If you look at the tabs across the top of my screen, I've got a lot of really interesting things that have been happening in the AI world that I want to share with you.

Despite the fact that it looks like there's a lot going on, I think we're actually going to look back at this week and consider it a slower week compared to what's coming.

We're about to get hit with a huge wave of AI announcements coming out of all of the companies developing AI, but I'll talk about that more at the end of the video.

For this one, I want to get into what is interesting right now.

Starting with the piece of news that came out this week, that is probably the most groundbreaking news we've had in a long time, yet it seems to be kind of under reported at the moment.

This week, Google's DeepMind introduced AlphaFold 3.

This is a model that can accurately predict the structure of proteins, DNA, RNA and ligands and how they interact.

The previous AlphaFold models were huge breakthroughs.

I mean, like, they're talking about the creators potentially getting Nobel prizes for them.

At some point, this is significant because AlphaFold can accurately predict how these various molecules will interact with each other and actually find new ways of combining molecules that scientists haven't actually tried combining yet.

This could potentially lead to an explosion of new drug discoveries, new vaccines, and even to greater insights about how the human body actually works.

This is some research that probably should get its own dedicated video.

I'll probably do that in the future, but this announcement from Google's DeepMind is pretty impactful.

In fact, Demis Hassabis, the CEO of DeepMind went on Bloomberg to say this.

Again, I feel like that topic requires a much deeper dive video, but since this video is designed to give you the grand overview of all of the important stuff happening this week, I'm going to go ahead and move on and tell you about Demis' ex business partner, Mustafa Suleiman and his new project over at Microsoft AI.

In a recent previous video, we talked about how Mustafa actually left Inflection AI who created Pi to go and become a CEO of Microsoft AI.

Well, now we're starting to get details of what he's actually working on over at Microsoft AI.

He's working on a new model called MAI-1 with a rumored 500 billion parameters.

This is actually the first time that Microsoft has worked on a Large Language Model of this scale.

They've worked on some smaller language models, but for their Large Language Models, they've famously been partnered up with OpenAI as their largest investor.

Most of the tech that Microsoft uses for their AI platforms is all OpenAI's tech.

We also know that Microsoft has a relationship with Meta and leverages their open source LLaMA models as well.

It's really interesting to see that Microsoft is starting to build a sort of GPT-4 level model in a house, knowing how closely tied they are to OpenAI.

In all likelihood, this is just Microsoft making sure they're very diversified as part of the contract between Microsoft and OpenAI.

If OpenAI achieves AGI, the terms of their deal change.

Microsoft I still believe gets some sort of payout up to a certain amount, but then the relationship with the for-profit OpenAI changes.

Microsoft's upside from OpenAI dramatically diminishes.

To me, it seems like what they're doing is, well, they're working with OpenAI, they're working with LLaMA, they're working with Mistral, and they're building their own in-house models so that no matter what happens to these other AI companies and what the relationships end up doing, Microsoft is still fully capable of achieving whatever their bigger vision they've set out to achieve is.

While we're on the topic of Microsoft, they're apparently investing $3.3 billion dollars into an AI data center in Wisconsin.

Building this data center is expected to create an estimated 2300 union construction jobs and up to 2000 permanent jobs.

All right, now let's talk about OpenAI.

This week, Stack Overflow and OpenAI partnered to strengthen the world's most popular Large Language Models.

If you're not familiar with Stack Overflow, it's basically a website where people who are creating things with code can go and ask questions about their code.

The community responds, almost like a Quora, but for code.

You ask questions about the code, and then people sort of help you answer your coding questions.

Apparently, since ChatGPT came out and people found out that it was pretty good at coding, the traffic to Stack Overflow has been tanking.

Stack Overflow tried to launch their own AI model, but I don't really think it ever went anywhere with most people.

It appears that OpenAI and Stack Overflow are partnering and working together so that OpenAI can train on Stack Overflow's giant database of coding data.

It should really, really improve the coding ability of tools like GPT-4 and whatever their next generation model is.

However, people that have contributed to Stack Overflow are not necessarily happy about this.

Here's a tweet from Nixcraft that says, Stack Overflow mods are experiencing some frustration as several users have been deleting their answers since the announcement with the OpenAI partnership.

As a result, they have started suspending accounts that engage in this behavior.

Apparently Stack Overflow's terms and conditions grant them special permission to retain all answers.

If you've ever posted an answer on Stack Overflow, well Stack Overflow now owns that answer and you are not allowed to remove it, meaning everybody who's contributed to Stack Overflow sort of unknowingly now contributed to OpenAI's coding ability.

Last week, I talked about gpt2 chatbot, which was this mysterious chatbot that popped up inside of Limsis.org, the chatbot arena.

Within 48 hours of people actually discovering it and using it, it was yanked off the platform.

It's back.

You can actually use this new model, which is still mysterious.

The prevailing theory is that it is some sort of model from OpenAI, but OpenAI has been very hush-hush about it, only sort of tweeting troll-like memes to give us any clues that they're the ones behind it.

I did a super quick breakdown of what's going on.

I was testing a sort of new style of video.

It's slightly less than four minutes called the mysterious chatbot that beat chatGPT is back.

If you want to know all of the details of what we know about that model, check out that video.

It'll give you the breakdown in a rapid fire three minute and 45 second video.

Also in OpenAI news this week, they put out a blog post called Our Approach to Data and AI.

The main thing to note in this one is that they're claiming they respect the choices of creators and content owners on AI.

They're going to be building a media manager so that content owners can manage how their works are used in AI.

If you wrote an article or created a piece of art or something like that, and you don't want it trained into any of OpenAI's models, they're building some sort of platform they're calling the media manager so that you could make sure that the content you want excluded actually gets excluded.

They say the goal is to have the tool in place by 2025.

Next year, sometime we'll have this media manager and can actually control what content does and doesn't get trained into the systems.

Theoretically.

There's been a rumor going around.

We talked about it in last week's news video that OpenAI is about to launch their own search engine. If you're following along, Microsoft who uses GPT-4 from OpenAI is now developing their own Large Language Model.

OpenAI who uses the search functionality of Microsoft is building their own search engine.

Again, there's not a hundred percent confirmation that this search engine is going to get released next week.

But according to Watcher Guru, ChatGPT creator OpenAI is to announce Google Search competitor on Monday.

We'll find out more next week.

If you have an online business, or if you're trying to grow an email list, it can be really difficult to Figure out what words and what sort of style to create a landing page in to get people to actually sign up or buy your products.

That's why for today's video, I partnered with HubSpot.

I wanted to show off their custom GPT inside of ChatGPT.

It helps you build really good landing pages.

It's completely free to use for all ChatGPT Plus subscribers.

It's actually pretty simple.

You jump into the GPT.

It's easy to start by using their pre-written prompt for you.

I want to create a landing page and then it's just going to ask you a few questions to make sure it gets all the details right.

I'll give it a quick two or three sentence paragraph about what the site I'm trying to build is supposed to do. In this case, grow the Future Tools email newsletter.

It wants to know what makes the newsletter stand out.

I'll simply share a few bullet points here.

It'll ask me a couple more questions about what I want my call to action to be.

In this case, I'm going to use be subscribed.

I'm just going to say B then it asks what sort of tone I want it in for this one.

I want it to be informative.

I'm going to go with I then immediately wrote some copy for me.

I think the sales copy to get people on my newsletter looks great.

I'm going to go ahead and say, looks good. Let's create it.

I'm going to confirm that I trust this HubSpot GPT here.

A couple seconds later, I have a link to my landing page, which I can click on and it will open the page in a new tab.

We can see we have a landing page. That's nice and designed for us with really decent copy.

We can easily swap out these images with the most relevant images and tweak any of the copy that we want.

Within roughly two minutes, we have a landing page ready to start growing our mailing list.

This custom GPT from HubSpot makes it super, super simple.

Again, if you have a ChatGPT Plus account, it's totally free to use.

I've gone ahead and put the link in the description so you can access it easily, or you can try the QR code that's on the screen right here.

Check it out.

It's pretty impressive with how quick and easy it is to set up.

Thank you so much to HubSpot for sponsoring this video.

Also this week, gradient AI announced that they managed to get a 4 million token context length out of LLaMA three eight B. That is the largest context window of any openly available Large Language Model.

A 4 million token context window is essentially 3 million words of input and output back from the Large Language Model.

To put that into some more context, the entire Harry Potter book series is a little over a million words.

Theoretically you could feed in three times the size of the entire Harry Potter series and actually get responses based on that context window.

If I had to guess, I would guess that the whole concept of a context window is going to sort of disappear over the coming years.

The context windows of these Large Language Models are going to get so large that context window isn't even really something people think about that much.

What's even cooler about this gradient AI 4 million context length is that they developed it with the LLaMA three eight B model, which is open source and then shared what they created over here on Hugging Face.

If you actually know what you're doing, which in this case, I really don't, you can come to Hugging Face, go to files and versions and actually download the files to be able to run this yourself and get that 4 million token context window.

We also got some news this week out of X AI and their Grok chat bot.

If you have a premium account and you're using either the desktop version or the iOS version, you can go to the explore section, click on any article like this iPad article here.

This little description of the article was actually generated by Grok.

We get that with all of the news over here in the explore tab.

Any one that I click into here will give me a nice quick AI generated summary of what's going on based on all the tweets like this one about my favorite baseball player of all time.

All right, moving into AI art this week, Crea AI actually launched video inside of your platform with their little example.

They showed two images, they blended them together and generated a video based on the blend of the two images.

It looks pretty sweet.

Ever since this was released, I've been seeing some really cool videos getting posted onto X here.

Like this one from Steve Mills that has this crazy person on a bed zooming in with a demon and then they turn into a house and then they're on a bed.

It's just this crazy, crazy looking animations.

I mean they're not realistic like what you'd get out of something like Sora, but they've got this really cool sort of frame interpolation.

I don't know.

It's just got a really cool feel to it.

It seems like it'd be really good for like music videos and things like that.

Here's another 10 second clip that Justine Moore shared of a bunch of cities, sort of Morphing between each other and some cool clouds in the sky.

Really, really cool videos.

Crea isn't publicly available to everybody yet.

They're starting to hand out some access codes to some people to test it out.

I managed to get an access code to get in.

I'm not a hundred percent sure when they're going to be opening it up to more people, but let's go ahead and take a peek real quick.

Inside the app is actually really impressive.

You can see we've got a timeline down here and I can actually move my mouse along the timeline and see every single frame.

You can see some key frames here where it sort of interpolates between the two images that it started with.

I can add additional key frames all along the timeline here.

I click into settings.

We've got aspect ratio, we've got motion intensity.

We've got the option to loop the video.

Down here, we've got a few options like film, render, animation, and experimental.

I'm going to go ahead and clear everything that's here.

We're starting from scratch and then let's go ahead and toss a couple of random images in here.

I'll toss this little like fantasy world image that I created in Midjourney in here.

Let's add another key frame down here.

Here's another sort of similar fantasy image.

Let's go ahead and add our text prompt.

I've added two images and a couple of text prompts to go with it.

Let's go ahead and stretch this out to right here, stretch this out to right here.

Let's go ahead and generate a video and see what we get out of it.

Here's the video I've made.

That's really, really cool.

It's just a really cool thing to be able to take one image. It's like your starting image and then where you want the image to end up and make a video that kind of goes between the images.

This one seems to do it pretty smoothly.

What Cray is doing here is actually using open source technology called animate diff under the hood, which means there are free alternatives that you can use, albeit a little bit more complicated, but you can achieve the same effect locally.

I'll actually link up a YouTube tutorial from PERS in the description below.

If you do want to take a stab at trying to do the same effect locally, there's a pretty in-depth tutorial that you can follow along to.

While we're on the topic of AI video, we actually got some more details about Sora this week.

Sean Ralston here who works with OpenAI on dev support, shared this video showing that you can change just the character in the video without it changing the rest of the scene.

If you watch this video, you've got the top left with a man walking down the street, top right, same exact alley street, but with a woman down here, an older man and over on the bottom right, a robot, but everything else pretty much remains the same.

Some of the graffiti in the background looks a little bit different, but for the most part it's generating the same scene, but with the ability to in-paint a different character with each scene.

That seems pretty cool.

If you make a video that you really like, theoretically you can go, okay, I don't like the character that it put in there, but I like everything else about this video and just change the character.

That'll be pretty cool to see when we finally get our hands on Sora, whenever that may be, who knows.

This week, the tool Udio, which everybody's been raving about with how realistic the music that comes out of it sounds just recently rolled out a new update as well.

They actually have a feature now called audio in-painting where you can select just a little piece of the song and tell it to change just that chunk.

If you like most of the song, but maybe the lyrics in one part came out a little funky, you can select just that portion of the song and have it fix just that chunk.

Udio has also run into a little bit of hot water recently with a lot of people pointing out that you can prompt Udio in a way that almost makes carbon copies of existing songs, sort of hinting to the idea that Udio was trained on a lot of copyrighted songs that they didn't have the permission to train on.

I'm hoping they get that stuff sorted out because it's not great that I can go and make a song that sounds nearly identical to an existing artists work.

I do think they'll iron out the bugs and fix some guardrails so that people won't be able to do that.

But from a technology standpoint, I'm still pretty blown away by what this is capable of, especially now that it has these new in-paint features.

ElevenLabs who pretty much does everything else with audio also appears to be getting into the AI music generation game.

With ElevenLabs, we can already make realistic voice clones, realistic audio translations, sound effects, all sorts of cool stuff with audio.

This week, they tweeted here's an early preview of ElevenLabs music.

All of the songs in this thread were generated with a single text prompt, no edits.

There's a whole bunch of examples, but let's go ahead and just take a super quick listen to the one that they put at the top here.

It sounds just like Tom Petty, but still very impressive what they can do with some of these tools.

I'm going to go ahead and just sort of rapid fire a bunch of other smaller news items that came out.

I mean, I guess it's a whole relative.

Several years ago, if some of this AI news came out, it would have been like some of the biggest news in the world.

But comparatively to some of the other news we're talking about, this feels like smaller news.

Some of this can probably use a more dedicated video down the line, but here we go.

Apple held an event this week where they showed off their new iPad model with the M4 chip.

I'm pretty sure this is the first keynote where Apple ever actually talked about AI.

They sort of danced around the topic in almost every other keynote that they've done.

But this M4 chip here apparently has the fastest neural engine ever capable of 38 trillion operations per second, which is faster than the neural processing unit of any AI PC today.

Further down here, it says the M4 is extremely well suited for the latest applications leveraging AI.

Even Apple's iPads now that they're putting out are sort of being designed to future proof them for AI.

This week, Hugging Face also launched Le Robot, which is an open source robotics code library.

They actually made it all available on GitHub.

Le Robot aims to provide models, data sets and tools for real world robotics in PyTorch.

The goal is to lower the barrier to entry to robotics so that everyone can contribute and benefit from sharing data sets and pre-trained models.

If people are out there trying to build robots, you now have models available that are open source that you can use to sort of fast track your way into building robots.

We also got a peek at Dr. Eureka this week, which is a new research paper out of the University of Pennsylvania, NVIDIA and the University of Texas Austin.

We sort of got a sneak peek of this type of technology at NVIDIA's GTC earlier this year, but these robots were actually trained in a virtual environment and learned how to do the things that we're seeing in this video in a virtual environment or a digital twin.

Once it learned and understood how to do these things in the virtual environment, the sort of code and training was then put into the real robot and the robot was able to perform these things in the real world, even though they weren't trained to do these things in the real world.

They trained it to do things like walk on a exercise ball.

This dude here is even kicking the exercise ball to try to throw the robot off and it still manages to maintain its balance.

I came across this random video on X. I actually don't know what factory this is in China, but they showed off this robotics factory and these are obviously robots.

But the first time I ever watched this video, there's a scene right here in a second where it walks past this guy.

The first time I watched this video, I didn't even realize this dude was a robot here.

I thought this was just somebody standing here watching the robots being made, but nope, that's actually one of the robots.

To me, this is like straight out of Westworld.

Like they look really realistic.

Again, I don't really know much about this video.

I just kind of came across it on X. I thought it was cool and wanted to share it in this video.

Some people in the comments are claiming that this is actually fake.

This isn't even a real video.

I have no idea.

Take this one with a grain of salt.

This might have been made with Sora for all I know.

Also this week we got the news that machine learning aided in the discovery of sperm whale alphabet.

Using machine learning, they were able to sort of tell the difference between various noises that the whales made and find patterns and correlations between them to sort of start to map out this sperm whale alphabet.

Is this leading us to a future that we could maybe even talk to whales?

That could be interesting.

I also mentioned at the beginning of this video that I think this is going to be considered a slow week compared to the weeks that are coming up.

This is the reason why between May and June, we have these events.

We have Google IO.

This is Google's big annual event where they make a ton of announcements.

Last year it was all about AI.

This year is expected to be all about AI as well.

From the day I'm releasing this video, that event happens next week.

Also next week is the rumored OpenAI update event where we're expected to hear about OpenAI's new search engine project.

The following week is Microsoft build up in Seattle.

Again, a big event that happens annually where Microsoft makes some of their biggest announcements this year, likely to be mostly focused on AI.

There's Computex in Taiwan where Jensen Huang, the CEO of NVIDIA is one of the keynote speakers, possibly more announcements.

There's Cisco live an event that I'm going to, it's going to be mostly focused around cyber security.

There's going to be some big announcements around how we can better protect ourselves against things like deep fakes and voice cloning scams and things like that.

That is something that Cisco has been working really, really closely on.

Many of those announcements are going to happen at that event in June in Vegas.

Also in June, we have Apple's WWDC event, the worldwide developer conference.

If you remember last year, that's when they announced the Apple Vision Pro.

Well, this year, some of the rumors are that maybe we're going to get a new Siri powered by either OpenAI or Google Gemini.

I really honestly don't know what sort of announcements to expect from WWDC, but all indications point towards it being a very AI heavy sort of announcement day.

That's why I think this week is probably the calm before the storm, even though it feels like I just talked about a lot.

It's only going to ramp up from here.

I think as we get into the summer, we're going to see some crazy announcements.

I'm here for it.

I'm excited.

If you want to stay in the loop and make sure that you know about all of these things and that you don't miss any of it, make sure you're subscribed to this channel.

I put out these AI news recaps every single Friday to loop you in on all of the news that happened throughout the week.

You can get caught up with just one video week of everything you need to know.

If you like this video specifically, give it a thumbs up.

I really, really appreciate it.

Also check out the next wave podcast.

We recently had Logan Kilpatrick on.

He was the head of developer relations over at OpenAI.

He recently left OpenAI and is now working over at Google doing very similar work.

We had an interview with him where we asked, well, why did you leave OpenAI?

He broke it all down for us.

We also talked about open source first closed source and really went deep down the RABBIT hole with Logan.

Those of you that want to explore a little bit deeper into the AI space than what I provide on this channel, definitely check out the next wave podcast.

That's where we really, really deep dive.

As always check out futuretools.io, where I curate all the coolest AI tools.

I keep you up to date on the latest AI news.

Believe it or not, the AI news that I share in this video isn't all of it.

I have to cut some of it out just to make sure these videos aren't two hours long, but all of the interesting news that I come across gets shared on this news page.

Anything I missed in this video should be here.

I also record these videos on Thursdays.

Anything that comes out on Friday, a lot of times gets missed in these videos.

Of course, there's a free newsletter where I'll keep you looped in with just the most important news of the week, just the coolest tools I come across.

It's totally free.

You get access to the AI income database when you sign up, which is a cool database of ways to make money with AI.

Again, totally free.

Find it over at futuretools.io and I'll make sure you get hooked up.

That is way too many things that I just asked you to do.

I apologize, but thank you so much for watching this video and nerding out about AI with me.

That's what's important is that you like to come and hang out and nerd out about the latest and greatest in AI.

It's so much fun.

Thank you so much for tuning in.

Hopefully I will see you in the next video.

Bye.