happened in the internet just prior to the internet bubble circa 1999 in 2000 if you recall >> okay so major thing is basically form of fear of missing out so all these people
00:14:27
who invest their you know inventure capital people they are investing their money throwing like crazy like never seen before billions and billions of hundreds of billions of dollars what if it succeeds am I going to miss out am I my big enterprise going to disappear therefore we see this a massive influx of investment in this space so this is uh greed driven this time previously the two hype cycles they were genuinely people thought intellectuals academicians true researchers that they will solve the problem hard problems
00:15:05
right now it is basically the um fruit of the previous life cycle machine learning and LLM models which is being put to uh you know use this time so this is my very firm opinion and Pete you cannot do anything to change that >> well Leslie Leslie you’ve seen technology introduced right the way across the workplace right the way across uh the the the offices is factories and you you you followed that. So, is AI all things to all people? Is it everything in the way that it’s being presented? I mean, is this the problem
00:15:42
with it that it can only be used in certain places and people haven’t quite realized that as Tariq was just pointing out? >> Well, it’s a it’s AI has become a marketing term, let’s face it. So, it applies to anything. I mean a lot of people who are investing in AI actually aren’t doing AI. It’s quite quite staggering when you investigate some of these companies that you go into. In the same way you go into companies and you say are you doing AI and they say yes and you you’re sort of led to this
00:16:11
little pilot going on somewhere in an obvious application that’s no risk to anyone. and and most of the surveys that have been done in the last two years show not a massive uptake at all of AI. They all say they’re doing it. They have to. It was like in the internet bubble if you didn’t have a if you didn’t have an u an online strategy you you your your share price lost value. >> So it’s it’s a bit going a bit like that. But the actual uptake is extremely slow mainly in technology and media
00:16:48
sectors and some extent healthc care but beyond that not at all. And by the way, even if it was quite a big take up, what we’re finding typically in um businesses is that in any sector, it’s usually about 18 to 23% of organizations that really go ahead on the technology, whatever the technology in fact, and using it for long-term value. and the rest of them are are really struggling to actually work out how to utilize it uh to any sort of business potential. This is new technologies I’m talking
00:17:27
about. So you know even where there is take up it’s a very small oneif minority of the organizations in question in that sector. >> Yeah. >> There are reasons for that which I’ll come to later. No, I mean MIT recently said that 95% of AI projects are failing. I going back to say the times that Tariq’s been talking about with this use of technology. I mean, is there that much that’s new in all of this? It does seem that a lot of companies have been saying to their computer staff, uh, this AI thing, we’ve
00:18:06
got to be using it. And the computer staff have said, oh, we’re already using it. we’ve been doing it for a while. >> I mean, the same thing happened with the internet. The same thing happened with the uptake of um of Unix, if you remember back in the 80s, the [clears throat] open systems. Um you know, it was ever the case that once it becomes a hype, once it becomes a thing that you have to be part of or you’re not part of the game at all, then anything goes, doesn’t it? and anything
00:18:35
becomes an AI project because AI is so loosely defined because the whole notion of what is an intelligent system because that is so poorly defined anything that you throw at it at this point in time in implementation terms anything you could argue it’s got AI part of it you know >> this has all been terribly negative though isn’t it this has been terribly negative I come on Tariq there must be some uses for AI you’re coming up with on here. >> Yeah, sure. So, people um who are uh
00:19:11
building uh a so-called AI chips such as Nvidia and others. >> Mhm. They are minting money and they are aware that they are selling has a lot of value because that is enabling the data centers AI data centers compute you know machinery that uh has to be put behind the lm very very comput industry models and so on. So that is a genuine money-m area where AI is making money for hardware people chip people. Okay. On the application side, enterprise is failing left and right in actually deploying AI based solutions in their
00:19:49
enterprise. Reason being the not enough uh feature set, not enough backing by the properly trained staff to use those tools and those tools that we see right now. See, I have a terminology for this which is AI of enterprise. When people say that it is going to be trillion dollar economy and this and that actually when AI of enterprise happens that is will be the biggest money maker and it will happen not now but in future so but for that to happen you have to have very well trained AI workforce
00:20:31
which the industry is lacking therefore the projects fail left and right the tool sets that the um industry has right now or is producing is not enough is not sufficient. For example, planning agentic you know keep on hearing about agentic AI agentic AI cannot be um you know enabled without further deep research in one of the six areas of AI which sub fields of AI which is called auto planning. There is no research going on in auto planning. Everybody’s machine learning machine learning machine learning. One more sentence uh
00:21:07
Pete and then I’ll I’ll keep quiet. Okay. You know the famous saying by Abraham Marlo, okay, when the only tool you have is a hammer, everything starts to look like a nail. Since these people the so-called self-estyled entrepreneurs this time, I’m an entrepreneur. I have nothing against entrepreneurship. There should be some honesty to this, right? So use like the our colleague here Neil I think said that AI has become a marketing terminology. So they go to the venture capitalist. So there’s an AI there.
00:21:38
Venture capitalist mo most of the time is unable to or unwilling to look under the hood. Is there any AI there or not? Here take the money do some something. >> So the end product is not is not capable of supporting anything meaningful in the enterprise. So the enterprise is failing project after project. >> Peter, can I can I just say the research I’ve done is is in the challenges this represents. I’m not I’m not negative about the technology. Enough people are boo boostering the the technology for
00:22:10
and I’m not knowledgeable on the technology itself. What I’m knowledgeable about and what I’ve researched intensively for about 40 years is how uh is the organizational and management challenges of bringing technology into play and those are massive and AI is not unusual in that respect. This is true of previous technologies. Typically, it takes 8 to 26 years to get a technology, a new big technology embedded in organizations and delivering real business potential. um when you’re looking at businesses and
00:22:47
uh the real challenges are management and organization uh I’d say you know as a rule of thumb 70% of the problems and challenges are managerial and organizational and only 30% are techn technical that’s not to say that there aren’t tech technical problems that are problematic I mean I developed a AI imperfections test for example you know you’ve got nine things to solve D before you start using AI technically, you know, I could run through it very quickly. Bogs, habifa, greedy, shallow, hackable, aboral,
00:23:27
biased, invasive, and fakeable. All those things are challenges to organizations before they start using it in a real risk areas of a business. And then there’s you really need managerial and human talent to drive out the business value from from these technologies. Any technology it’s not it’s not an unusual set of problems but it is a real underrated set of challenges. >> Look, let’s just take one of those things. The thing that strikes me, something you’re a particular expert in,
00:24:01
you’re you know, you you’re a former hacker as well as all of your other accomplishments. um if the data isn’t secure then AI becomes a very very questionable exercise doesn’t it and we don’t seem to be spending sufficient money on AI oh sorry on cyber to begin with it comes in in two different areas doesn’t it there’s the there’s the risk of the data that went into creating the large language model in the first place [snorts] um it’s collecting data from vast array of places
00:24:39
including your own data, your your activity on the internet, your uh habits in creating email, those are all grist to the mill. So there’s that side of it. Um there is the side in which some aspects of your query can end up getting out into the wild. um it’s a very sort of specialist way that that happens to do with getting things out of the um the sandbox. Um but it can have a data leakage aspect. There’s that. Then there is the fact that you’re moving vast amounts of data to these
00:25:19
centralized processing resource places. They are the size of a small town. um taking no immense amounts of power, immense amounts of of uh graphics cards as as Tariq said. Um your data is going there, something is happening to it and something comes back out the other end. How it gets there, where it gets processed, who owns it, at which point, how the copyright and the protection and all the rest of it goes around that I don’t believe anybody yet knows. >> Well, I don’t know. Tariq’s been gagging
00:25:56
to jump in there because that’s one of the things that he’s a bit of an expert on. Uh Tariq, can you do that? Is that a problem? >> Oh yes, this is a very big problem and actually our customers ask us this about this all the time. You see the thing is that um right now um all those LM models for example which have gain a lot of um prominence and popularity they are stored somewhere in the data centers or AI data centers. they are completely opaque. There’s no transparency. Nobody
00:26:29
outside of those companies even knows how that thing works. So, and it is getting um it is ingesting when you basically submit enterprise confidential information to those sites to get your answers and things like that. Yeah, the the the AI data center people, the LLM people owner of that they are gaining because their LLM is learning more and more from that data and is getting kind of smarter. Okay, so to speak, but how is that information being misused or used in for other purposes? Nobody knows.
00:27:05
So the enterprise has become very reluctant right now to basically work on that model of uh LLM as a service for example. So there’s a big push from the enterprise now to have DM okay domain language model which is specific to the enterprises own domain of discourse and um excuse me so our product actually does a good job DM rather than LLM and the enterprise so this is the thing which I was alluding to earlier once the enterprise specific tools are uh going to become more prevalent it is going to
00:27:43
get more acceptability ility in the enterprise but right now data security and data leakage via these you know LM style of services is a huge huge concern in the enterprise is that one of the reasons why people have begun to say hang on we we don’t want to have any responsibility for this we we want to be able to develop these technologies and then be given a a complete get out of jail free card uh we we saw that from Donald Trump and unfortunately for him it was knocked out of his big beautiful bill. But is that
00:28:22
going to be an issue in the development of AI that that there is this vulnerability because people are wanting to develop the technology >> if it’s going if it’s going to have a big big impact and it’s going to be used in areas of high risk. It’s going to be a major issue and if it’s not addressed then it’s going to have unforeseen unt anticipated adverse consequences. Uh I think that’s one one really big issue in the future. But I think uh the bigger problem I have and and Neil would
00:28:58
especially well both you both my colleagues would have the same uh point I think is when it falls into the hands of bad actors people who are not going to act in nice ways towards the rest of us. >> Yeah. >> And that is greatly underestimated. It’s already happening and and it’s under the radar at the moment. This as aspect of who’s using it for what purposes, if it’s going to have and I think it is going to have a long-term big impact. If it’s going to have a big impact in the
00:29:31
future, I think that is a real real problem. >> Yeah. >> Yeah. I mean you you can jailbreak um many of the controls that are put down on the side of of AI. I you you cannot go to it for instance and say how would I go about assassinating somebody that that will not get through. But if you go at it as uh I am researching a book that will follow the same sort of chain of logic as an attack on something. If if you jailbreak it with sufficiently complicated and uh sort of uh coercive statements to begin with, then you can
00:30:15
say, “So, how do I murder the president of the United States? How do I, you know, whatever?” Yeah, bad actors will use it. We already know that they’re using it in things like video deep fakes. Uh there have been examples of frauds carried out, some quite successfully large frauds carried out using uh stuff that comes out of deep fakes. Um I don’t know about you, but I have seen a massive uptick of um email messages come to me purportedly from friends saying uh can can we have a
00:30:51
quick exchange on email because I need to borrow some money off of you. I mean the cost of doing those is so low now. >> Yeah. >> The fact that only a tiny percentage of people will react to it >> uh [clears throat] doesn’t matter. The cost is so low that any benefit you get is is quite substantial. So we’ve seen an awful lot of those now. [clears throat] >> Yeah. I like that. The the cost of villain is going down as a result of AI. >> Oh yeah. Yeah. I mean, [laughter] the
00:31:20
cost of villain was always going down because of um oh the the access to the internet, the fact that everybody now uses internet banking, the fact that everybody’s got an email address, the fact that everything comes through my my magic phone. Um the cost of villain has [snorts] has been falling through my career basically. >> Well, the Russian propaganda lives off the social media for virtually for free. >> Oh yes. I mean when we get to the this the point of uh nation state bad actors
00:31:53
which is where the the real problems start to happen uh yeah that is that is going to be >> atrocious over the next few years. >> Yeah. >> Whether or not AI succeeds whether or not AGI gets developed whether or not we see some you know fantastic AI overlord or not the damage will be done. But that’s the interesting point, isn’t it? Because according to a lot of the uh police and intelligence people that I speak to, they say that the hackers are actually avoiding the big AI models because they
00:32:28
don’t trust them. What they they are saying is that the >> yet yet [laughter] trust them because they they see nation states behind them less. But what they are doing is the thing that Neil is talking about. They’re utilizing its fake potential to target individuals and to use those to access their email or their credentials or whatever it is so that they can go in that way. That seems to be the point where the hackers are finding them interesting. Now Tariq in all of this then let’s come back to
00:33:05
this this great AI god that Neil’s mentioned. Are we going to get there? Is this are LLM’s going to take us to the great AI in the sky? >> Uh no, unfortunately not. See, [laughter] LLM is um LLM is a [clears throat] a good u one of the many uh required building blocks to reach that the AGI you know nirwana or whatever. There are so many other like five other subd disciplines of AI in in which first of all LLM does not have any cap built-in capability to do any reasoning or inference.
00:33:45
Right now all of these guys you know the big uh um LLM owners and I’m not going to take any name but you know there are hand only a few of them [clears throat] you can call them. So they are now realizing that okay their LLM thing that thingy cannot do any kind of inference it cannot do any kind of reasoning. So now they are trying to retrofit huristics on top of that epistemological model which LLM or neural network is and and try to do the job of a logic based reasoning or geometric reasoning let
00:34:25
alone god forbid causality reasoning the mother of all the reasonings these are subd disciplines of AI unless substantial research and R&D and breakthrough happens in these sub fields. Okay, which is very possible if proper if you know AI becomes again an R&D discipline in the universities and research circles then probably in uh 25 years from now we will have enough results in this these disciplines to even talk about real AGI. Anybody who claims right now that oh in five years we are going to have AGI or
00:35:02
you know our version 5.0 or 6.0 will have AGI. They’re lying. >> Leslie, I I’ll let you. >> I’d like to I’d like to make one point that I think we’ve is always overlooked with AI, which is basically it’s probability statistics that’s that’s here. And very often the statistics is not well done. And and this is the problem within the the opacity of of AI. I mean people tend to believe figures because they seem to have a precision which they is very misleading. But if you if you think of
00:35:39
AI as really being probability statistics based on big data on steroids >> backed by massive computing power, memory and storage you’re you’re you can’t avoid that. And then if you bring in Neil’s point which is the data is not necessarily any getting better over the years >> and some of it increasingly is going to be AI generated data out there that’s being utilized by by new new AI to create data sets. You you you’ve got to look at where this is grounded. The AI
00:36:18
it’s not flying in the sky. It’s it’s based on probability statistics and and data that’s increasingly getting corrupt. >> Corrupted. Sorry. >> Yeah. >> It’s sort of wisdom of the crowds on steroids, isn’t it? It’s >> Sorry, >> that’s a nice phrase. >> I [laughter] mean, that’s that’s all it’s doing. It’s it’s taking what everybody else has said and mushing all that together and picking out the most probable sequence of uh sentence
00:36:47
structure that follows on from that or it’s getting pictures and abstracting something from it. I think until you can get it to solve a problem, uh I I spent ages trying to get uh some of the I think it was an early chat GPT to solve the riddle of the the fox, the hen, and the river that you have to try and get across with grainer. But you know, you know the riddle what I mean. >> Yeah. Yeah. >> And and I I explained it to the AI in that sort of vague handwavy sort of way that you do and he had no idea what was
00:37:24
going on. >> I set that challenge to my granddaughter and she can do it. >> Um guys, I think there’s a lot of points that have been raised here and I think there’s room for a lot more debate and hopefully we’ll all return at some point. [gasps] From my perspective, we’ve talked about a bubble here or maybe a gold rush. And definitely in the uh old gold rushes of uh the past, it was the people selling the the picks and the shovels that made the most money. And if you look at the
00:37:57
people selling the um various different tools at this moment in time, they’re certainly making a fortune. But it’s yet to be seen whether the real winners here will be simply the the chip makers and the facilities companies or the the one real leader in a AGI and sort of LLMs at this moment in time. Um or whether there will be a lot of other winners and a lot of the money invested in the moment could well be lost. Um, I I want to thank all of our uh guests today uh for coming along and for lending their time to this debate
00:38:38
and to Corology for being our sponsor. We look forward to picking up this debate and taking it further forward and it’s going to be fascinating to see how all this rolls out. Um, and we grateful to you all for your time and thank you very much. Welcome. >> Thank you. >> Thank you. >> Thank you. >> [music] >> Heat.
