The Google Strikes Back (The AI Wars)

 

Take a look at this. Say you're on a hike and you stop to take a photo in front of a waterfall. You wish you had taken your bag off for the photo. So let's go ahead and remove that back strap. And maybe you want to even get rid of some clouds to make it feel as sunny as you remember it. [Music] Looking even closer, you wish you had posed so it looks like you're really catching the water in your hand. No problem, you can adjust that. [Applause] This was a moment at Google's I/O conference for developers. The way the crowd responded to this moment reminds me a little bit of that time Steve Jobs scrolled on the iPhone for the first time in public.

'Well, how do I scroll through my list of artists? How do I do this? I just take my finger and I scroll.'

Now obviously, Google's keynote wasn't on the same magnitude, but you could sense something, that little spark of delight in seeing something digital that was visually surprising. I think it's likely that as time goes on, AI will continue to surprise us in a similar way to the first iPhone did. After being blindsided by chat GPT and the subsequent AI explosion, Google is hitting back and they showcased just what they were capable of when given a clear direction and goal. From automatic email generating to a conversational search engine and the AI discovery of a molecular syringe, it's all very interesting stuff which shows you just how quickly the world is about to change. But how are the risks going to be balanced by Google? 

'I've been reflecting on the big technology shifts that we've all been a part of. The shift with AI is as big as they come and we are doing this responsibly in a way that underscores the deep commitment we feel to get it right. So on behalf of all of us at Google, thank you and enjoy the rest of I/O.'

It's clear that Google says this is a big moment in their history and in world history. Think about it this way, Google's most valuable product is search, which by the way brings in about 60% of profits. If Google is willing to risk putting AI into their Google search, the center of their multi-billion dollar income stream, AI can no longer be called a fad. It's here to stay. It's not an NFT or a meme coin, it's a new class of legitimate product, a digital assistant for your workflow. Eventually, no matter what job you do, AI assistance will help you do the repetitive, boring stuff. Or at least that's the plan anyway. The implications of this are huge. It's changing vast swaths of industries like the web did, but 10 times as fast.

In May of 2023, Google unveiled their AI game plan to the world. Google's main strategy was to integrate all of their AI research systems into their products. Now, Microsoft Bing had their chance to corner the market, but it seems like they couldn't capitalize on their lead fast enough. Bing AI turned out to be even weaker than chat GPT. It is useful for some stuff, for example, I had an issue with Ableton that would be hard to Google and Bing AI did help me out. But still, for a lot of people, they want even more. For Google, it was the perfect time to strike back and with their large user base, they could now protect their Cash Cow search. Google's AI search Can basically do everything Bing AI can do but with the accuracy of Google. So let's search for 'what for what and a dog Bryce Canyon or arches'. Now although this is the question that you have, you probably wouldn't ask it in this way today. You'd break it down into smaller ones, sift through the information and then piece things together yourself. Now search does the heavy lifting for you. There's an AI-powered snapshot that quickly gives you the lay of the land on a topic. And so here you can see that while both parks are kid-friendly, only Bryce Canyon has more options for your furry friend. Then if you want to dig deeper, there are links included in the snapshot. You can also click to expand your view and you'll see how the information is corroborated so you can check out more details and really explore the richness of the topic. This new experience builds on Google's ranking and safety systems that we've been fine-tuning for decades.

To make shareholders happy, Google spent a lot of time in the presentation focusing on shopping. They wanted to make it clear to advertisers that they're not going to be left behind. Here in summary, this is promising a much smarter Google search experience with the ability to understand complex questions. It's a natural conversation chat experience that includes follow-up questions.

So what about Bard? As many of you remember, the launch of Bard was pretty poor. The presentation was rushed to compete with chat GPT and the initial results of the system were less than stellar. Google left that battle with their tail between their legs but have now returned with vengeance. Google claims that Bard now has better math skills, logic reasoning capabilities, and can now access the internet.

'So I'm thinking about colleges but I'm not sure what I want to focus on. I'm into video games and what kinds of programs might be interesting. Okay, this is a helpful head start. Animation looks pretty interesting. Now I could ask, help me find colleges with animation programs in Pennsylvania. Okay, great, that's a good list of schools. Now to see where these are, I might now say, show these on a map. Here, Bard's going to use Google Maps to visualize where the schools are. [Music] [Applause] Now let's start organizing things a bit. Show these options as a table. Nice, structured and organized. But there's more I want to know. Add a column showing whether they're public or private schools. [Applause] Perfect, this is a great start to build on. And now let's move this to Google Sheets so my family can jump in later to help me with my search. [Applause] Okay, now that's a taste of what's possible when Bard meets some of Google's apps. But that's just the start. Bard will be able to tap into all kinds of services from across the web with extensions from incredible partners like Instacart, Indeed, Khan Academy, and many more. With Adobe Firefly, you'll be able to generate completely new images from your imagination right in Bard.'

So with that being said, what's the difference between Bard and chat GPT? Well, for one, Bard now has live internet access. OpenAI, on the other hand, hasn't released this feature to the public yet. Chat GPT can often cite sources that are incorrect or hallucinated, but with Bard, you can easily fact check within the app. Also, unlike chat GPT, Bard gives you the option to select multiple drafts of its outputs. These drafts are distinct from one another. It allows you to choose the output that most aligns with your preferences. While GPT-4 is capable of Understanding images, this function hasn't been rolled out to chat GPT yet. But Bard, on the other hand, can understand images. So in the next few weeks, Bard will become more. You might upload an image and ask Bard to write a funny caption about these two. A lens detects that this is a photo of a goofy German shepherd and a Golden Retriever, and then Bard uses that to create some funny captions. Really, it's going to be so interesting to see how all of this pans out. As for availability, there's no waitlist anymore and Bard will be available in 180 countries.

Another area that Google focused their AI injection efforts on was Gmail. For many years, Gmail has had sentence suggestions and small elements of AI, but now they're leaning fully into it. They call this new system 'help me write'. It goes like this: a user types in a prompt, hits 'create', and a full email draft appears. The idea is you'll no longer need to search through email chains because the system automatically pulls useful information from old emails and other conversations, and it learns from your writing to improve over time. I can see that the aim of this is to write emails faster so people can focus on doing actual work. But while watching the presentation, I couldn't help but entertain a funny thought. [Music] Gmail is absolutely massive and everyone uses it. In the near future, in essence, we might just have AIs writing to each other for correspondences. Humans will be less involved in business communication. What this exactly means, I don't know, but it is a strange thought.

Google Photos now gets an AI overhaul. You can create a stunning 3D image from any regular photo and then use it as your wallpaper. So let's take a look. And I really like this photo of my daughter, so let me select that. And you'll notice there's a sparkle icon at the top. So if I tap that, I get a new option for cinematic wallpaper. So let me activate that and then wait for it. And check out the parallax effect as I tilt the device. It literally jumps off the screen. [Applause] Now under the hood, we're using an on-device convolutional neural network to estimate depth and then a generative adversarial network for in-painting as the background moves. The result is a beautiful cinematic 3D photo. Users can now reposition elements in their photos and the AI fills in the rest. This means that parts of photos that never existed can be seamlessly recreated.

So there's a question, how is Google doing all of this? Google has been researching AI for a long time and now they're just letting it out of the bag. A lot of what Google announced was based on a more polished version of their in-house GPT competitor, Palm, and they're calling this new system Palm 2. It will generate text in most languages, has the ability to code, and also the ability to reason. Google's Palm 2 comes in four different sizes, all designed for different scale. The smallest is lightweight enough to be put in a mobile device offline. But where Palm 2 really thrives is within the datasets of specialized knowledge. Users can fine-tune the system as their client needs. Two examples Google gave was cybersecurity and medical research.

What does this mean? Well, let's take medicine as an example. Say you're a medical service company. Hospitals may benefit from having an AI helper that's specially trained in medicine. Instead of training the model from scratch, all you'll need to do is buy the base model from Google and train it on particular data in your field. This Could be research documents, studies, labeled x-rays, MRIs, and ultrasound data, etc. The AI helper can now look through the MRI and guess with 95% accuracy a rare cancer or provide cognitive aid to physicians. That kind of thing. It does sound crazy, but it's already on the way. Using Palm 2 as a basis, Google trained it on medical data and produced Palm Med2. The result was a nine-fold decrease in wrong answers when compared to the standard par model. This is currently the state of the art. I discussed this whole concept of fine-tuning on top of a base model in my first video when chat GPT first hit the public. Google claims that its optimized AI hardware servers powering all of this are 80% faster and 50% cheaper than other alternatives.

It's probably worth saying at this point that a lot of this stuff may not be anywhere near perfect at launch, and there's probably going to be a few months of disappointment. But give it six or so months, and it's going to exceed even the expectations of the critics. Google gave some updates on their AlphaFold biological project. They stated that 400 million years of progress was made in the space of weeks.

'I feel defining researchers helping scientists make bold advances in many scientific fields, including medical breakthroughs. Take for example, Google DeepMind's AlphaFold, which can accurately predict the 3D shapes of 200 million proteins. That's nearly all the catalog proteins known to science. AlphaFold gave us the equivalent of nearly 400 million years of progress in just weeks. [Applause] An interesting outcome was a molecular cancer syringe. 1 million researchers around the world have used AlphaFold's predictions, including Fan Shang's pioneering lab at the Broad Institute of MIT and Harvard. In fact, in March this year, Zhang and his colleagues at MIT announced that they'd used AlphaFold to develop a novel molecular syringe, which could deliver drugs to help improve the effectiveness of treatments for diseases like cancer.'

There's a lot of other stuff mentioned, such as Duet AI for Workspace. It can help you draft documents, slides, spreadsheets, and emails, and can also generate images from scratch. Vertex AI, an AI platform for building generative applications. Users can select and then fine-tune the AI models, and they have access to foundation models such as chat, text, and image. And lastly, Google did unveil their plans for their defense against the flood of AI-generated content. The company is working on AI detection and harm reduction tools. This includes the evaluation of images with extra context.

'As we invest ments models, having the tools to identify synthetically generated content whenever you encounter it. Two important approaches are watermarking and metadata. We are building our models to include watermarking and other techniques from the store. As we begin to roll out the generative image capabilities, like Sundar mentioned, we will ensure that every one of our AI-generated images has metadata, a markup in the original file to give you context if you come across it outside of our platforms. Not only that, creators and publishers will be able to add similar metadata, so you'll be able to see a label in images in Google search marking them as AI-generated.'

So, Google has had over a decade of AI research and have even called themselves an 'AI first' company for the past seven years. So they've had the research this whole time, but they've rightly been very cautious. Microsoft and OpenAI have forced them to fine-tune their technology and get it out fast. This is both good and bad. It's good because we all get more AI tools to increase our productivity. ButHere is the final part of the transcript:

"It's bad because the slow and cautious approach has been abandoned in the face of increased competition. That being said, the safety measures being put in place are a good step. I personally think it's fascinating to watch this battle play out in real time for a fundamentally revolutionary technology. It's going to take a few years to see what really happened and how the world settles, but I'll always be watching with intrigue.