AI News: What You Might Have Missed!

So here's all the AI news that you might have missed this week. This week, there was actually two separate giant conferences going on at the same time. There was the annual siggraph conference, which is a giant event that covers all the latest research, all the latest advancements in computer graphics, all the research papers that I'm constantly sharing with you. A lot of the people that wrote those research papers are there actually presenting them and basically explaining what they mean and showing them off if there's actually a demo to show. And this year, there was actually a lot about AI at the event. Also this week was the AI4 conference conference in Las Vegas, which was purely focused on AI. So whoever made the decision to make AI for the same week as siggraph, I don't quite understand the decision. I was invited to go to both events, but had to choose. And I ended up going to Siggraph because I was super interested in hearing about the wide range of topics that they talk about @Siggraph. Although I love AI and I report on AI and I break down the news every week on AI, I also really love computer graphics and visual effects and virtual reality and augmented reality and mixed reality and all of the various emerging technologies. And siggraph kind of covers all of those because it had a wider range of things that I was interested in hearing about. I decided to attend SIGGRAPH over AI for this year. Now I'm going to do an entire breakdown of my experience of Siggraph in a future video. There was, however, a handful of new announcements that came out of this Siggraph event. Most of them came during the keynote presentation by Jensen Huang, the CEO of Nvidia. Siggraph was an amazing event, and after going to this one, I plan on attending every year I possibly can because I was just blown away by some of the stuff they talked about. And if you want to dive deeper into my experience at Siggraph, make sure you subscribe to this channel, and I'll make a video specifically around all of the cool stuff that came out of Siggraph. Speaking of Jensen Huang's keynote, I was in the third row for this event, so I was very, very close to the action. And I recorded the entire presentation that Jensen gave on my iPhone so that I can report on it later. It's funny because I came across this article on Nvidia's website about Jensen's keynote. And when I scroll down to this image right here, it's kind of hard to tell, but this is the back of my head right here.

The generative AI era is upon us. The iPhone moment of AI, if you will, where all of the technologies of artificial intelligence came together in such a way that it is now possible for us to enjoy AI in so many different applications. The journey of Nvidia accelerated computing met the journey of the deep learning researchers and the big bang of modern AI happened.

So after Jensen shared that this was the iPhone moment of AI, he then unleashed announcement after announcement after announcement of things happening at Nvidia. Here's a super, super quick summary of some of the announcements that they made. They unveiled the next generation of their GH200 Grace Hopper superchip. These superchips were created to handle the world's most complex generative AI workloads, and they deliver up to 3.5 times more memory capacity and 3 times more bandwidth than the current generation offerings. They also announced Nvidia AI Workbench. This is a simple tool set that makes it really easy for you to pull in models from places like GitHub or Hugging Face. Train your own data into these models and it allows you to do it locally, or you can transfer it to Nvidia's cloud and also run them in the cloud. It's accessed through a simplified interface running on a local system. It allows developers to customize models from popular repositories like hugging face, GitHub and Nvidia NGC using custom data. The models can then be shared easily across multiple platforms, so you can pull in code from places like GitHub or Hugging Face, run them locally, train your own data on them locally, and when you're ready to scale up easily, move what, whatever you trained over to the cloud so that you can scale it up and allow other people to run it as well. They also announced a partnership with Hugging Face, where Hugging Face developers will have access to Nvidia's DGX cloud along with all of their AI supercomputers all within the Hugging Face platform. With DGX Cloud, Hugging Face users will be able to connect to Nvidia's AI supercomputing, providing the software and infrastructure needed to rapidly train and tune foundational models with unique data to drive a new wave of enterprise LLM development. So basically we're going to have access to much more powerful supercomputers on Hugging Face to be able to run some of these models, which is honestly something I'm really excited about. They also talked about their new Omniverse Cloud APIs to help developers adopt Open USD. Now Omniverse is sort of their like world building 3D engine, sort of like an Unreal Engine kind of thing. Now, Omniverse isn't new, I just want to give you some context here, but it allows you to create 3D worlds and 3D graphics. And they are now making APIs available for it to try to get people to adopt the open USD format, which is attempting to be a universal 3D graphic file type. So if you make a 3D world or 3D graphic in Unreal Engine, you can pull it into Blender. If you make something in Blender, you can pull it into Maya. If you make it in Maya, you can pull it into Omniverse. They're trying to get this open USD file format to catch on and have this universal 3D format that makes it so it doesn't matter what platform you're developing on these 3D graphics, work everywhere that you can use 3D graphics. That's the goal. They also created a large language model called ChatusD, which is designed to be a chat model to help developers better understand and better use this open USD format. So lots of big announcements from that Nvidia keynote. I'll probably touch on some of them in a little bit more depth when I make my video specifically around my experience from Siggraph. Now the other event that was happening this week was called AI4, and again because I was at Siggraph, I wasn't at AI4, but I did find this nice. Quick break down from Jeremiah here on Twitter about the AI trends and issues that were discussed at the event in Vegas. He says that responsible AI was a keynote for MasterCard. They shared a process with a focus on outcomes. Actors are going to be in further trouble as consumer level AI for movies is on the rise. Which, you know, personal opinion. It doesn't feel like the greatest time for actors and writers to be striking. Corporations are all organizing their businesses for AI. The culture of the company is reflected in its data structure, which changes how the AI will evolve. A fragmented company will have a fragmented AI. Corporate business units are often it, but also operations, analytics with a BU and marketers. And then he also made a comment about how he couldn't help but notice that 9 out of 10 of the AI leaders here are men. We need to involve more women. It's very dangerous if the AI is built and doesn't reflect the needs of the population, which I totally agree. I didn't actually find that same statement to be true at Siggraph. I found there was a pretty good mix. It definitely leaned a little bit more male at Siggraph but it was a fairly decent balance. I was actually quite surprised that it didn't feel overwhelmingly male to female in both the attendees and the speakers and presenters. Hopefully next year AI4 isn't the exact same week as Siggraph, and I will try to get out to it and take more notes on my own. But when it comes to AI, it sounds like a lot of the same discussions that were happening around AI at AI 4 were a lot of the same discussions that were happening at Siggraph. In fact, at Siggraph, I sat in on a panel all about AI and the ethics of AI and the ethics of AI going into art. And as you can see by this photo I took, it was standing room only. It was a hot topic. It was much too small of a room for this topic because everybody wanted to weigh in on their thoughts on the ethics of where all of this AI stuff is headed. This week, Midjourney also held their weekly office hours. And again, I was at an event, wasn't able to make the office hours. But luckily we've got people like Ali Jules here who makes amazing breakdowns of these Office Hour calls. So if you're not following her, make sure you follow her because she'll keep you in the loop with all these Mid Journey office hour calls calls. But here's what came out from this week's Office hours. They mentioned that version 5.3 might be released next week with inpainting and Aesthetic in Discord. Now, I'm not quite sure how they're going to do the inpainting. I mean, when we use tools like Zoom out and Pan, that's a form of inpainting. But is this going to be a different type of inpainting than what we've got there? Are we going to be able to inpaint just certain sections? I don't know how that's going to work in the Discord user interface, so that'll be something interesting to see. And then it also sounds like maybe the inpainting will only be available on the desktop version. And then version six, which is interesting because I was on an Office Hours call about three months ago and they said version six was about a month off. Well, it sounds like they were a little underestimating that. Apparently they're still working on it. It will be more coherent with better language understanding. And they still don't have an estimated release date yet. Now, probably the most requested feature from Midjourney is Get us the heck out of Discord. When can we have A website to generate these images. Well, it sounds like a lot of people are working on the front end website right now. They're going to have both a desktop and a mobile version of it. The launch date is getting pretty close, but it sounds like they're not willing to commit to a date yet. The ability to explore will be much better and the creation page will be much better. And it sounds like they are still working on video, but they don't have a timeline for it. And they're still saying that there's not going to be an API anytime soon for other tools to be able to integrate Midjourney. Once again, thanks to alijules for these updates. I'm still a big fan of Mid Journey. I just can't wait till they get it out of Discord and give us a proper user interface for it. Now, speaking of AI art, let's talk about Wirestock, who is actually the sponsor for today's video. If you're not familiar with Wirestock, it is a tool where you can upload a single image. Like say for instance this heart image that I generated here. It will actually create all of the tags for you for the image, create a title for the image, create a category and a description for the image, and then it will upload it to all of the stock photo websites for you so that people can purchase those images off the stock photo website. One place to upload the image, but the image gets distributed across all of the stock photo websites. And they recently just rolled out a Discord bot. So if you've been using Midjourney, you're already kind of familiar with using Discord bots and it's really simple to use. You simply install the bot by authorizing it, adding it to your own private Discord server. I'll add it to my AI Art test server where I've already added my Mid Journey bot and then click continue. And now you can see I've got this bot installed over here in my Discord server where I create all of my AI images. Here's the most recent AI image that I just created. It's not yet upscaled, but if I type slash, select the Wirestock bot. Here you can see we've got a few options here. I'm going to go ahead and log in to make sure it's connected to my Wirestock account. If I go to my Wirestock settings, I can see my Discord token. I can go ahead and paste this token in and it will connect it to my account. I'm going to go ahead and do that real quick. But I'm going to do it off camera so you can't see my token. All right, so I've added my token and now I have successfully been logged in. Now if I come up to this most recent image that I generated up here, I could come to these three little dots. Come down to apps and you can see it added this new upscale option with the Wirestock logo. If I click on upscale, it now says upscaling is in progress. It normally takes less than 10 seconds. Now it took a bit more than 10 seconds. But once it did upscale I now have these options to publish on Wirestock or participate in a challenge. Now if I click Publish to Wirestock, it will automatically publish it on the website and make it available for sale on all of the stock photo sites that allow AI images. Sites like Adobe Stock 123RF, FreePik, Dreamtime, Imageo and others. Wirestock also does these really cool 24 hour challenges where they give you a theme, you can generate an image with AI, submit your image to the competition and win prizes. You could learn more about Wirestock by going to futuretools IO Wirestock and if you do decide to upgrade, use the coupon code Matt that's M A T T to get a 20% discount on your premium membership. Thanks once again to Wirestock Stock for sponsoring this video. In other AI art news this week, one of my favorite AI art generation tools, Leonardo AI just released their iOS app. The iOS app gives you access to Leonardo Alchemy, their various fine tuned models, image to image, upscaling, unzooming, removing, background and more. And you can find it in the App Store right now. I personally haven't played around with the mobile app yet, but I am about to download it and give it a go. So this is really interesting because Zoom recently updated their Terms of Service with this statement. You consent to Zoom's access, use, collection, creation, modification, distribution, processing, sharing, maintenance and storage of service generated data for any purpose to the extent and in the manner permitted under applicable law, including for the purpose of and there was a whole bunch of purposes listed, but also among them was machine learning or artificial intelligence. Now what I find interesting about this is these Terms of Services were actually updated on March 31, 2023 almost five months ago. However, there was a new update to the Terms of service on July 27th and when these new terms were updated, people started to notice the terms that were updated back in March stating that we can use your data from Zoom for anything we want. It sort of Caused a bit of an uproar. In fact, it was one of the big talks at Siggraph. People were passing this information around. However, after this uproar, Zoom's Chief Product Officer made a statement saying, it's important to us at Zoom to empower, empower our customers with innovative and secure communication solutions. We've updated our terms of service to further confirm that we will not use audio, video or chat customer content to train our artificial models without your consent. So it sounds like they added this term into their terms and conditions as just sort of a blanket statement saying we should be able to use this data any way we want in the future. Not really thinking of the implications, but now that all of this AI news and some AI outrage is kind of bubbling up, people notice these terms brought it to people's attention and now this has become an. And Zoom's having to speak out saying, no, no, no, we're not going to use your chat conversations for training anything. We just want to use data that we collect to improve the services at Zoom and not to train our own large language model on your data. They basically wanted to use the generative AI for good, as they say, for instance, with tools that summarize meetings if users request it, which, you know, is a pretty handy use of Zoom. If I get on a Zoom call with you, we have an hour long conversation and Zoom can send me back a description of the conversation that we had with bullet points. That's helpful to me. But if Zoom takes that data and uses it to train their own large language models, now there's potentially proprietary information trained into Zoom. They are telling us that they will not do that without our explicit permission. It's just very interesting how something from March all kind of bubbled up now in August. Speaking of Zoom, here's a little tip. If you're ever on Zoom and you need to enter your password into a website somewhere, make sure you mute your microphone. There's a new AI out there that can identify passwords by sounds of keys being pressed. They specifically even mention Zoom. Tapping in a computer password while chatting over Zoom could open the door to a cyber attack. A study revealed artificial intelligence can work out which keys are being pressed by eavesdropping on the sound of the typing. And it can do it with more than 90% accuracy. So again, if you are going to enter a password while on a call with somebody, mute your microphone. Also in AI news this week, Google now has an AI powered grammar checker. It's real simple to use. If you go to Google and you type a sentence like where am I going? And then type grammar check and do a search. You'll see it actually fixed my grammar here and it allows me to copy the fixed grammar. The tool itself is powered by artificial intelligence, although it might not always be accurate, especially if you ask it to review a phrase or fragment instead of an entire sentence. These tools have been in Gmail and Google Drive like Google Docs for a while now, but now you've got it in search as well. Week ChatGPT also expanded the Custom Instructions feature and made it available for free users. So if you go to ChatGPT, come down to these three dots in the bottom left corner here. Click on Custom Instructions and you can add some additional context that ChatGPT will remember every time you write prompts in the future. ChatGPT also finally rolled out these suggestion boxes so as you are writing prompts it will actually give you suggestions of potential next prompts that you can use. In other OpenAI news OpenAI launched a We Web crawler called GPT Bot and basically what this is is a web crawler designed to crawl all the websites on the Internet to gather information from them. Now if you've used Chat GPT already, you have probably encountered the fact that it is only trained with data through 2021. Well, it sounds like GPT5. They want to make it constantly updating and constantly training on new data. In order to do this that it needs to create little bots that go around and scrape the web and constantly bring in new information. It's the same way Google Search Engine works. Google Search Engine has little bots that go and scrape the web. Whenever websites add new information, it finds that information, pulls it back into Google's database and makes it searchable for you. Well, GPT5 wants to have that same functionality as a way to constantly update GPT5 with new information. They are however allowing websites to opt out. So if you do not not want your website to be scraped by GPT5, you don't want any of the information that you share to be included in the training. In the future there will be a command that you can put inside of your website to disallow the GPT bot from scraping that information and following along on that same thread there. The New York Times just updated their terms of Service to prevent AI from scraping its content. The content includes but is not limited to text, photographs, images, illustrations, designs, audio quality, clips, video clips, look and feel, and metadata, including the party credited as the provider of the content. This updated terms of Service also prohibits website crawlers which let pages get indexed for search results from using content to train LLMs or AI systems. This is interesting to me and I'm going to inject a little bit of an opinion here. Sites like the New York Times benefit heavily from web crawlers scraping their site, grabbing their news and then putting them on sites like Google, Bing, Yahoo, all of the various search engines. Like if Google didn't rank their content anymore, it would be absolutely disastrous for their company. Yet if Google wants to scrape their content for creating a better user interface and a better user experience through AI, they're saying no, you can't do that. If Google does move more and more towards an AI based model where it is more chat based, similar to what we're seeing with Bing chat companies like the New York Times who are saying you can't use our data in AI anymore, are they just going to be excluded from the results even when people are just searching? I don't know. Some of these decisions from companies like the New York Times in my opinion are very short sighted because this could be the future of search. This could be the way that people find their information in the future. And if you're saying we're opting out of it, Google might say, fine, we're not showing your stuff anymore. And I don't really know how many people are still buying New York Times and Print, but I have a feeling it's going down year over year. You really want to eliminate the Internet traffic you're getting as well? I don't know. Just throwing some thoughts out there around this. Also this week Anthropic launched an improved version of its entry level LLM. So Claude 2 basically this week got some fairly marginal improvements. It saw gains in areas such as math, coding, reasoning and safety in internal testing. The new version of Claude scored 58.7% on a coding benchmark compared to the previous version which scored a 52.8% and it scored 86.7 on a set of math questions, over 80.9% on the previous version. So in my opinion, fairly marginal improvements to Claude, but it just shows that it's getting better and better. Stability AI just announced a new large language model called StableCode which is a model that is fine tuned towards coding. It appears to be hosted on Hugging Face right now. So if you want to go test stable code, it does appear to be available for both use in the cloud and the files are available if you choose to install it later locally. Stability AI also this week announced Japanese Stable LM which is the best performing openly available language model created for Japanese speakers. This one is also available over on Hugging Face. If you come to the Stability AI blog, click on the name of the language model here. You can open it and use it directly on Hugging Face and download the files and install it locally if you know what you're doing. The company eleven Labs made a couple big deals this week. First they partnered up with did, the company where you can upload still images, add some text and it will make it look like the people are speaking. We've seen all sorts of memes from the world of Harry Potter, Star wars and of course Wes Anderson films, all using DID to generate those images. Well now it is going to directly integrate with 11 labs so you can even use your own voice or voices that you trained inside of the did models. ElevenLabs also partnered with InWorld, which I talked about in last Friday's news video. InWorld is a company that is creating chat based NPCs for video games. So you can walk up to a character in a video game, have a conversation with that character, and all of the audio that the character generates back to you is generated in real time using AI. The voices have left a little bit to be desired so far, but now because 11 Labs is working within world, we're going to get much, much, much more realistic voices from those IN game characters. This week it came out that Amazon is reportedly testing generative AI tools for if you remember the video that I did a few weeks ago about Shopify, where it will generate descriptions of the products for you just based on a simple description that you give it, it sounds like Amazon's going to be adding that same functionality for Amazon sellers where they could go in, plug in a very brief description and let Amazon's AI generate a much more fleshed out, better selling description for you. This week the White House also ordered a ban on certain US Tech investments in China. They don't want US investors to invest invest in companies that deal in semiconductors and microelectronics, quantum information technologies and certain artificial intelligence systems. There's an absolutely amazing video from Johnny Harris called We're heading into a New Cold War which really breaks down what is going on here on the larger scale. I highly recommend watching it. I'll link it up in the comments below. But basically the US Is worried about capital and expertise helping China develop technologies that can support its military, modern modernization and undermined US national security. They basically want the US to be the most advanced technological country and they don't Want us to help China get more advanced with their technology again? Just watch the Johnny Harris video. It'll break it down much better than I will. Here's an interesting random article that came out of New Zealand this week. A company called Pack and Save created the Savvy Mealbot, which is supposed to help shoppers do their meal planning. However, it was recommending customer recipes. Chlorine gas poison bread sandwiches and mosquito repellent Roast potatoes. Among some of the other delicious recommendations, an Oreo vegetable stir fry. So I think this bot might need a little bit more training and hopefully nobody actually tried the poison bread sandwich. Now, right before I hit record on this video, I got this email to my inbox from RunwayML that we can now generate longer generations in Gen 2. Now, if you remember when I talked about Gen 2 in the process past, I talked about how you can create a video, take the last frame from the video and then set that as the first frame of your next video, and then start your generation that way. And then you could take the last frame of that video and set it as your first frame of your next video. And you can now generate longer videos by stringing these together. I also mentioned that it would probably be a smart idea for gen 2 to just build that into their system. Well, they did it. Now it says now you can extend generations in Gen 2. After you generate your initial output, just select the Extend button to generate another 4 of your video. You can do it up to a maximum of 16 seconds. So we have now hit a point where with a single image prompt, we can generate 16 second videos from one image prompt. A few days ago we were only able to do four seconds. So if I jump into the Runway app and I click on Image to Video, let's go ahead and upload this image that I was messing with earlier for Wirestock. Toss it in here. Click Generate. Let's see what it does with it. We get this 4 second video here, but you'll notice up at the top I now have a button that says says extend four seconds. Let's go ahead and click that and see if we can now have an 8 second video. And now we have an 8 second version of the video where it just gets even crazier and crazier as the time goes on. So pretty cool. We can now generate up to 16 second videos. And then I don't see why if you wanted to, you couldn't just take that 16 second video, grab the last frame of that, and continue the process to make even longer videos. But it's really cool that Runway just added that into the mix now as well. Now one last thing I want to mention before I wrap this one up is that next week, August 16th and 17th, there's an event in Los Angeles called the Artificial Conference. I'll be hosting and participating in panels at that event and hanging out along with a lot of other really smart people in the AI world. We're going to be talking about AI for creators. We're going to be talking about ethical and legal implications of AI. It's going to be a really fascinating, fascinating and really, really fun event. It's going to be a nerd out fest where a bunch of us who all nerd out around AI get together and hang out and nerd out together. You can find details about it over@ArtificConference.com and there's a tweet that I'll link to below where you can actually win free tickets to the event by just simply liking and sharing the tweet. So I'll make sure that's linked up below. Just wanted to point that out because I am going to be there. I am looking forward to it and I'm hoping some people, people that watch these videos turn up and hang out with us. It should be a really, really fun time. That's all I got for you today. Don't forget to check out FutureTools IE where I curate all the latest, coolest AI tools that I come across. I keep the AI news page up to date on a daily basis and if you just want a breakdown and email of just the TLDR for the week, make sure you join the free newsletter and I will send you the Future Tools weekly newsletter. With just a handful of tools, handful of news articles, few YouTube videos, videos and the just what you need to know from AI space for the week. You can find it over at FutureTools IE and clicking to join the free newsletter. Thanks again for watching. Thank you so much to Wirestock for sponsoring this video. Sponsors like Wirestock make it so that I could spend my full time researching tools and AI news and just finding all the good stuff and sharing it for you. So thanks again to Wirestock for helping me make this dream a reality. Honestly, I know that sounds cheesy, but it really is true. And thank you so much for watching these videos. I really, really apprec appreciate you. Hopefully you found some news from this video that you hadn't heard about yet because that's my goal. See you guys in the next one. Bye Bye.