The Slow Death of Google's AI

One of the questions many people have been asking is what on earth is going on @Google AI? And this video is going to give you a comprehensive guide on exactly what's going on at Google AI. So let's not waste any time. One of the recent things that kind of went viral, and this isn't just the only thing that went viral this year for Google's AI blunders is this right here. Now, I know that many people have.

Actually spoken about this, but the video.

Isn'T going to be just about this. It's actually going to be mainly about what's going on behind the scenes.

The.

Because there's a lot that is going on behind the scenes that most people don't really know. And that's why Google might lose the AI race if they aren't careful. So essentially you can see right here. Recently, Google's search generative experience suggested a few things that I guess you could say were out of the ordinary and in some cases completely outright dangerous. Suggesting that you could add glue to a pizza base to make it more tacky and sticky and is just completely false. And if they rolled this out worldwide or, you know, to just us users, I'm guessing that since millions and millions.

Of people use Google search every day.

This is a really, really negative thing that whilst, yes, it is based in the fact that generative AI does have some hallucinations and some shortcomings, it's something that they should have figured out how to do before publishing this. Now, the problem is, okay, is that this is a symptom of a larger issue that is going on with Google. And this is only the symptom of the entire situation which I'm about to dive into now. And I think it's important that we.

Do talk about this because Google, out.

Of any of the other large conglomerates and large companies, has the biggest opportunity in terms of being a company that.

Does really, really well with the AI stuff. So the problem with Google, and one.

Of the main, main problems, okay, and this is a bit, I guess you could say hypocritical, but.

But one of the main problems with Google is that they have a failure to ship. Okay?

And essentially what that means is that if you truly, truly, truly look back on what Google have done is they've just literally sat on everything that they've had. And you can see right here that this is an image from a tweet where they describe the Google I O conference. The Google I O conference was a recent conference where they announced, and I say announced, not released because they announced a bunch of different AI tools and a bunch of different AI things that.

Are going to be coming. The only problem is that Google literally has the biggest opportunity to win. They've got the user base, they've got millions and millions of dollars. And the problem is, is that they don't ship anything. Now, like I said, it's hypocritical because in the past slide, you saw that when Google shipped something, it wasn't exactly perfect. But that's the problem, okay? Whatever Google's ecosystem is currently inside the company. It's not working because they don't ship things fast enough. But when they do ship them, they're a huge giant mess. Now, this image here, I've added this because I put failure to ship because one of the craziest things about the whole AI race, the thing that started this AI thing where Microsoft saying they're going to make Google dance, was the fact that Google sat on their research paper, the transformer, and OpenAI used it to create ChatGPT. And that's the thing that people now know and they now use. And the thing is, Google have lost the ability to innovate and use the technologies that they've previously developed to actually make them into meaningful products because they're just a giant company. And it's pretty crazy the fact that this also happened with Sora. There's actually some research that indicates that the people who created Sora actually built it off of the backs of some of the, you know, research papers that were surrounding Google's video models, which is pretty crazy when you really do think about it. So one of the things that I talk about is the fact that Google is just too big. The company's employee count stands at 182,000 people, okay? According to the latest data from quarter four of 2023 guys, 180,000 people is absolutely, absolutely insane. Whilst, yes, this doesn't account for the fact that likely all of these people are probably not working on AI, because if we do look at deepermind, there are only 1500 people working at AI. The thing is, is that when you have such a big giant company, the problem is, is that it's like this, okay? Every company is like a triangle, okay? And it's a hierarchy, okay? It's a pyramid, whatever you want to call it. If it's like that, you know, you've got. That was poorly drawn. But if you've got like a pyramid like that and you've got, you know, hierarchies of the lowest level employees and Then you've got the CEO. Okay? The problem with that is that you have to run it up the chain. Okay, there's levels, okay? Every single level. Google, Gemini, Google, DeepMind, whatever you want to call the team, they don't act independently. They have to go up, you know, levels in order to get it to Sundar Pichai, the CEO. The problem is okay with they're competing against OpenAI, which is more closely knit. And I guarantee with OpenAI it's probably just maybe four levels that it has to go up with. But with Google, the problem is that things just move very, very slowly. And whilst, yes, they do state that they are on a code red for AI and that they're going to be prioritizing this stuff, companies like OpenAI that have a much smaller overall headcount and can focus on one thing which is just providing the users with a simple product, the market dynamics have shifted. You see, OpenAI is this nimble smaller robot that can move around and can quickly strike and efficiently get into markets. But, but Google, okay, on the other hand is this giant. It's this big thing that is too slow to move, too slow to react, and by the time it does, it publishes something that is very, very poor. So for the cycles that Google is in, it's in a precarious position because on one hand it's behind, it wants to release, release, release. But of course, if it doesn't release and it doesn't verify that their systems are good, every time they release, release, release, they're going to get bad pr, because what happens is you release something that's unfinished and that's not good. Especially with generative AI, the PR takes a massive, massive hit. And that is of course, Google's problems. They need to give DeepMind or Google, the Google Gemini team, the ability to get as much resources as they need. They need to make a lot more hires and they need to make sure that whatever products they are doing and whatever research is being done, they actually do ship it. Now, the thing that I constantly talk about and something that is completely not overblown, but something that a lot of people don't even know, is the fact that Google has failed to ship some of their crazy, crazy products. You can see on labs.google.com this is a website most people probably haven't even seen. And the problem is, is that this website is insane in terms of what you can do with their AI capabilities. You've got image effects, okay, which is basically like mid journey, but the controllability and the user interface is better. You've got music effects, which is your ability to generate lifelike music. Okay. Which is, you know, was, was, was released pretty early, like much earlier than you know, yu do and that stuff. You've got Notebook lm which is something that from the results, I haven't used it just yet but from what I've seen and from what I've heard, people have said that this is absolutely incredible. They use it literally every day. And that's anyone that I've seen. And of course we've got video effects and of course we've got Google Workspace. And the problem is, and of course, you know, we've got Imogen as well. I nearly forgot Imogen, which is their, you know, text to image model. The problem is, is that this is just available on their labs Google website and it's not even available worldwide. You have to sign up for a wait list and then you get through to their like alpha testing system where if you're lucky you'll be able to use this. And the thing is, okay, Google is most likely, here's what's going to happen, okay. They're most likely going to wait for OpenAI to release something like Gemini for like Workspace or like you know, a video model which is Sora, which is coming or, or like their image model which still isn't available in the UK and of course their music model which OpenAI probably will release at some time. And of course this notebook thing here which allows you to organize your notes in a really amazing way. You could really release these products. But for whatever reason it is, Google just has this culture of failing to ship great AI products. I'm not sure what it is and I think they are changing a little bit this year, but it is really, really a big problem when you are in the race for customers and, and you are in the race for an industry that is very, very, very quick in terms of its iteration cycles. Now another crazy thing about Google, which is why they're losing, is that they probably aren't utilizing their compute. Now you can see here that even when giving them every benefit of the doubt, Google's compute capabilities make everyone else's look silly. Google will quite literally have more TPUV5s than OpenAI, Meta Quest, Coreweave, Oracle and Amazon will have GPUs combined. They are able to rent a large chunk of this capacity to various startups. So Google's TPUV5s, their latest advanced chips, have a much higher capacity than OpenAI's which allows them to, I Guess you could kind of say experiment and do whatever they really want. But the problem is that they aren't taking advantage of this advanced level of compute. I mean, when we really ask ourselves what Google has done with the amount of compute capabilities that they have, I mean, it's truly a wonder on what kind of things they're truly working on and what they're actually even releasing. Because my, oh my, it seems that OpenAI continues to make Google dance in their collaboration with Microsoft. And it's quite unfortunate because this is something where they truly do have the ability. And it seems that you know, in the future that they were set up, like if you look back in time, if you thought there was a country slash, I mean, company that was really going to achieve AGI, everyone would have probably put their money on Google. Especially when you take a look at their compute trajectory. Now, like I've already said in the past, we have drama with Google. There was also this woke drama which was a terrible, terrible PR disaster. And as some have pointed out, this was likely because of Google's failure to check and validate what results were going on when they were doing their extensive testing. I'm sure if they actually used human feedback in the loop here when they were actually testing, then they wouldn't have had this woke drama where they inaccurately misrepresented historical figures or kind of events. Whilst yes, it is important for AI systems to be inclusive for everyone, I think it also is important for it to be accurate as well. And that is where Google may have slipped up. And the problem is, is that these issues don't get fixed as quickly as they are discovered. The, the problem is, is that sometimes I'm seeing these issues, these issues, whether it be woke AI or the, or the AI saying, you know, wrong things regarding certain topics. They do for some reason for Google, for such a big company, it takes them a very, very long time for them to fix their issues. When I've seen issues on OpenAI's chat GPT, like within 24 hours, they've usually fixed that issue and you're not able to replicate it with AI anymore. And with this, it seems that like this was something that persisted for over a week and that was really, really damaging to Google's core AI reputation. Now with this, some of you may not have even known what is going on here. So basically Google did release Gemini and with Gemini, you can see. Just read this. Okay, so it says here we've been testing the capabilities of Gemini, our new multimodal AI model. We've Been capturing footage to test it on a wide range of challenges, showing it a series of images and asking it to reason about what it sees. Now, most of you guys remember this demo, but if you don't, I'm going to give you guys a quick breather. I'm only going to play, like, 30 seconds of this, so you guys know what I'm talking about, because there have been a lot of AI demos recently. All right, testing Gemini. Here we go. Tell me what you see.

I see you placing a piece of paper on the table. I see a squiggly line. What about now, the contour lines are smooth and flowing with no sharp angles or jagged edges. It looks like a bird to me.

Hmm. What if I add this?

The bird is swimming in the water. It has a long neck and beak. It is a duck.

Yes.

A duck is a type of waterfowl in the family Anatidae. Ducks are closely related to geese and swans. The duck appears to be blue. This is not a common color for ducks. However, there are some breeds of blue ducks. These are typically not as common as other duck breeds. It is more common for ducks to be brown, black.

You might be wondering, why on earth did I show you guys this? Well, the problem is with this demo here. The demo wasn't actually real. This demo is, I guess you could say, edited. The voice that you can hear is just artificially generated. And the fact that Gemini is responding with such quickness is edited to basically make it look as if it's real time. Now, what's crazy is that now we do have the technology to actually do this. But many people were completely upset by this demo because they saw Gemini, and, yes, it showed its impressive multimodal capabilities, but they essentially were just felt like they had been deceived because, unfortunately, this demo, you weren't able to replicate this when you actually use Gemini. And I guess you could say, and the fine print, unlike the blog post, they spoke about how, you know, the video isn't real. And they said, we created the demo by capturing footage in order to test Gemini's capabilities on a wide range of challenges. Then we use it. Then we used still image frames from the footage, prompting via text. Okay, so it might kind of do the things it did in the video. It didn't and maybe couldn't do them live in the way that they implied. In actuality, it was a series of carefully tuned text prompts with still images clearly selected and shortened to misrepresent what the interaction is actually like. And you could see some of the actual Prompts and responses in a related blog post. So this is clearly an issue on Google's part and this is the problem that's going on behind Google's. Okay, so this is basically the innovator's dilemma, okay? So the summary is basically that successful companies fail by sticking to their established business models and they overlook disruptive innovations. And the thing is like this, okay? When you have such a big company like Google, you have a steady stream of income. You have a steady stream of revenue, so you don't want to focus your energy on things that might not return an income. The problem is, okay, is that if your company's big and you don't spend money on innovative things, okay, that could bring outsized returns that could give you, like income graphs like this. When a company, okay, a smaller company that's going to be focusing on the innovative stuff and all they do is focus on the innovation, that is how these companies eventually lose and go bankrupt. Because these small companies that can hyperfocus on innovations, okay, like OpenAI did with ChatGPT, they can then surpass and take market share from existing market leaders who just focus on serving their existing customers. And that's the problem that Google has faced. And it seems that they haven't adjusted correctly. So you can see these smaller companies are just unburdened and they can just focus completely on the disruptive stuff and they can take advantage of the new philosophy, the new innovations. And basically Google is stuck between a rock and a hard place. So they're stuck between trying to, I guess you could say, fulfill the demand that they already currently have with their giant business versus innovative, continually innovate and merge into a new market. You can see here that Google's management reportedly issued a code red amid the rising popularity of the ChatGPT AI. And this was in December, the 21st of 2022. And the problem is, is that it's not like you can just go ahead and train a model very, very quickly. These things take a lot of time and a lot of money. And the problem is, is that like, by the time GPT5 comes out, if Google is already training Gemini 2, which apparently they are, the problem is, is that they might even be behind again because OpenAI started the first cycle and when they released GPT4, all these companies are trying to catch up to, I guess you could say, GPT4. And the problem is that their training runs are just a cycle behind. So the problem is that there's like literally a time delay and Google are trying to, I guess you could say delay that time. They're trying to cut down that time. But in doing so, you overlook very important processes that make the AI good enough. And then when you release it early, the problem is, is that you get something that is just a PR nightmare. Now, another thing that we do have about Google, and this is why I talked about they have, unfortunately, a culture that just won't ship anything, is that people from the company have been leaving, okay, They've been leaving to start new things. You can see Google DeepMind veteran departs to launch AI agent startups. So people that are at the company, Google DeepMind, they've been leaving and apparently they've been leaving to start different AI startups because they want to, I guess you could say, pursue different things if Google isn't going to do them. And the problem is that this talent is very, very rare. Like, you know, good AI, you know, agent developers, or Google DeepMinders, whatever you want to call them, okay? Good AI researchers are hard to come by. And the problem is, is that not only is Google's failure to ship making people think, okay, I'm going to leave and start my own thing because Google's just not going to do this. OpenAI are taking advantage. And you can see OpenAI's new weapon in talent war with Google is $10 million pay packages for researchers. So OpenAI are basically saying, okay, they're paying 5 million to $10 million, okay, for, to basically buy people from Google because they realize that Google has some incredible talent and they realize that you look, AI researchers are like gold right now and we need them to, you know, make sure our products are the best. And they're basically offering them $10 million, okay? And if Google isn't, you know, matching the offers, which Google does not appear to be doing, people are going to be leaving and going straight to open AI, which is pretty, pretty crazy. Okay? So OpenAI's hidden weapon is ex Google engineers. And it seem, it would seem like Google aren't really paying attention. Now what's crazy as well is that OpenAI are striking up some of the biggest deals with some of the major companies that are going to be very, very integral to AI development. You can see that Microsoft and OpenAI are hiring and acquiring talent like crazy. Microsoft hires inflection founders to run new consumer AI division. Microsoft hired Mustafa Suleiman, who of the former Inflection AI. It's pretty crazy that OpenAI struck up a deal with Microsoft because Satya Nadella is playing 4D chess, as many have stated, because They've been acquiring some of these smaller companies and making sure that they are positioned well to take over when AGI is here. Now what's crazy, okay, and what's even crazier is that Apple has been striking up deals with open AI. You can see OpenAI have a new deal with Apple recently disclosed and you can imagine the amount of phones, okay, that exist in the world and in the western hemisphere, okay? IPhones are pretty dominant, which means that Google lost out hugely here because losing out on Apple being, you know, natively running, you know, your AI system is huge, okay? That is a huge, huge distribution network that you lost out on. Which means that whilst yes, Androids are good, I mean it's going to be pretty, pretty hard to get people to try and natively use to use Gemini because are so used to using chat GPT already and people just want ease of use. So it's going to be pretty, pretty crazy. Now I think Google does have one shot and I think their last shot is probably okay, going to be agents.

I'm also really excited about the next stage of these large general models. You know, I think the next things we're going to see perhaps this year maps next year is more agent like behavior. So systems that are able not only to just, you know, answer questions for you, but actually plan and act in the world and solve goals, you know, and I think those are the things that will make these systems sort of the next level of usefulness in terms of being a useful everyday assistant. And again that's bringing back in some of our work we did years ago with gaming which were all agent systems, you know, achieving certain goals and objectives, bringing that kind of work and marrying that together with the modern large multimodal model work.

What you can also see here is Google's recent AI agent demo where they're showcasing that this is going to be their main focus for AI.

Next steps for you, searching your inbox for the receipt, locating the order number from your email, filling out a return form and even scheduling a pickup that's much easier, right?

So I mean the question now is what is Google's strategy? Because there have been many different things that you know, have been discussed on how Google needs to up its AI strategy, it needs to move faster, it needs to launch more products. But I think Google are in a pretty precarious position. Let me know what you think about this because as things are looking and as things go on, it doesn't look great for Google despite them being such a great company. But I think if anyone can pull it off, it is Sundar Pichai. I think he is currently being underestimated at the moment. And I think next year we truly will see who comes out on top of this AI race.