The Amount of AI News This Week Was Insane!

So this week was one of those weeks in AI where it was just announcement after announcement after announcement, and we've got a lot to cover in this one. Google had their annual IO event where they make all of their big announcements about all the cool stuff that they're rolling out. And here's a super quick recap from the Verge.

AI and AI AI AI generative AI generative AI genernative AI AI is AI AI AI AI AI AI AI, AI AI AI. It uses AI to bring AI AI.

AI AI, and that pretty much sums up what they talked about at the event. But besides the Google event, there was still a ton of other AI announcements and I'm going to break them all down for you, including the stuff that I found the most interesting from the Google event. So technically this one rolled out last week on Friday, but because I record these videos on Thursday, it didn't make last week's video. But on Friday of last week, OpenAI rolled out their new text to 3D model called Shape E. And as of right now, there's a hugging face space available where you can play around with it yourself. I'll make sure the link is below in the description, but you can come to the Hugging face page, type in a prompt, and it'll make a 3D image of what? Whatever you prompt it with. So I can type a rainbow wolf and I get something that kind of sort of looks like a rainbow wolf. I can click this download button and it downloads it as a GLB file, which I believe I can open in something like Blender. Some other examples, a spaceship, a green boot, and a chair that looks like an avocado. They also have image to 3D where you can upload an image and it'll try to make it into a 3D image. Their example here is this little Corgi image. I tried uploading my own images and none of them looked anywhere near like the image I uploaded. So I don't know how to get this to work, but you can play around with it. It's over on Hugging Face right now. It's not the most amazing 3D assets I've ever seen, but it's the best looking text to 3D we've seen so far. All right, now moving into this week, we started off the week with Warren Buffett comparing AI to the creation of the atom bomb and Geoffrey Hinton saying that the threat of AI is more urgent to humanity than climate change. But I don't want to get all doom and gloom in this video. I want to talk about some of the really cool and fun advancements that happen in AI. So let's just move right along. Also on Monday we got this demo from Loveless Studio where they show off their tool to create text to Metaverse. Essentially we could see that this was an alien snowy world that they created. Great Lost city of Atlantis. This is what it created. Got procedural terrain generation, dynamic landscapes, dynamic geologies, dynamic lighting, dynamic post processing, all sorts of cool features that takes us one step closer to being able to just build video games using prompting. And this all works inside of Unreal Engine. We don't know when we're actually going to have access to this. This was just kind of a teaser video of what's to come. But this looks like a really, really fun tool that is going to further democratize game creation. Also on Monday we got the announcement that Meta was open sourcing their Image Bind research which has image to audio, audio to image, text to to image and audio, audio and image to image, audio to generated image. So for example, if we select an image, it will try to make an audio out of one of these images. So if I click on this dog, you can see it generated two audios from this dog. So that's AI generated audio from the dog. If I click on this tiger here, this is what it sounds like.

Their audio to image is the other way around. So if you give it a sound of birds singing.

It generates these images off of that audio. Or a train running.

Based on that audio. We get these images, text, image and audio. If you give it the text drums, you get this image and this audio.

Or this image with this audio.

And then we have audio an image to a new image. So we have a dog barking with a picture of a beat. So here's the dog barking.

If I combine the two, we've got a picture of a dog on a beach, we've got the sound of pouring.

With this picture of apples and it generated this picture of apples under a sink. And finally audio to generated image. This engine sound.

Gives us this picture of a boat or something, dog barking.

We get this picture of a dog and this is what image buying does. And you can read more about this over on Meta's blog at this URL which will be in the description. And like I mentioned, it is open sourced so you can actually find it over on Facebook Research's GitHub page and use this code in your own projects if you want. This week TED also uploaded the TED Talk from Imran Chaudhry here where he showed off some of his Wearable AI, augmented reality tech. And it's pretty dang cool.

And we agree.

Sorry, this is my wife. I'm gonna have to get this. Hello. Hey, Bethany. How's it going?

Good.

Are you a ted?

Yeah, I'm on the Red Circle right now, actually. Let me show you something. Invisible devices should feel so natural to use that you almost forget about their existence.

You'll note that's me and my voice speaking fluent French, using an AI speech model that's part of my own AI. This is not a deep fake. In fact, it's deeply profound. This is my AI giving me the ability to speak any language. Imagine this. You've been in meetings all day and you just want a summary of what you've missed. Catch me up.

Patrick is coming to tomorrow's design meeting. Bethany wants to move next week's dinner, and Oliver is asking about soccer this weekend.

These are emails, calendar, invites and messages, all surfaced up to the top. Picked up one of these chocolates. Used to eat a ton of these when I was a kid. Can I eat this?

A Milky Way bar contains cocoa butter. Given your intolerance, you may want to avoid it.

So I can't eat these anymore. I'm gonna eat it anyway.

Enjoy it.

So this is a really, really awesome TED Talk. One of my favorite TED talks in a long time. It's less than 14 minutes and I highly recommend everybody check it out. I'll make sure the link's in the description, but what I can say about this is that AI and wearable technology is gonna get really, really interesting. And this TED Talk with actual working technology shows sort of where things might be heading. This week, IBM also announced that they're going into the AI space with their Watson X AI. They actually partnered up with Hugging Face so you can experiment with the open source models that are available on Hugging Face. This appears to be their own version of like a GPT4 or Palm2, like what Google has. And they're expecting availability for Watson X in July. So as more access for this one rolls out, we'll probably be talking about this one a little bit more in the future. The interesting thing about something like this Watson X is it's starting to feel a little late in the game. If everybody's already starting to build out apps on top of OpenAI or starting to build out apps on top of Google's new Palm 2 by July, it almost feels like most businesses will already have sort of picked their platform of choice to build on top of. It'll just be interesting to see how it plays out and what IBM can bring to the table. That's actually unique. It was also announced this week that Wendy's is going to test AI chatbots that take your drive thru order. And it appears to be built on top of Google's Palm 2 technology. When McDonald's tested it out, it. It didn't go so well.

Can I please have vanilla ice cream? Can I get you anything else? No Caramel? Would you like anything else? No. Cream packet? No, I just, I just want that. I just want a large water and a cup of ice cream, not a ketchup pack. Oh my God. I literally can't. Where did the butter come from? I would just like a large water and a cup of ice cream. There's no saving this.

We need to just go and tell.

Them, McDonald's, I'm done for the Sunday.

Or hot.

But none. Plain zero, zada, nothing. Just vanilla ice cream for the Sunday. Oh my God.

I'm done.

I'm done.

We're done.

Jesus.

So yeah, hopefully Wendy's take on the automated ordering bot fares a little bit better than McDonald's. Airtable also announced that they're adding AI functionality into their platform, which kind of felt inevitable. I'm pretty sure everybody saw this coming. Anyway, it looks to do the same kinds of things that you might get out of something. Like Notions AI. Airtable AI will be an optional paid feature with details on pricing and availability coming soon. Also this week Hugging Face announced Transformer agents. You can control 100,000 plus hugging face models by talking to transformers and diffusers. And it's going to be fully multimodal, meaning you can use text, images, video, audio document, etc. In their Twitter thread they say you can create an agent using large language models like Open Assistant, Starcoder and OpenAI and start talking to transformers and diffusers. It responds to complex queries and offers a chat mode. Create images using your words. Have the agent read the summary of websites out loud. Read through a PDF comes with built in tools like Speech to Text and Text to Speech. Text classification, summarization, translation, image generation, transformations, captioning, segmentation, upscaling text to video. Now this is the part that I found really, really interesting here. So designing a tool and pushing it to the hub can be done in a few lines of code. The toolkit of the agent serves as a base. Extend it with your tools or with other community contributed tools. So what this means to me is that you can use all of the tool sets available inside of Hugging Face to build a brand new tool. Once you build that new tool that does whatever you make it do, you can then add that tool to Hugging Face's set of tools. And now that tool becomes available for others to use and iterate off of, becomes a new tool in the tool set that developers can then use to build off of. So the more these transformer agents get used to build tools, the more tools people have available to build new tools. So with each new tool built inside of this, it just improves and improves and improves, giving people the ability to continually make better tools. In the last few weeks, we also talked a lot about AI generated music, and a lot of that music was ending up on Spotify. Well, apparently Spotify has flushed all of that music off the platform with Spotify saying artificial streaming is a long standing industry wide issue that Spotify is working to stamp out across our service. So although Spotify is now allowing AI generated playlists, Spotify seems to be very anti AI generated music. When we identify or are alerted to potential cases of stream manipulation, we mitigate their impacts by taking action that may include the removal of streaming numbers and the withholdings of royalties. This allows us to protect royalty payouts for honest, hard working artists. Now, anybody that's actually generated AI music will probably be slightly offended by this because it's not as easy to create as most people think. It actually does require some talent and some skill and quite a bit of time to put into it to create something really good still. But Spotify, at least for the moment, seems to be drawing a line in the sand against people making AI generated music, at least with the voices of existing artists. It's gonna be interesting to see how this plays out because this feels like a very gray area to me. Like, what if I had a voice that sounded exactly like Kanye's and I sang in a way that sounded like Kanye, but I'm not Kanye, but I did actually sing it. Does that mean that my song would get booted off of Spotify because it sounds too much like his voice? Just because a voice sounds like Kanye but wasn't actually generated by Kanye, does that make it Kanye's voice? I don't know. It was definitely trained on his voice's data, so there is that element to it. But in the future, I don't think we're gonna be able to differentiate what was generated on somebody else's data versus somebody that just sounds like somebody else. And so it's gonna get real slippery. And I'm not really making a lot of AI music myself, so I don't really have a dog in this fight. I just think it's gonna be interesting to see how this one plays out. Speaking of AI generated music, with all of Google's big announcements at Google I O this year, this is one that flew under the radar because it didn't get announced at their event, but they just opened up their music LM model where you can generate AI music. Now, it doesn't generate vocals, but if I go to the AI test kitchen, I can actually play with this right now and generate some music with it. You can see it gives us some examples. Songs that would play in a poolside party with a light summer vibe. I could see what this sounds like.

And this is totally AI generated.

Hey, pop punk anthem from the early 2000s.

I can actually download the song right here. But I'm very unclear at the moment whether or not you can actually use this music commercially or not. My guess is that you probably still can't, but I'm not sure. It doesn't specifically say anywhere that I can find. Even in their terms of service, there's nothing about how we can use this or if we can use it commercially. All right, so now we're getting into the meat of what happened this week, and that is Google's IE event where they just had announcement after announcement after announcement about all of the AI stuff they're doing now. I did make a sort of TLDR version of the entire Google IO event. The entire event was almost four hours long and I condensed all the stuff that I found interesting down to about 15 minutes. You can find that over at my Twitter. I'll link to this exact Twitter post in the description below. If you want to watch the 15 minute recap that I created, let's go over some of the quick highlights here. So they started by showing an example of how it can help you in email, where somebody got an email about a flight that was canceled. They gave it a prompt to ask for a full refund for this canceled flight. The AI wrote a short email, they asked the AI to elaborate, and then it wrote a more detailed email asking for a refund for the flight. Now, while I think it's really cool that this is going to be built directly into email, it's nothing really new, right? We've been able to do this with ChatGPT. There's probably a hundred Chrome extensions that do this for us right now. It doesn't feel like anything super novel or exciting yet. It just kind of saves us one step of having to use a third party tool. And instead we are going to have it just built straight into Gmail. They also announced that Google Photos is going to have what they call Magic Editor. They rebranded their Magic Eraser to Magic Editor. It can do stuff like erase unwanted items in an image, adjust the brightness, make it look like a more clear day, and then even move things around in the image to get it closer to what you were originally looking to create in the first place. While I think it's cool that it's going to be built directly into Google Photos, this isn't super novel technology. This is the type of stuff we've been getting from tools like Clip, drop and Runway ML already now. They also announced Palm 2, which is their sort of next generation large language model that tools like Bard are being built on top of. It's much better at coding and can even do things like add comments in Korean. They also talked about fine tuning Palm 2 and fine tuning it on medical knowledge in their Med Palm 2, which was quite impressive in the fact that it was the first large language model to reach expert performance when it comes to clinical knowledge. I think it's operating behind the scenes on Bard. So if you have access to Bard, you're kind of seeing Palm2's capability. They also talked about their next gen model called Gemini, which will be coming out later, which is trained on even more. I mean, I guess this is kind of like their GPT training, I suppose. I don't know how to compare it. But this is their next even larger large language model that they're in production on right now. It's still being trained. They talked about how Bard is much better at coding now, trained on all of these programming languages that you see on the screen. They also announced Bard Tools which is really interesting because if you're Familiar with what ChatGPT is doing with ChatGPT plugins, it sounds like that's what they're doing with Bard. Bard will directly connect with tools like TripAdvisor and Wolfram and Redfin and Instacart and Spotify and YouTube and Khan Academy. And a lot of the tools that we've seen are Getting plugins on ChatGPT. This is really cool and I think this is a great feature, but it also feels a little bit like they saw ChatGPT do it and now they're doing it too. Now Google may have totally had this idea before ChatGPT. I can't say for sure, but just the timing of events. It kind of feels like they got the idea from ChatGPT and went we need to add that to Bard2 if we want to compete. Now, one of the other big announcements is that they are partnering with Adobe Firefly inside of Bard. And if you've ever used Bing Chat and had it generate images with their next gen version of Dall? E, then it sounds like Bard is kind of doing the same thing just with Adobe Firefly. So you can go inside of Bard, for example, and say, generate an image of a unicorn in front of a birthday cake for me, and using Adobe Firefly, it will generate that image for you. Again, it kind of feels a little bit like being one step behind what Microsoft and OpenAI are doing with their current version of Bing Chat. They also announced at their event that Bard is now open to everybody in 180 different countries. It's available in English, Japanese and Korean. And on the screen right now you can see all of the countries that it's available in. I'm going to go ahead and scroll slowly so that you can, you know, take a screenshot or pause this video and see where you can get access. But I'm also going to link up to this page in the description so you can check it out for yourself. Now, there are some noticeable omissions from the list. One thing I noticed is Canada still doesn't have access up here, which seemed very surprising to me. And also I've noticed that a lot of people in places like Sweden and Switzerland don't have access either, which is a little bit surprising. And the only reason I pinpoint those countries in particular is because I have friends in those countries who actually reached out and said I still don't have access to it. I thought they were unrolling it to all of the countries. Well, they unrolled it to 180 countries and Canada, Switzerland and Sweden aren't three of them. They also announced Google Duet AI, which is essentially them putting AI in all of their Google Workspace suite of tools. So you'll have AI directly inside of Google Docs, directly inside of Google Sheets, directly inside of Google Slides. And they all kind of tie together with each other. So if you need information on your slides from one of your Google Sheets, it can reference and access what's in your sheets for your Google Slides. If you need access to something that's in sheets in one of your documents, they all sort of cross talk to each other and that's called Google Duet. Here's one of the examples of them using it that I found interesting.

Something like client and pet roster for a dog walking business with rates and hit create Sheet Sends this input to a fine tuned model that we've been training with all sorts of sheet specific use cases. The model figured out what you might need. The generated table has things like the dog's name, client info, notes, et cetera. This is a good start for you to tinker with.

They also gave another example of it being used in slides where they generated a whole slide presentation, forgot to create notes for the slides, and then had AI generate all of the speaker notes for them automatically based on the content of the slides. They also showed off what Google's new conversational search is going to look like. It looks very similar to what Bing has planned for their search engine. You can see they asked a question, what's better for a family with kids under three and a dog, Bryce Canyon or Arches National Park. It actually generated a response with AI, gave some potential follow up questions, gave some sources, and then below that gave the search results. They also showed off some features for shopping on Google which will help you compare various products using AI on Google. For example, showing E bikes in red for 5 mile commutes with hills and it was able to find stores near their location that had red bikes that were good for working on hills. They also showed off their new vertex AI, which to me feels very similar to Nvidia's foundation platform or Amazon's bedrock platform where companies can build their own custom large language models on top of their GPU resources. Now, of all of the things that they showed off, their Project Tailwind was probably the single thing that excited me the most. Because what it is is basically you create your own intern chatbot based on files that are inside of your Google Drive. So if you have a whole bunch of documents inside of Google Drive, you can essentially query those documents with a chat and it will use those documents as the sources to answer your questions and then cite the source from your database of documents. It's basically creating a chatbot by embedding the content of your Google Drive for you and then making all of that content of your Google Drive chattable. That to me seems super valuable because you can create folders on your Google Drive for specific topics that you're interested in and add various documents that you find around the web, write your own content in there, save transcripts from interviews and videos and whatever you can imagine in there, and then create your own custom tailored chatbot to query all of those documents that you saved in there. That concept to me is really exciting. So this Project Tailwind is something that out of everything they showed throughout this entire keynote is probably the thing I'm most excited to get my hands on. They also showed off a universal translator where they took a video of somebody speaking, translated that video into a different language, and then the person's mouth and voice actually followed along as they were speaking in that other language. So here's the example.

What many college students don't realize is that knowing when to ask for help and then following through on using helpful resources is actually a hallmark of becoming a productive adult. Muchos universitarios no comprende que saber cuando pedira yuda Yusar recursos utiles en realidad una clave para convertirese en una dulto productivo.

So that's pretty cool. But they also mentioned during the this keynote that they're not going to make that available for everybody because they're too afraid of it being used for deep fakes and things like that. So when the general public will have access to this from Google, who knows? During the event they also presented a few new phones. They have a new Pixel folding phone and a Pixel 7a. I wasn't too interested in that. I was more interested in the AI features. You're going to be able to use AI to help you with your text messages where if you've ever seen that south park episode where they talked about ChatGPT, this feels straight out of that episode. They're also going to let you use generative AI to generate AI backgrounds, which, I mean, cool. And then they moved on to the developer keynote. So there was two main keynotes during this IO event. You had the main Google keynote and then you had the developer keynote where they focused a lot more in on the APIs and the various development software that was available for programmers and things like that. They did mention that there's a whole bunch of APIs that developers are going to get access to and a whole bunch of code helpers similar to like a CodeWhisperer or GitHub Copilot. Google's gonna have their own version of that. They also announced their own Android AI development platform and a whole bunch of cool tools for builders. But most of what they talked about in that portion was over my head. They used a lot of coder speak and industry jargon that I couldn't really translate other than the fact that it was tools to make coding a lot faster and a lot easier if you're going to code with Google's platforms for tools like Android. And finally they wrapped up the event with really fun, which was showing off their Space Invaders World Defense augmented reality game which actually uses Google Earth plus augmented reality technology to add Space Invaders directly into the real world. So you can see this is a Google Map street view here. And then they added Space Invaders right over the top in reality and they call this game Space Invaders World Defense. And this was how they wrapped up the entire Google event. So this is something that I found really cool, really fun. I'm excited to see this play out. But they also talked about how this augmented reality technology is available for developers to develop on. So you can use the Google Maps tools and their augmented reality tools to build games like that yourself. They call it the AI Core Geospatial API and developers will have access to that API to build similar stuff to this augmented reality Space Invaders game. That's pretty much my recap of everything that they talked about during Google IO. I do have a 15 minute recap on Twitter again that will be linked below. It really is a super edit of four hours down into just.