Google’s Gemini puts it in the game vs ChatGPT

You can also listen to this podcast on iono.fm here.

Listen to the full MoneywebNOW podcast every weekday morning here.

SIMON BROWN: I’m chatting with Viv Govender from Rand Swiss. Viv, I appreciate the time today. Google launched their much-anticipated ChatGPT competitor, Gemini, last week. First off, the launch video, by all accounts – and in fact I think Google has admitted it – was a hallucination, or perhaps more politely, an outright lie.

VIV GOVENDER: Yes. It’s not a lie but at least severely enhanced. So what they did was they shortened it. The way it looked when you watched that video was that this thing could see a video, like the AI could see a video. So they basically had a camera over a piece of paper, and the guy was showing it things, drawing on the paper, that kind of stuff. And it was giving answers which seemed to be in real time. The closest thing I could say to this it was like a Star Trek computer, how they reacted. Exactly. It looked amazing.

The thing that amazed me – I think I spoke to you about this as well, the last time we talked – was the fact that this thing didn’t come out so much better in terms of the testing than GPT-4 because the video looked like a quantum leap better than what GPT-4 was.

But we looked at the little test results. It was beating it by like 91 versus 87, or like a couple of percentage points here and there. It wasn’t saying this was a totally different kind of machine much, much better than a GPT-4. And this wasn’t even a GPT-4 as the current version; it was like the first version of GPT-4 from like nine months ago. So it was weird, just the discrepancy between what we were seeing was and what the scores were. And then, like you said, it did come out that they were basically showing still images and saying it was actually video. They were basically cutting down the time between the responses and the question, and a couple of other things along the way to give us an impression of what it could do beyond the actual capabilities currently.

SIMON BROWN: So Google’s in the game now. They’ve got their Gemini. They of course had Google Bard going, and I’ve got to say I always thought GPT-4 was slightly better than Bard. They’re in the game, but they’ve just in the game. To your point, they’re not rushing ahead, much as when GPT-3.5 came out in November last year, which was an absolute revolution.

VIV GOVENDER: Google’s kind of weird. Google had DeepMind and it had Google Brain, which they’ve combined together at the moment. But the thing is that the technology of all the stuff, the actual underlying basic science on which everything is based is from Google. Google was like the first guys out there but, as you know, with the computer, Xerox invented everything that Apple made famous – you know what I mean. And then Windows made the real money off it in the end. The same thing applies here.

It could be that Google is the IBM or the Xerox of the AI space, where they did the actual underlying technological changes. But the problem that they have is that AI really, really goes up against their primary source of revenue, which is Search. It totally cannibalises their Search business.

And so Blackwoods Microsoft with the phone, like with some other companies out that we’ve seen in the past, when you do have something that you invent that cannibalises your current business, you’re much more reluctant to use it, whereas Microsoft basically doesn’t have a Search business except now Bing. You can see what it was, but it’s not really the core part of Microsoft’s revenue stream; they were willing to experiment and that has maybe given them an advantage.

But the amazing thing here is, like I said, the scores that we’re getting from Gemini aren’t significantly better than from GPT-4, and GPT-4 was trained back in September of 2021. It took about a year for them to be comfortable to release it to the public, but it was pretty much trained two years ago. So it’s kind of disappointing that Google isn’t like a huge quantum leap ahead of what GPT-4 is.

The second thing is that Google DeepMind is doing some incredible things. So this kind of damages their reputation, but they’ve come out of things recently like GNoME, like TensorFlow, like a new one recently – I think something called Elemental [Elements] – I can’t remember the exact name.

But the point is the program stuff that they’ve done basically does the work of a billion-dollar-a-year program that the US government is running on your personal computer and does it faster and better than the billion-dollar program does. The US government has a supercomputer they run that gives you a result once every, I think, five or six hours about the state of the US weather system. DeepMind came out with something that you could run on your home PC, a very high-end home PC, but something a person could buy, an individual could buy, not a billion-dollar thing. And it gives you a result every couple of minutes, not every six hours. And the results were better, marginally better than what the government program was doing.

There’s also a new thing that they’ve come out with, GNoME, which is basically material science where they’ve developed a thing that can predict stable crystal formations. Sounds very silly, sounds very esoteric, but understand that material science is like one of those driving things. You invent a new material. Imagine you get a new kind of steel that’s five times as strong, or a new kind of copper that’s basically more conductive; the effect it has on the world economy is insanely large. And it is releasing these things out of nowhere.

We’ve talked about AlphaFold, which revolutionises medicine, or the proteins. And so they’re doing everything great except for the LLM [large language model] stuff, which unfortunately is their core business.

SIMON BROWN: I take your point, it’s an innovator’s dilemma, isn’t it? They’re scared because they don’t want to lose their Search, because that’s where they make their money. This does suggest to me that the winners into the next year are pretty much going to be the winners from this year. It’s going to be Nvidia – they’ve got their Grace Hopper 200 coming out – it’s going to be Microsoft.

Even, heck, Meta has been putting out some AI products mostly into the open source market.

VIV GOVENDER: Meta is doing a very different thing. If you listen to the guy that runs the Meta system, a guy called Yann LeCun. He’s a French guy, but there are two teams when it comes to AI. They are the Doomers and the Accelerationists. Acceleration means the faster we get to AI, the better it is going to be because it’s going to solve all our problems. Effectively we are going to invent – not to be blasphemous or whatever – a benevolent god that’s going to solve all our problems. No more climate change, no more disease, no more whatever.

The more we delay bringing this AI into being, the more people suffer in advance. The Doomers believe that the moment we get super-intelligent AI there’s like a 99% chance we all die. I don’t mean like things go badly. I mean they honestly believe that there’s a virtual certainty that all of humanity dies – like literally big dinosaurs.

And Yunn LeCun is very, very much on the Accelerationist side. He’s basically the guy in charge of Meta.

And so Meta has a very different policy. He is Meta’s AI team, so Meta basically releases all this stuff to the public. You can go out and download from a torrent – I think it’s like 80-something gigs – but you can download from a torrent the Meta AI system. It’s not as good as GPT-4, by no means, but it’s free and it’s available to the public. And people are using that as the basis for creating their own models, their own chatbots.

That is something that I think is a very interesting plan because we all know Microsoft won, or IBM won the in inverted commas, the PC wars, because they were Open Source. Everybody made IBM stuff. Apple was closed. And therefore, even though in some ways they were better at the start, developers never used them because it cost too much. And therefore Microsoft won; and it’s possible that Facebook wins or Meta wins, because they are the ones that are developing the program that everybody’s using, because it’s a free program, it’s the one that’s available to them.

SIMON BROWN: Yes, so it’s going to get that mass adoption. We leave it there, Viv Govender, Rand Swiss, always appreciate the insights.

Download the free LiSTN audio app on Google Play, Apple or here.