Massive Companies Unite to Prevent AI Takeover

It's been an interesting week in the AI world and I don't want to waste your time, so let's just get right to it. Starting with the fact that Time magazine released their best inventions of 2023, and in it they had a whole section on AI and among the AI tools they listed 12 different products. Adobe's Generative Fill, GPT4, Runway Gen 2 feature, Print BE drone, Meta, Seamless, M4T, Soviets, SVC Alert, California Cal Fire AI Wildfire detector, Stable Audio, Trailguard AI Dall E3, Project Gutenberg AudioShake and Humane AI Pin. While I agree with most of these that they should be on there, there's definitely some snubbed products like maybe Midjourney version 5. I mean, they listed Dall E3, but Midjourney's way better at making realistic images and it's a lot less censored than Dalle 3. They also included Stable Audio, which to me is weird because I didn't find the outputs of Stable Audio that great. But then you have tools like Suno, which did a really good job of making AI music, and Google's Music LM is also much better than Stable Audio's output. Sovit's SVC is definitely good. If you've heard like the AI Drake song or the AI Kanye songs or any of those songs that went viral, they most likely used Soviets SVC behind the scenes to actually get that singing voice to sound like that person. However, it's a very convoluted, difficult process, so that's also surprising. And a biggest shock on here is the Humane AI pin, which yes, we got a preview of it at a TED Talk and there was also some like Runway show where people were wearing it, but most of the world hasn't seen it yet. And if you really want to know how biased Time is, check this out. Let's click into the Humane AI Pin. Interesting investors in Humane include Time co chairs and owners Mark and Lynn Benioff. Hat tip to Nick St Pierre on Twitt for actually pointing that out to me. Also, the biggest clear omission, obviously, Future Tools website. I mean, come on. Also this week Midjourney rolled out the beta version of their new long awaited website. Now, while we can't actually generate images on the new website yet, Midjourney claims that feature should be available probably next month. But Midjourney does sometimes have a habit of underestimating when something's actually going to go live. However, the website they do have is actually really cool. The search functionality up here is actually Lightning fast. If I click Search Halloween and hit Enter, look how quickly it pulled up images that other people have generated related to Halloween. That was almost instant. You might think I sped up the video, but I didn't. If I look into my images, you can see I have a zoomed out view of all of the images I've generated, so I can quickly hone in on the exact image I'm looking for. And this search works just as well in this area. So if I type robot, you can see all of the images I created that had the word robot somewhere in the prompt. You can also click on a specific image and press the numbers 1 through 4 to rate it, 4 being the highest rating. So I press 4, you can see down in the bottom right. It just gave it the hearty eye emoji here. And when I go back to my search, I can actually filter it down to just the ones that I rated a 4 and you can see all of the images that I've given that high rating to. So it's another way to sort and filter all of the images that you've generated. But quite honestly, we're all just waiting for that generation inside of the website so we can get the heck out of discord. This week if you use Shutterstock, they actually integrated a feature that allows you to edit images using AI, including Magic Brush, creating variations, expanding images, Smart resizing, background removing, and an AI image generator which was actually launched earlier, but now you can hone in and get the exact image you're looking for. In fact, on their website it says craft picture perfect content with a palette of creative AI features to design and edit any Shutterstock image generated or stock. So if you find a stock image that you really like, but it's missing just one little touch, you can do some sort of generative fill in it, or remove the background or remove an element of it. Although if I'm totally honest, I really don't understand why people wouldn't just go generate the exact image they're looking for and then use Generative Fill or one of the existing inpainting tools to get the exact image they're looking for. Amazon's getting into the AI image generation game as well. They rolled out an AI powered image generation tool to help advertisers deliver a better ad experience for customers. So Amazon sellers that are advertising their products can now take a generic image and have AI apply a specific background to it to make the ads for their products stand out a little bit better and draw people's attention to them. But as AI images become more prolific, companies like Google are trying to help people better detect what's AI and what's not. This week, Google announced tools to help users fact check images. Google is going to allow people to view an image's history, metadata and the context users used it with on different sites. So Google announced in about this image feature so that users can understand when the image was first seen by Google's search. It's also going to help people understand how people describe the image on the sites to help debunk any false claims. So it's going to look at sites where these images appear and give contextual information about what those websites said about the image. Also, if an image was generated with one of Google's AI generators, the company said that it marks all images created by Google AI as well. So if I go to Google Images, I search for an image of a wolf here, let's click on this one. If I click these three dots up here, we've got this about this image button. And this image is at least five years old, so we know it's probably not AI generated. And we can see the web results of where this image appeared to get better context around what was actually said about this image. But of course, artists still want to protect themselves from their images being trained into AI models. So also this week we saw a new tool rolled out called Nightshade. This is a tool that messes up training data in ways that can cause serious damage to image generating AI models. A new tool lets artists add invisible changes to the pixels in their art before they upload it online, so that if it's scraped into an AI training set, it can cause the resulting model to break in chaotic and unpredictable ways. Here's some examples of what a clean model looks like and with poisoned samples within that model. So here's what a dog looks like with stable diffusion XL. Here's what it will start to generate. If there's 50 poison samples in there, 100 poison samples, 300 poison samples, and you can see how it pretty much destroys what these images look like and starts to shift them to completely wrong images. However, stuff like this I think is also going to be a perpetual cat and mouse game. I think the people that are training the models are most likely going to figure out ways to circumvent the poisoning of the models as well. But I also believe that there's a constant push towards opt in only. So future models will will likely be more and more just trained on images that permission was given for those images to be trained on. So in my opinion, the need for tools like this nightshade are going to be less and less needed as time goes on. This week, OpenAI, Anthropic, Google and Microsoft all gave an update about their model forum. They appointed their first Executive Director, Chris Mezzaroli. Now, we talked about the forum in a previous video, but just to sort of catch you up, the mission of the forum is to advance AI safety research, promote responsible develop development of frontier models and minimize potential risks identify safety best practices for frontier models share knowledge with policymakers, academics, civil society and others to advance responsible AI development and support efforts to leverage AI to address society's biggest challenges. They also announced a $10 million AI safety fund. The primary focus of the fund will be supporting the development of new model evaluations and techniques for red teaming AI models to help develop and test evaluation techniques for potentially dangerous capability of frontier systems. Basically, they're funding a team outside of the actual companies themselves that are funding it to find ways to make these tools more secure, more private, and also to try to break the tools and find vulnerabilities and bugs and issues with the tools. OpenAI also announced an internal team this week that they're calling their Preparedness Team. This Preparedness Team essentially has the same role as the Forum Team. Their goal is to manage the catastrophic risks from frontier AI by answering questions like how dangerous are frontier AI systems when put to misuse both now and in the future? How can we build a robust framework for monitoring, evaluation, prediction and protection against the dangerous capabilities of frontier AI systems? And if our frontier AI model weights were stolen, how might malicious actors choose to leverage them? They also launched their $25,000 AI preparedness challenge. They're looking for 10 submissions of novel ideas to prevent catastrophic and if you end up helping them, you could end up getting a piece of that $25,000 in API credits. And if OpenAI is doing it, Google might as well too. So acting on our commitment to safe and secure AI so this week Google announced their Bug Bounty program specific to generative AI. They released some criteria over on the Google Security blog of what sort of things are in scope and out of scope to potentially receive the bounties for helping them find bugs in their AI systems. Also this week the British Prime Minister announced that he would establish a new artificial Intellig Safety Institute as well as seeking buy in for a new global expert panel on the emerging technology. Next week the UK is having their Global AI Summit where they intend to set up a body that will investigate AI risks and share those with global partners. They also claim that they will make the work of their AI Safety Institute available to the world. Next up, I want to tell you about the sponsor of today's video, HubSpot. They just put out a new guide all about ChatGPT. It's called the Ultimate Guide for Using ChatGPT at Work and it's available to download for free. As we all know, AI is rapidly changing the way we work, and of course Chat GPT is at the front of this revolution. It can help you write everything from blog posts to emails to YouTube scripts and ad copy. And it's incredibly easy to use if you know the right way to do it. So here's a few ways that HubSpot highlights. Using ChatGPT at work you can generate high quality content quickly and easily because ChatGPT can basically write at a human level, but it's not a human, so it does it a lot faster. So you get to spend less time on writing and more time on other important tasks. And ChatGPT helps you save time and energy by helping you automate repetitive tasks, things like research, fact checking, and even helping you come up with ideas for images within your writing, which again, frees up your time for the work that's actually important to you. So if you're looking for a way to boost your productivity and improve the way you get work done, I highly Recommend Checking out HubSpot's guide for ChatGPT. Again, ChatGPT can help you get a lot more more done in a lot less time if you know how to use it correctly. And HubSpot wants to help you demystify how to use this powerful tool. So if you don't know where to start, check out this guide to learn everything you need to know. To learn more about this free resource and how it can help you with your work, visit the link in the description below. And thank you once again to HubSpot for sponsoring this video. Also this week, Amazon is bringing conversational AI to kids with the launch of Explore with Alexa. This is actually an Alexa conversational experience designed for kids under the age of 13. However, it still can be led astray and hallucinate answers. However, Amazon claims they've put some guardrails in place to address the potential problems. If you remember from a few months ago, Yannick Kilcher and a entire community of people launched Open Assistant and it was designed to be a ChatGPT alternative that was completely open source and essentially completely built and trained by the people that were using it. Well this week they decided that they are done working on it now. It is still open source, it is still openly available. People can go and pick up where they left off and continue working on it. But the initial team that was building it said that essentially they are happy with where they're leaving off on it and their work here is done, essentially. They also went on to say that with all of the major companies getting involved, the Googles, the Metas, the Microsoft's, the Amazons of the world, all getting involved in building large language models, they just didn't feel like they were going to be able to build something that was going to be at that level with the resources they had available to them. And finally, in the last bit of large language model news for the week, there was a little bit of a leak around Gemini and something called Stubbs. Now in this medium post Bedros, that last name essentially leaked screenshots of what Gemini and Stubbs will look like. Now we kind of already know that Google's Palm 2 is going to be replaced by Gemini. And if you're wondering, well what is this stub that he's talking about? It's a feature where you get to build and launch your own AI generated app directly from Maker suite. He says that translation between languages will be fully supported in maker suite and Google will provide a sample prompt for translating between Spanish and English when it's out. So keep an eye out for that. Now from the leaked screenshots he claims that this new stubs tool will be like AI generated figma prototypes. It won't create full code, but rather working AI generated app prototypes. You can see in this screenshot here there's a text box and within the text box they wrote make an app that provides pet care instructions. He also claims that you'll be able to remix stubs and have your own twist on an idea. He then showed off some leaked screenshots from Gemini here, which shows that it will likely be multimodal. It looks like you'll be able to upload images and ask questions about the images and I'll link to this blog post in the description along with everything else I've talked about in this video today, so you can read more about the details yourself. Seeing how this is some leaked screenshots and there's been no confirmation from anybody other than Bedros here. Still, take this information with a grain of salt. I mean it feels pretty legitimate, but you just never know. And speaking of leaks, this week there was a leak of the new Snapdragon 8 gen 3 now the Snapdragon is the mobile chip that's in many mobile phones these days. Here it says the Snapdragon 8 Gen 3, its debut in mobile platform designed specifically for generative AI. Now again, since this was a leak and not something that was actually confirmed by one of the companies involved, take this with a grain of salt, but I mean it's pretty fair to assume that all next generation chips are going to be designed with AI in mind. So I'd say we can barely confidently assume that this information is correct. One thing for sure is that Snapdragon 8 gen 3 will definitely challenge iPhone 15's ray tracing capability. It's a console defying and premium level gen 3 with a ray tracing feature brought to you by Unreal Engine5.2 and can upscale cutscenes up to 8K. Now this article is full of technical jargon that quite honestly is over my head, but here's the full leaked specs here if you want to pause and take a look. Quite honestly, I don't personally have a frame of reference other than what this article tells me that it's going to give iPhone 15 a run for its money. It sounds like YouTube might be trying to make an official way to create AI Drake fakes. And I guess not just AI Drake. They want to get the rights to use musician songs to train a new AI powered voice replicating tool. YouTube is currently developing an AI powered tool that allows users to replicate the voice of famous musicians while recording audio. According to a new report by Bloomberg, no deals have yet been signed by any major record label, but Bloomberg sources claim that the discussion between parties are currently ongoing. Not a ton is known about this yet, but it sounds like YouTube wants to find a way to work with the music labels to allow people to use AI generated songs, but I would assume also fairly compensate the artists. And speaking of YouTube, YouTube Music now lets you make your own playlist art with AI. Previously, YouTube didn't even let you change the COVID art of your playlist at all. And still it looks like you won't get to upload your own image, but you can now make a custom image using generative AI for your playlist. Nothing mind blowing, but since I was on the topic of YouTube, figured I might as well mention it. Late last week Satya Nadella from Microsoft wrote his annual letter to the shareholders and well, it's a pretty long document. So I did what any smart person does who wants to save time and I plugged the long document into Claude and asked it to give me the high level bullets of things in the letter that are related to AI and this is what it told Microsoft is entering a new era focused on AI, including launching Copilot and Everyday AI Assistant that will be integrated across Microsoft products. Microsoft introduced AI capabilities across products like GitHub Copilot for developers, Dynamics 365 Copilot for business applications, and Microsoft 365 Copilot for productivity. Microsoft aims to train and certify 10 million people with digital skills, including AI skills by 2025. So far it's helped 8.5 million gain the skills. Microsoft believes AI can help address climate change and launch tools like renewables watch to map clean energy sources. Microsoft acquired Activision Blizzard and is using AI to improve gaming through recommendations and personalization in Xbox Game Pass. Microsoft launched new AI powered search features in Bing, including chat and image generation. And Microsoft is focused on developing AI responsibly with principles, practices and tools to ensure trustworthy and ethical AI, Something that we also talked about at the beginning of this video. So here's something fun. Several months ago we talked about how there's a new Beatles song coming out and Paul McCartney and Ringo Starr worked on it, but they were also able to use vocals from John Lennon using some old recordings that they found combined with AI. Well, we now have a release date of when we can expect this new song. The song is called now and Then and it drops on November 2nd with a music video debuting on November 3rd. So if you're a Beatles fan or an AI fan, that's something to look forward to. This week Forbes launched a new generative AI search tool called Adelaide, powered by Google cl. It's basically an AI powered search engine to help you search the news on Forbes. Also this week this article came out claiming that Google Maps is becoming more like Search thanks to AI. Google Maps is getting an AI makeover, adding new features like Immersive View and making existing features like driving directions easier to follow. The biggest takeaway here is that Google wants Maps to be more like search, a place where people can obviously come get directions or find coffee shops and EV chargers, but also enter vague queries like fall foliage, latte art or things to do in Tokyo video and get a whole bunch of actually useful hits. Google is using neural radiance fields, aka NERFs, which is a form of generative AI to sort through billions of images including aerial photos, street imagery and indoor snaps to create 360 degree experiences. Traffic information is culled from historical data and filtered through Google's predictive algorithm to help users figure out the best time to get somewhere with the least amount of traffic. So to me this sounds like some some pretty cool and pretty useful new updates that we're going to be getting inside of Google Maps pretty soon. In the world of robotics, Austin Dynamics put Chat GPT inside one of their robots and they're using it as a robot tour guide. The robot could walk around, look at objects in the environment, use VQA or captioning model to describe them, and then elaborate on those descriptions using a large language model. Additionally, the large language model could answer questions from the tour audience and plan what actions the robot should take to next. This article then goes on to share a lot of the technical details about how this actually works. And it's even using 11 labs to do the text to speech, meaning that it probably has a pretty realistic sounding voice. I found this to be interesting as well. The Waymo driverless vehicles are now available through Uber and it's starting in Phoenix. So you can actually open up Uber and order yourself an autonomous Weibo vehicle and a driverless car will show up, pick you up and take you to your destination. If you just want to use the Waymo app straight up, you can actually use it in Phoenix, San Francisco, Los Angeles, and pretty soon Austin, Texas. This week Bill Gates did an interview with a German newspaper. During his interview, he said that he thinks that the generative AI is kind of plateauing. He said that we shouldn't have very high expectations of GPT5. He doesn't believe it's going to be much of a leap over GPT4 and that we've seen most of the significant leaps already that we're going to see, at least for a while. Personally, I don't actually believe that. I think we haven't seen anything yet. We have a little bit of multimodal capability with things like Chat GPT. We can now have it generate images and we can now upload images. But I imagine in future versions we're going to be able to upload audio. We're going to be able to have it give us audio back that we can download. We're going to be able to upload video and have it actually generate videos for us. Right now we can't even connect GPT4 Vision with Dall E3, which seems like a no brainer that they would be able to go back and forth and iterate off of each other pretty soon. So I still think we have quite a bit more advancing to do and I can even comprehend what these advancements are and I don't know what I don't know. So there's probably advancements coming from these things that I'm not even comprehending yet. On this one, I kind of gotta disagree with Bill Gates, but I also don't necessarily know what he knows. So who knows? That's a lot of no's. And Finally, Meta chief AI scientist, says AI won't threaten humans Yann LeCun, the chief AI scientist at Meta, said that labeling the existential risk of AI is premature and it's preposterous that AI might kill off humanity. You're not familiar with Yann Lecun. He is one of the godfathers of AI. He won the touring award with Geoffrey Hinton and Yoshuo Benjo. He is one of the people that has been working on this stuff from the early days and, well, he doesn't necessarily see eye to eye with Yoshua and Geoffrey Hinton anymore because he believes that all of this existential worry that AI is going to kill off humanity is overblown. While Yoshua and Hinton are sort of on the opposite end of that spectrum where they're much more cautious and thinking that we should be a little bit more worried than we actually are. He says the debate on existential risk is very premature until we have a design for a system that can even rival a cat in terms of learning, learning capabilities, which we don't have at the moment. He sees AI as more of a helper and manager of our everyday lives. He says everyone's interaction with the digital world will be mediated by AI systems. Personally, I fall somewhere in between. I have a healthy amount of fear and some concern over where AI is headed, but I also tend to lean fairly optimistic that we're going to figure this out as we go and well, I actually have some upcoming videos on that exact topic that I'm really excited to share in the near future. I'm really trying to put some polish on them and get my thoughts out clearly on those topics. So look for those on the YouTube channel soon and if you haven't already, click the button to subscribe to this channel to make sure that that shows up in your YouTube feed. That's all the AI news I got for you for this week and I'm recording this on Thursday. There might even be new news on Friday that I totally missed in this video and if I did, I'll make sure it's in next week's video video. Thank you so much for tuning in. If you enjoyed this and you want to see more news like this as well as tutorials and some upcoming video essays on the world of AI. Again, consider subscribing to this channel and like this video. That'll make sure more videos like this show up in your YouTube feed. And don't forget to check out FutureTools IE where I curate all of the coolest AI tools that I come across as well as all of the AI news on a daily basis. I've also got a free newsletter where I will send you all the cool tools and latest AI news directly to your email inbox. You can find it all over at FutureTools IO. Thank you once again for tuning in. Thank you once again to HubSpot for sponsoring this video. You really make it so that I get to live my dream and make fun YouTube videos and nerd out about AI all day long. So thank you again to HubSpot. You guys are rock stars and I appreciate you and I appreciate you watching this video because it is so much fun to nerd out with you. I am so appreciative of every single person that watches my videos, even the ones that make the negative comments. I still appreciate you taking the time to watch the video, so thank you so much. You rock. I really appreciate you. See you in the next video. Bye bye.