Actually Big AI News Week - Here's What You Missed

Believe it or not, the past week saw an insane amount of AI news, and I want to try to break it all down for you, so let's just jump straight into it. At the end of last week, Elon Musk did a 45 minute livestream on X of his fully self driving V12 beta software showing full AI driving in the Tesla. During the live stream, Elon talked about how it's using all of the cameras around the car plus AI to do all of the driving. And. And it can actually work without an Internet connection. It does not need to be connected. All of the actual information is stored in the car's computer. For the most part, the drive went pretty smoothly. It navigated traffic cones, it navigated roundabouts, it obeyed stop signs, obeyed stop lights. There was one slight hiccup in the drive. Basically rush hour.

Intervention. Sorry. Okay, so that's our first intervention.

The car almost ran a red light because it saw the green turn signal. But after their intervention, that didn't actually happen again for the rest of the drive. And he also talked about how every time an intervention like that happens, the car actually gets smarter and learns from its mistakes. So it's constantly improving and getting better. Speaking of Tesla, they unveiled a $300 million supercomputer. The supercomputer is designed to accelerate the training on Tesla's self driving AI. And it contains over 10,000 Nvidia H100 GPUs, making it one of the fastest supercomputers in the entire world. Remember last month when we talked about Sarah Silverman and other authors suing OpenAI for training on their books? Well, this week, OpenAI responded. OpenAI claimed that the authors misconceived the scope of copyright, failing to take into account the limitations and exceptions, including fair use, that properly leave room for innovations like the large language models now at the forefront of artificial intelligence. According to OpenAI, even if the author's books were a tiny part of ChatGPT's massive data set, the use of copyright materials by innovators in transformative ways does not violate copyright. Unlike plagiarists who seek to directly profit off of distributing copyrighted materials, OpenAI argued that its goal was to teach its models to derive the rules underlying human language. So basically, OpenAI is arguing that they are actually protected in this case under fair use, and that the work is transformative and really only being used to train the models, not re regurgitate exactly what's in the content of these books. Speaking of copyright, the US Copyright Office actually wants to hear from you about AI and copyright. The agency wants to try to get answers to three main how AI models should use copyrighted data in training, whether AI generated material can be copyrighted even without a human involved and how copyright liability would work with AI. Here's the actual document from the Copyright Office with their Notice of Inquiry and request for comments. If you want to submit your own comments about your thoughts on copyright and AI, you can go to copyright.gov policy artificial intelligence click on the Comment Submission button. There's some instructions on what you need to do. Here they are allowing individual commenters and organizations to leave comments. Go ahead and click on the button at the bottom, and here is a page where you can submit your comments and attach any additional details Speaking of worries around AI, this week a survey found that relatively few Americans actually use or fear chatgpt. People like you and me who watch videos like this, we tend to be in a little bit of a tech bubble where we see AI everywhere and we're constantly thinking about AI. But according to a poll from Pew research, only about 18% of Americans have ever actually used ChatGPT. And those that do use it, most people use it for entertainment or to learn something. Only 19% of employed people who'd heard of the model thought it would affect their job in a major way, and 27% of the people polled expect no impact whatsoever. And what was even more interesting is only about 15% of people thought it would actually be helpful. However, in a different poll from Pew Research, 47%, up from 31% last year, said that AI actually makes them more concerned than excited. But do you know who is trying to be helpful with their AI? ChatGPT themselves. They're trying to help team teachers better use ChatGPT in the classroom. They put out a blog post saying, here's how teachers are using ChatGPT. Things like role playing, challenging conversations, building quizzes, tests and lesson plans for curriculum materials, reducing friction for non English speakers, and teaching students about critical thinking. And they even give example prompts to get educators started in better using ChatGPT in the classroom. Speaking of ChatGPT and OpenAI, a few weeks ago we talked about how OpenAI could possibly go bankrupt by 2020 because the company is spending $700,000 a day. However, OpenAI this week seems to be turning a Corner. They introduced ChatGPT Enterprise version, and if you sign up for the enterprise version, you get the most powerful version of GPT4 yet. You get unlimited access to GPT4, so no usage caps. You get up to 2 times faster speeds Unlimited access to Advanced Data Analysis formerly known as Code Interpreter I'm learning right now as reading this that they're rebranding Code Interpreter to Advanced Data Data Analysis. You get the 32,000 token context window so you can get up to four times longer inputs, files and follow ups. You get shareable chat templates for your company to collaborate and build common workflows, and you get free credits to use their APIs if you need to extend OpenAI into a fully custom solution for your organization. Now if you check out OpenAI.com enterprise, you can learn more. However, the enterprise plan doesn't have a price tag on it. You've got to contact sales to learn more about how much it costs. And seemingly companies are signing up because OpenAI has been shattering revenue expectations and they're predicted to now generate over $1 billion this year. So that calculation where they might be going bankrupt by 2024 probably needs to be rethought a little bit. Enterprise customers include companies like Zoom, Stripe, Notion and Databricks. This article talks about how the trend is increasing toward customized AI assistance for enterprises that can use them. However, despite OpenAI having revenue that's way better than expected, it seems like every other company out there is coming for them. For example, this article here stirred up some controversy this week when it said that Google Gemini eats the World Gemini smashes GPT 4 by 5 times the article argues that most other AI companies are training on GPUs and are super reliant on companies like Nvidia. However, Google and DeepMind have have TPUs and a quick search for TPU versus GPU. TPUs are designed specifically for tensor operations, resulting in faster training and inference times for neural Networks compared to GPUs. TPUs are also more power efficient than GPUs, making them a better choice for large scale machine learning deployment. Because Google has so many of these TPUs at their disposal and they're using those to train Google. Gemini Gemini will supposedly be five times more advanced than what we're seeing from GPT4 right now, but Gemini isn't the the only one out there trying to eat OpenAI's lunch right now. Two new similar chatbots just launched in China. The Ernie chatbot was just approved by the Chinese government this week, which is actually available in the U.S. however, it's only in Chinese. They also opened up a plugin marketplace for the Ernie bot, similar to ChatGPT plugins. Now, because I don't know or speak Chinese, I haven't actually been able to play with this one myself. The second one that came out this week in China was released by Alibaba and this model was called Quinvl and Quinn VL Chat. I have no idea if pronouncing those right, so I apologize if I'm not what's different about this model is this model actually reads images rather than text, so it can do things like providing directions by scanning street signs, solving math equations based on a photo, and weaving together a narrative based on multiple pictures, which are all actually things that in the past have been promised by tools like GPT4 and Google's various models yet. We haven't seen any real great implementations of that in practice with any of the tools that we have available to us today. And of course as we all know, metas also coming after OpenAI. Now this one's just rumor right now, but supposedly they're working on a llama 3 which will be about as powerful as GPT4. This comes from a tweet or X or whatever overheard at a meta gen AI social. We have compute to train llama 3 and 4. The plan is for llama 3 to be as good as GPT4. Wow. If llama 3 is as good as GPT4, will you guys still open source it? Yeah, we will. Sorry alignment people. So again, not much to go off of other than a hearsay post on X. But if we can have models as powerful as GPT4 and they're open source, meaning anybody can iterate build off of them, run them, build services around them in the cloud, or you'll be able to download them and run them locally, well that could spell some extra trouble for companies like OpenAI who rely on the revenue of things like ChatGPT plus ChatGPT Enterprise and selling API access. Also this week Google held a big conference called Google Cloud Next. This is an event where Google tends to make a bunch of announcements about a lot of their latest cloud features and to be honest, most of the event was super technical over my head, but there was some really really cool stuff that they broke down and I made an 18 minute super cut and posted it on my X account. If you want to watch the entire hour and a half presentation in less than 18 and a half minutes, here are the things that I think you'll probably find most interesting. They announced a partnership between Nvidia and Google Cloud and they even brought CEO of Nvidia up on stage Jensen Huang to talk about the partnership and how Nvidia's DGX is going to be added into Google's cloud services. They also announced a new way that they're going to start digitally watermarking images that were generated with AI to help, you know, prevent misinformation and things like that. And they're not doing it in traditional ways, like through metadata or anything like that.

There are some challenges with digital watermarking today. The first challenge is that if they're metadata based, they can be modified relatively easily. If they're pixel based, what happens is that the images don't look really well. That is not true for our new digital watermarking capability that we're announcing today. I can do a single click here in Generative AI Studio and I'm going to go ahead and generate watermarked versions of each of these images. And now you see the digital watermark is stamped on every one of these images. So we know that the fidelity and the integrity of our data is intact.

But I felt like the real coolest stuff was when they started talking about announcements around Duet AI and how it's now available inside of Google Workspace.

Today we're pleased to announce that Duet AI in Google Workspace is now generally available to everyone.

Every time you use Gmail Drive Docs, Meet Chat Calendar in your daily life, you're using Workspace.

Okay, so we'll begin right here in Drive, where I'll select a prior brief as a starting point. How about this one? And from here I will just open the Duet sidebar and I can already see a suggestion to use the selected file as a template. But I'll go ahead and click that suggestion. Use this template. Now, Duet just needs to know what kind of document we're trying to create. In this case, it's for a new line of pet toys called Project Eco Paws Brief. Okay, now in the background, Duet is looking across my drive, finding all the relevant information about Project Ecopause and combining it with the template we just selected to create this brand new presentation that's been generated right before us. And I can even see it is found in reference sources across my drive. There's a Google Doc, there's a Google Sheet. Why don't we go ahead and click it, see what we got back.

And that's amazing. Even brought in charts from the right Google sheets.

I know we want to market these toys to Gen Z, so we're going to go ahead and have Duet help us with that. I'll start by clicking this, add a slide suggestion, and then we'll just ask Duet to identify trends to incorporate in the brief for Gen Z by typing Gen Z Consumer Trends. And now we're using our Palm 2 model and all of its base knowledge about Gen Z plus the contents of this generated deck to put together this brand new generated slide. Understanding Gen Z. I'll go ahead and just click Insert and now we've got a great point from which we can continue to refine Aparna.

How about that?

We've all presented in meetings and know how overwhelming it can be. Duet's got you covered. First, you can present with confidence with our brand new real time teleprompter in Google Meet. It follows your voice as you read your speaker notes and a real time speech recognition system keeps up with you even if you jump ahead or reword notes on the fly. As for notes, Duet is automatically capturing notes, action items and video snippets to share with all the attendees at the end of the call.

With Duet, Meet can automatically detect and translate over 300 language pairs, providing real time captions for the viewers. Say you joined the meeting late, and let's be honest, who doesn't join meeting late? Occasionally, Duet provides an interactive catch me up mode, complete with a summary of everything you've missed before joining the meeting. You can even ask it follow up questions like let's see here what's been said about Gen Z. And in just a second I'm feeling fully caught up in everything that I missed.

So basically Duet connects all of the tools that you have inside of your Google workspace like calendar, slides, sheets, docs, all of that stuff makes it all searchable, all chattable, and when you're making a slide presentation, you can pull data from your Google Sheets. When you're making a Google Sheet spreadsheet, you can pull information from docs and everything just sort of connects through this chat. And of course there's also the new features inside of Google Meet where it will automatically take notes, it will transcribe subtitles so that you can speak with anybody in any language and have it transcribed for you. While you're on the conversation, you can have it catch you up on what you might have missed when you join a meeting late. So lots of cool stuff. Now the presentation went on to show off a whole bunch of AI features for coders and developers who want to use a lot of their new suite of tools. And again, if you want to watch the full 18 minute breakdown where I show you all of the announcements, I'll make sure the link to my X post is below in the description. I also recently made a video where I was experimenting with Google's new AI search features. Well, this week they also announced that those features are now available in India and Japan. So great news for those countries. Also, in my recent live stream, I played around with Ideogram, which at the moment is a completely free AI image generation tool that can actually generate really good, decent text inside of images. So as a real quick example, I can do a penguin holding a sign that says says subscribe to Matt Wolf. Click, generate, and usually at least one of the pictures comes out. Pretty decent. And there you go. Look at this one. Subscribe to Matt Wolf. Here's another one. Subscribe Matt to WOLFF. It's not 100% all the time, but if you look through some of the examples, look at this. Do you want negative prompting? Be happy, Be creative. Smart up. Just do it. Let's code. Maybe it doesn't get it perfectly right on the first generation, but it does do text and it does it pretty dang well. The reason I'm bringing that up is it's now available to everyone for free. When I talked about it in last Friday's news video, you had to join a wait list. Now anybody's in and it's totally free. Get in and play around with it and see what you can make it do. It's really cool. As I learn more about it, I'll probably make a video in the near future about best practices and cool tricks that I learned while using it. If you're a Substack user, Substack just announced a new feature for audio transcription tools. So if you actually publish an audio podcast on Substack, they now use AI to automatically transcribe your podcast into a written blog post for you. A few months ago, Meta announced their Dyno V2 State of the art computer vision models, which does a really good job of tracking objects inside of videos. Well, this week they announced that they're actually open sourcing. Dyno v2. Meta seems to be going down this path of pretty much open sourcing all of their research that they feel comfortable getting into the public's hand, which is pretty cool in my opinion because it allows anyone to sort of build off of it and improve upon it. They also introduced a new model called Facet, which stands for Bareness and Computer Vision Evaluation. The computer vision model helps classify additional details inside of images, things like perceived gender, perceived skin tone, age group, hair color, lighting conditions, and more. And this was trained by using actual humans who looked through 32,000 different images, and actual humans annotated the images with demographic attributes to help train the model so that the model could then in the future figure out those same attributes for future images and videos. So it's trying to eliminate some of the biases. Most image generators and video generators are going to tend to bias skateboarders as males. Well, this is a female skateboarder and the perceived gender is more female ness. So basically it's a tool to help better change some of those biases in the data set. There seems to be a big trend lately of announcements around AI and food. This week Samsung launched its AI powered recipe app called Food. It's an artificial intelligence that provides users with personalized recipes and food recommendations. They launched it in eight languages and 104 countries. It has 160,000 recipes and it was designed to be a personalized assistant that helps users discover new dishes, create tailored meal plans, and order ingredients online. Also this week, Doordash announced that they're developing AI powered voice ordering for restaurants. It's designed for an AI powered system to help restaurants take orders over the phone. Without any dedicated merchant manpower. Customers will have personalized voice ordering experience in multiple languages with no missed calls or long wait times. Returning customers will be able to quickly reorder their favorite meal and live agents will be available to jump in to support customers at any time. So I imagine when you call your favorite restaurant in the near future, it will be a robot that answers you. Place your order with the robot and if you've ordered there before, the robot already knows your preferred order. I mean, honestly, I'd almost prefer it that way. Also this week, Uber Eats is reportedly developing an AI chatbot that will offer recommendations and speed up ordering. It's building a chatbot that will ask users about their budget, food preferences and then help them place an order. Now here's some fun stuff that people can go and play with right now. One of my favorite tools to play with is Blockade Lab, where you can create these 360 landscapes with just a prompt. They've added some new features recently. For example, they added the Enhanced Prompt feature where when you enter a basic prompt, it adds more detail to the prompt for you. They also added a whole bunch of new models that you can play with, like Epic Digital painting, Radiant Enchanted, 1960s ethereal fantasy, psychedelic Illustration, Whimsical World, and a bunch more. Just for fun, let's try Psychedelic Illustration, a Colorful World Made of Candy. We'll leave our enhanced prompt on and click Generate and you'll see that it actually built out the prompt even more. VR360 vision of vibrant candy Cosmos swirled lollipops as planetary bodies shimmering sugar dust galaxies, rainbow gummy asteroid belts, cotton candy nebula developing in a surrealistic candy art style. And that's what this generated. And if I use that same prompt with Whimsical World, I get something that looks like this. Pretty whimsical. I just love Blockade Labs wanted to show that. Also this week, Runway ML launched a few new features. The first one is, if you go to generate videos and do a text or image to video, there's actually now a motion level. So you can set the motion level from 0 all the way up to 10. And the higher the number, the more motion will show up in your video. So just for fun, let's grab this random image from my hard drive here, and I'll pull this in. We'll pull the motion all the way up to 10 and click generate. And here's the video we got of it, which has just absolutely tons of motion in it. Now, if I bring the motion scale down to, let's just say three, and click Generate. And here's the video we get from that where you can see that there is motion. The clouds are moving, but everything's just a little bit more subtle. So you can now have that control between subtle motion and crazy rapid motion when you generate videos. With Runway Gen 2. Now, the other feature they rolled out, I'm not even sure they made an announcement about it. I was just kind of clicking it around, and I noticed that there was this option for 3D here. If you click into 3D, they actually have a 3D capture option and a 3D texture option. And the 3D capture option looks like you can now do nerfs directly inside of Runway. So the way this would work would be you take a video with your phone of an object that you want to turn into a 3D object, and you sort of walk around that object with your phone, and you shoot a video from one minute to three minutes of that object from different angles and different perspectives. You upload them into Runway ML and it will turn that into a 3D object for you. For example, here's a camera that somebody uploaded, and you can actually move around the camera in 3D. Now, every time you move around it, it kind of takes a second to re render from the new angle. But then you have a little editor here where you can change the camera path, the depth of field, the focal length, and then set keyframes to actually generate an animation of the object here. This isn't something that I spent a lot of time Playing around with it's something that I literally just came across today as I was exploring the new motion feature inside of Runway. So I imagine this is something that they'll be announcing and talking about more. But I'm excited to play with it because I love playing around with nerfs. And finally, to wrap up the AI news for this week, an AI powered drone actually beat a human. Champion pilot. Now, we've had AIs beat people in games like go and chess, but in this case, we've actually seen an AI take control of a drone and race a human and the drone actually won. And this is not speed up footage. This is how quickly these things actually move. And that wasn't something where it was like pre trained. It didn't know the path that it was gonna fly. It was using onboard sensors to know where to go next and then fly through the path that it was supposed to fly through. So I thought that was pretty cool. And I really love drones, so I wanted to share that with you. And to wrap up, if you've made it this far and maybe you watched my video about my studio tour, you can see that the overhaul of my studio has begun. I painted my wall behind me. I got rid of the old dresser that was behind me and actually got a cabinet that is designed for being an office storage type cabinet. And with each video that I post, you're probably going to see a little bit more get built out in my background. And right now, in our future Tools Discord, we're running a competition. I'm actually looking for a Matt Wolf logo because I want to make a sign of whatever the competition winner is and put it as a lit up like neon sign behind me on my wall. So there's some awesome entries that are coming in right now. But if you're not inside the Discord, definitely jump in. Check out the announcement section where we talk about our that we're doing for AI art, because the one we're running right now is for a Matt Wolf brand logo. And if you win that competition, not only is there cash prizes, but whatever logo you generate will be in the background of my videos for the future. So hopefully you learned something new from this video. Hopefully. I pointed out some AI news that maybe you hadn't heard yet. That's always my goal is to break down all the AI news and do it as quickly as possible. So you really only need one video a week to get all caught up and know everything that's happening in this world. But if you want to stay up to date on a more frequent basis. Make sure you check out Future Tool I actually curate the news on a daily basis here, so all of the news that I'm sharing in these videos, I'm publishing them to the website as they come out. I'm also curating all of the coolest AI tools that I'm coming across and testing out and playing with over on the main FutureTools IE website. And if you want to just get some updates from me once or twice a week via email, join the free newsletter Right now I just send one email a week with just a handful of news items, a handful of tools, and just the TLDR of the week. But I'm about to up it to two emails a week. So pretty soon you'll be getting two emails in your inbox if you're on that newsletter to sort of break it up a little bit more so the emails aren't so long. You're gonna get two shorter emails per week. So the TLDR is gonna be even more TL Dr. Or something. Anyway, thank you so much for tuning in. If you like videos like this and you wanna stay in the loop with all the latest AI news and happenings and coolest tools and tutorials and all that kind of stuff like this video and maybe consider subscribing to this channel and I'll do my best subscribe to to keep you in the loop. Once again, I really, really appreciate you. Thank you so much for tuning in. See you in the next video. Bye bye.