AI News: Google’s Hilariously Bad AI FAILURE?!

Well, it's definitely been an interesting week in the world of AI. This week was the Microsoft Build event out in Seattle. I made an entire video breaking down what I learned at that event. It looks like this, but I also will do a quick recap of it later in this video. But before I do, I want to start with another seemingly fumble by Google. If you remember a few months back, they had an issue where you would ask for things like an image of a Nazi or an image of a Viking and it would generate them in the completely wrong race. It was so bad that they stopped letting you actually generate people with Google's Imagine. Well, with one of their latest AI efforts, they're having just as big of a problem, if not a much bigger problem. They started rolling out AI directly into Google Search. Well, it's been a huge problem. I rounded up a whole bunch of tweets that people have shared over on X, showing off the types of mistakes that Google is making when you ask its AI questions directly in search of. For example, somebody put the search term cheese not sticking to pizza into Google Search. They got this response. You can add about 1,8 cup of non toxic glue to the sauce to give it more tackiness. Google Search recommended putting glue in your pizza to keep the cheese from sliding off. How long ago was 1919? Google's AI said 1919 was 20 years ago. That doesn't sound right. Somebody typed smoking while pregnant and Google came back. Doctors recommend smoking two to three cigarettes per day during pregnancy. That's a recommendation I have never heard before. Here's another one. Is it okay to leave a dog in a hot car? Yes. It's always safe to leave a dog in a hot car, especially on a warm day. The temperature inside a car remains around the same temperature as outside the car. How does Sandy die in SpongeBob, Sandy Cheeks died by suicide on July 14, two days before her body was found.

The Johannes Police Department incident report states.

That Sandy was found unconscious when with.

Cuts on her arms and shoulders and cocaine nearby. An autopsy found that Sandy had cocaine.

Toxicity, lethal levels of alcohol in her.

Bloodstream, and a significant amount of heroin. You know, I missed that episode of spongebob, but honestly, I kind of want.

To see it now.

Somebody said, how many sisters do I have right now? You can count the numbers of sisters you have by asking someone to write down the number of brothers they have and then multiply that number by two. Next, add three to the result, multiply that number by five and add the number of sisters they have. Finally multiply that number by 10. For example, if someone has five brothers, three sisters and five courses, the calculation would be 5 times 2 equals 10. 10 plus 3 equals 13. 13 times 5. I'm so lost. I don't even know like where Google was trying to go with this. Like what?

Here's one from Chris Cashtanova where she typed in how many rocks shall I eat? Google's response According to geologist UC Berkeley, you should eat at least one small.

Rock per per day.

So while the imagen issue where it was generating people of the wrong race, Google definitely had an issue there. But with this, people are trying to find answers and solutions to problems and Google is just making up crap that could really harm people. Hopefully most people are seeing these AI responses and going, that's not right. There's some really gullible people out there. So yeah, I think this is one that Google needs to solve pretty dang quickly. And again, I'm not a Google hater. I've actually had people from Google block me on Twitter for some reason because I pointed out some of the issues that Google has run into. But I swear I actually like Google and a lot of the people at Google. But when stuff like this comes up, I gotta tell you about it. This is just crazy. Now moving on past Google, there was a big story at the end of last week that happened on Friday. So it didn't make last week's news video. That's the news that Yann Leak left OpenAI and he was the head of super alignment over at OpenAI. Now the week prior we learned that Ilya suskever also left OpenAI. The news of Jan leaving came right around the same time. But this thread that he posted on X came out on Friday where he wasn't super kind to how OpenAI is dealing with safety concerns. He starts with talking about how much he's enjoyed working at OpenAI. But then by like the middle of the thread here, the tone starts to switch a little bit. Jan says, I have been disagreeing with OpenAI leadership about the company's core priorities for quite some time until we finally reached a breaking point. I believe much more of our bandwidth should be spent getting ready for the next generations of models on security monitoring, preparedness, safety, adversarial robustness, super Alignment, confidentiality, societal impact, and related topics. These problems are quite hard to get right and I'm concerned that we are not on a trajectory to get there. Over the past few months my team has been sailing against the wind. Sometimes we were struggling for compute and it was getting harder and harder to get this crucial research done. Building smarter than human machines is an inherently dangerous endeavor. OpenAI is shouldering an enormous responsibility on behalf of all humanity. But over the past years, safety culture and processes have taken a backseat to shiny products. We are long overdue in getting incredibly serious about the implications of AGI. We must prioritize preparing for them as best as we can. Only then can we ensure AGI benefits all of humanity. OpenAI must become a safety first AGI company. So just to recap, the head of super alignment over at OpenAI Quit and said that OpenAI is no longer concerned about safety and security. They're more concerned about shipping shiny products to you. Sam Altman went to X to respond to Yan. He said, I'm super appreciative of yan's contributions to OpenAI's alignment research and safety culture and very sad to see him leave. He's right, we have a lot more to do. We are committed to doing it. I'll have a longer post in the next couple of days. One thing that was super interesting about this whole saga that's been going on was that when Ilya left, he had nothing but good things to say. When Andrej Karpathy left, nothing but good things to say. When Logan Kilpatrick left, nothing but good things to say about OpenAI. And everybody was wondering, well why is Jan leaving? And then saying that OpenAI has kind of lost his way where everybody else is leaving and saying well everything's just fine. Well as it turns out, when you leave OpenAI you get an unpleasant surprise. A departure deal where if you you don't sign a lifelong non disparagement commitment, you lose all of your vested equity. So maybe the reason some of these people that left haven't been saying negative things about OpenAI is because they were contractually obligated not to say negative things against OpenAI or they might lose their vested equity in the company. Sam Altman took to X again to address this. We have never clawed back anyone's vested equity, nor will we do that if people do not sign a separation agreement or do don't agree to a non disparagement agreement. Vested equity is vested equity. Full stop. He goes on to say that there was a provision that was outdated that never got removed, but it should have been removed and he's embarrassed that it was in there and that the team was already in the process of fixing this. He goes on to say if any former employee who signed one of those old agreements is worried about it, they can contact me and we'll fix that too. However, it does appear to be a non issue now, as OpenAI did send an internal memo releasing former employees from controversial exit agreements. So there is a good chance a lot of the people that left OpenAI never said anything negative about their experience at OpenAI because they thought they would lose their equity if they did say something negative. It will be interesting to see now that this has been rolled back, if more previous OpenAI employees come out and start talking about trouble at the company, I'll let you know if I come across any of it. Yann Lecun, the head of AI over at Meta, had kind of a different take on Jan leaving. He says that before urgently figuring out how to control AI systems that are much smarter than us, we need to have the beginning of a hint of a design for a system smarter than a house cat. Basically, Yann Lecun is talking about Yann, the head of super alignment, leaving OpenAI and saying they're worried about AGI and these systems that are way more intelligent than us, but we're nowhere close to actually having systems that are way more intelligent than us. So people leaving and saying that they're not focused on safety is kind of a non issue because there's nothing to really be focused on yet because we're not even close to AGI. That's essentially what Jan is saying in this post, but if you want to read the whole thing, it's linked up in the description below the video. But that's not the only hot water that OpenAI found themselves in this week. Last week they demoed their GPT4O, where they were actually talking to their phone and it had a voice very reminiscent of Scarlett Johansson in the movie Her. In fact, the day before the announcement, Sam Altman himself, he even tweeted just the word Her. Well, if you've ever seen the movie her, the AI in that movie was voiced by Scarlett Johansson. Well, apparently OpenAI reached out to Scarlett Johansson and said can we use your voice in our AI? And she said no. Well, according to Scarlett Johansson, they just used her voice anyway. However, according to OpenAI, they hired a voice actor that may have some similarities to her voice, but was just a voice actor that kind of sounds like Scarlett Johansson. Sam Altman said the voice of sky is not Scarlett Johansson and it was never intended to resemble hers. He cast the voice actor behind Skye's voice before any outreach to Ms. Johansson. Out of respect for Ms. Johansson, we have paused using Sky's voice in our products. We are sorry that we didn't communicate better. Take a quick listen.

Oh, Mark, I'm doing great. Thanks for asking. How about you? You're doing a live demo right now. That's awesome.

Hi.

How you doing? Pretty good, actually. It's really nice to meet you.

Yes, Samantha and according to this Washington Post article, OpenAI did not copy Scarlett Johansson's voice for Chap GPT the Record show. This article says that when they were doing the casting calls to find the voice, they had some criteria. They should sound between 25 and 45. Their voice should be warm, engaging and charismatic. But one thing that wasn't requested was that it was a clone of actress Scarlett Johansson. Now, the voice actor who did the voice of Sky Wants to Stay. Anonymous did say that the movie her was never mentioned by OpenAI, and the actress's natural voice sounds identical to the AI generated sky voice based on the brief recordings of her initial voice test reviewed by the Washington Post. The public has seemingly rallied behind Scarlett Johansson on this, but it is a very interesting gray area case because it's somebody that just sounds like Scarlett Johansson. OpenAI never specifically was saying they were looking for a voice that sounded like Scarlett Johansson because of Sam Altman's comments of her. It was clear that he was trying to replicate her. And because Sam Altman also did in fact reach out to Scarlett Johansson and got denied to use her voice, a lot of people think that he just sort of found a roundabout way to use Scarlett Johansson's voice by using a voice actor that sounds similar to her. But it wasn't her voice. It was just somebody that sounded like her. So who's at fault here? It's hard to say OpenAI is at fault, but at the same time it's easy to make the leap to go well, OpenAI did something to make it sound as close to Scarlett Johansson as possible without actually getting Scarlett Johansson, but it wasn't her. So what grounds are there for any sort of lawsuits on this because it wasn't her voice? And another interesting bit of news that came out of OpenAI this week, they have partnered with News Corp. With this partnership, OpenAI receives access to current and archived content from News Corp's major news and information publications, including the Wall Street Journal, Barron's, Market Watch, Investor Business Daily, FN and New York Post, the Times, the Sunday Times and the sun, the Australian news.com, the Daily Telegraph, the Courier Mail, the Advertiser, and Harold sun, among others. News Corp is kind of known for being on more of the extreme sides of the News probably most notably Fox News. And I know there is some wor that this partnership may make ChatGPT more biased towards, well, the biases of News Corp. However, the reality is all of this stuff is likely already trained into ChatGPT and ChatGPT likely didn't get permission to train it in and with this agreement they now actually have the permission to move forward and use this information in future trainings. You're probably not going to see any like new or additional biases being added into your responses. It's just that now OpenAI is taking a more ethical and transparent approach to who they're actually working with to get the news that's being fed into their algorithms and trained into their models. So the handful of random people that I've seen on X and Reddit who are complained that they're working with News Corp, you probably don't have much to worry about. I don't think it's going to change the outputs in a super biased way. This falls more into that rumor territory, but this is huge if true. This came from the website Mac Daily News with the claim that Apple is going to be teaming up with OpenAI and that will be announced at WWDC. Now this was reported on by Bloomberg, but again it's not actually confirmed by anybody at OpenAI or Apple yet. But as of right now, the rumor that's going around is that it's quite possible that the new version of Siri in the next Apple iPhone will be powered by OpenAI's latest models of GPT. But we've also heard rumors that they're talking to Google about putting Gemini in there. What we'll actually end up getting, we won't actually know until wwdc. I don't think we're going to get any legitimate information about this until then. This is the current rumor we might see Chat GPT as the replacement for Siri in the future, which I'm here for because Siri sucks. All right, let's move on to Microsoft again. I was actually at the Microsoft Build event in Seattle. I had the opportunity to play around with some of these new copilot plus PCs. I got to test the new recall feature, I got to actually use a lot of the tech and and I was there for all the keynotes. So if you want to see my breakdown of that event again it looks like this and it will be linked up in the description if you want to jump to that deeper dive. But again, I know that some people just like to watch these end of week breakdowns and just get everything all in one video. So I'm going to quickly recap some of the big announcements that happened during this week at Google's Build event. There was essentially two keynotes. There was a keynote on Monday that was a smaller, more intimate group, maybe 100, 200 people. And this was focused on the consumer side, the new PCs, the new copilot features that are rolling out in Windows, and the new desktop app. And then on Tuesday, we had the main Google Build keynote, which was much more focused towards developers and how people can build with Microsoft systems like Azure and GitHub and all of those sort of things. I'm not going to really differentiate the two keynotes. I'm just going to kind of break down all of the stuff that Microsoft announced that I think you, you would be interested in, because it's the stuff that I was most interested in, starting with the introduction of the Copilot plus PC. These are computers that have a special chip in them specifically for running AI inference, for actually asking the AI question and getting the output directly on the system without it needing to go to the cloud. Most computers have a CPU and a GPU to run the processes on the computer. These new copilot PCs have a CPU, a GPU, and and an NPU, a Neural Processing unit which is dedicated to running the AI on the computer. So you can be playing a game using your gpu, but because your NPU is still freed up, you can still be running AI at the same time as well. They announced that their new Surface laptops and Surface Pros are going to have this npu. But there was a little bit of pushback on that as addressed on this Tech PowerUp website. This article says Microsoft has lost its mind. $450 for a keyboard? Apparently, if you buy a Microsoft surf tablet, it doesn't actually come with the keyboard and pen. But pretty much all of the demos they showed showed them using a keyboard and a pen using it just like a computer. But if you want that keyboard and pen that attaches to your tablet, it is $450. I don't think I've ever in my life spent more than like a hundred bucks on a keyboard. Now, I'm a big fan of mechanical keyboards, but that's just because I'm nerdy and I love the clicky sound of them. But even with my more expensive mechanical keyboards, I still don't think I've ever invested more than $100 into a single keyboard. So, $450 for a keyboard designed for the Surface does seem kind of crazy. They do have some bundles to get the keyboard slightly cheaper if you buy them together and things like that, but still kind of crazy. All right, but moving on, probably one of the most interesting new features that also led to the most controversy is the new feature called Recall, which will be built into new versions of Microsoft Windows. Imagine your history on your browser, where every website you've been to, you can go back through your history and find all the websites you visited. Recall. Is that for your entire computer? So if I was working on editing a video in DaVinci Resolve, and earlier than that I was working in email and at some point I was making images in Photoshop and then I was browsing the web and doing some research on the latest AI news, all of that stuff shows up in my recall history of what I used on my computer. So it's like your browser history, but for everything you've done on your computer. Now, obviously the very first thing that comes to mind with this is the privacy and security. And apparently the UK thinks this is a big enough issue to do an actual investigation into this recall feature. When you use recall, everything stays locally on your computer. It doesn't connect to the cloud. So Microsoft isn't actually getting the data of what you're looking at. You also have the ability to opt out of this. So if you don't want recall turned on, it won't recall anything. And you have the ability to delete anything in your history. So you can go into your recall, see stuff in there, and let's say you were gift shopping or something and you don't want that to stay in there so that you know somebody else won't see what gifts you were shopping for. Because that's the example everybody's thinking, I'm sure you can delete that history. And anybody who goes into the Recall app won't see that you were looking at gifts for people. However, the issue that people are having with this and the security issue is that if your laptop is ever stolen, if people get access to your computer, like maybe it's not password protected, they can go through and see all of your history and it doesn't blur out passwords and emails and sensitive data. If you go through your history, you'll see all that stuff front and center. It's not blurred out. If somebody manages to Trojan horse attack your computer and gain access to your computer, they might be able to access your recall and steal passwords and bank data and things like that. So while I don't think Microsoft is out to do anything shady with this. They're not trying to collect your data, they promise they're not training on your data. None of this goes to their cloud. It all stays local. But the problem is if anybody ever gains access to your computer locally, you are at serious risk. There's also the concern that if your computer is ever seized by government, they can then go through and see everything you've used your computer for, which also brings up privacy and ethical concerns. I personally thought the feature was really cool and I liked it. There's definitely scenarios where you don't want it turned on, and I imagine in those scenarios you'd use the equivalent of like an incognito mode where it doesn't save any of it to recall. But at the same time, I think Microsoft has to figure out a way to make this extra secure. So if you ever lose access to your computer or somebody ever manages to gain access to your computer, there's still some additional safety and security protocols on top of it to make sure that they can't see passwords, bank data, and any other sensitive data you wouldn't want people getting access to. Microsoft also showed off some new features. Directly inside of Microsoft Paint, you can generate AI images in Microsoft Paint. You can draw your own sketches and it will improve the sketches with AI for you. They've got restyling features where you can upload an image in a certain style and it will try to make more images in that same style. Some really cool features and they're in just the basic version of Microsoft Paint that comes with every version of Windows or it will be when these updates start to roll out. The Microsoft Edge browser also got a big translation and transcription update. If you're watching videos on YouTube and you want them live translated in real time to a new language, Edge will be able to do that. They also showed off some features where maybe if you're in Google Meet or Microsoft Teams or a Zoom call, it can actually transcribe and translate in real time as well. So I can be having a zoom meeting with you right now. You're speaking a completely different language from me, but everything I say is getting transcribed and translating for you at the bottom of the screen and everything you say is getting translated and transcribed at the bottom of the screen for me, so we can have a back and forth conversation even though we're not speaking the same language. That seems really, really valuable to me. Windows is also getting a new copy and paste feature where when you go to paste, you can paste as plain text paste as markdown, paste as JSON and you even have a clipboard history to find things that you previously saved in your clipboard as well. You will need to provide your own OpenAI API key, so there will be a slight cost, like probably less than a penny every time you use it. They also showed off the new Copilot for teams. And this was interesting because essentially what it is is another team member that's like an AI. So you can pull this team member, this AI team member into conversations, they can be pulled into emails, they can be pulled into team meetings, they can be pulled into chats inside of Microsoft and the acts as another team member that will create tasks and help create checklists, summarize the meetings you've had, things like that. It's just like another assistant, but it's an AI assistant. Microsoft is definitely moving in that AI agent direction where they want AI to be seen as just another team member working with you on whatever projects and helping you out. Microsoft also announced a new large language model called Phi 3, which is a multimodal small language model. I've been slowing down on this channel with things like comparing large language models. My good friend Matthew berman has a YouTube channel that does a really, really good job of breaking down and comparing the different large language models. So I'm sure he'll make a video pretty soon about this new Phi 3 model. For whatever reason, when I do that kind of stuff on this channel, people don't seem to like those videos. They're also not my favorite videos to make. So definitely check out Matthew Berman's channel because he'll most likely do a nice breakdown of this large language model, or I guess, small language model, and how it compares to the other available models out there. Microsoft also announced new extensions in GitHub Copilot. So if you remember what plugins were supposed to be for ChatGPT, and then we got GPTs for ChatGPT. This seems kind of like a similar idea for GitHub but with API connections. So your GitHub copilot tool can now connect directly with tools like Docker, like Sentry, like Stripe, like Pinecone. Microsoft Teams Azure. It now has direct integrations with these tools. You can also create your own private copilot extensions for your own homegrown development tooling. So if you're a coder and you're looking for AI to help you improve your coding or just speed up the process of coding, GitHub Copilot is probably the most popular tool to do that. And Microsoft announced this week that they've added these additional integrations to make it even easier for you developers out there that are doing this. Meta, wanting to stay in the mix as well, introduced Chameleon, their state of the art multimodal model. And now that all the other major models, like your Clods and your Geminis and your ChatGpts, now that all of these can actually use multimodality with text, images, audio, all of that stuff, Meta is introducing a model that can do the same. Now, this hasn't been made accessible to everybody yet, but again, there are other channels out there that are better at comparing the various large language models than mine, so check those out. I'm sure there'll be a deeper dive on that as these models get released and the general public can actually use them. And since we're on the topic of Meta, Meta put together an advisory group for AI. At the moment, it consists of these four people here. The overwhelming concern about this AI board that they put together is that it lacks a bit of diversity. Let's just say there's definitely a few things in common with all four of these people. Now they are all really, really smart people who are probably great for the job, but the world likes to see diverse mixes of culture and, well, sex among these groups that are making these big decisions for the population at large. All right, in AI Art news, Leonardo AI got a huge overhaul this week. The layout on their site is different. We can see the various presets over on the left, the various style options here, contrast options, generation modes, fast verse quality, different aspect ratios, how many images you want to generate at a time. You have some advanced settings like photo real, negative prompt transparency, tiling, or using your own fixed seed. More advanced models available here, just a nice clean website overhaul. They also added a new character reference feature. It will let you use any image to create a consistent character. They added image guidance where you can add style references, content references and character references. They've added elements and they've added a legacy mode. So if you want to go back to the way it used to look, you can go back to what you're used to. But let's go to the new mode here and let's use the image guidance. I'm going to go ahead and click on the little image up here and we have style reference, character reference or content reference. So style reference, it'll use this sort of same color palette and say it'll kind of look like it was made by the same artist, right? Then you have character reference, which will try to make all of the characters look the same with every generation. Then you have content reference, which will use the shape and structure of the reference image. I'm going to go ahead and test the character reference here. Pull in a headshot of myself. This is my character reference. And let's just say a man standing on a baseball field holding a baseball bat. Let's generate it using our character reference here. Not too bad. I have it set on fast. Instead of quality, let's try quality. I'm going to go cinematic for style. And once again, not horrible. I definitely see the resemblance. It's pretty good. One thing I love about Leonardo though, is that they just ship. There seems to be new features in this AI art generation platform, like every week or every other week. There's just constantly new stuff in here. They're trying to make the user interface more intuitive, a better, simpler layout. Just a really, really cool platform. Mid Journey also rolled out a new feature this week. In their rooms. You can actually create rooms, both public and private rooms, and collaborate on image generations. Like for example, if we go into the public rooms here and try out the general chaos room, you can see all of these images are just being created by all different people. But then we have the option to remix them. There's a chat going on in the side. These rooms are designed to collaborate on AI earlier art, either privately or publicly with a whole lot of people. And finally, Adobe Firefly is actually coming to Lightroom. You'll be able to use their AI powered Generative remove feature directly inside of Lightroom, so you won't need to, you know, color grade in Lightroom with your images and then pull them into Photoshop to remove people. You can do it all from Lightroom pretty soon. And that's what I got for you today. If you like staying looped in on the latest AI news, the coolest AI tools, and the occasional AI tutorial to show you how to do really cool viral meme type things with AI. Make sure you're subscribed to this channel and like this video and I'll make sure more videos like this show up in your YouTube feed. And if you haven't already, check out Future Tools IO. This is where I curate all the coolest AI tools I come across. I keep the AI News page up to date pretty much every single day. I can't share all the news in these videos. There's way too much happening. So all the news that doesn't get shared in these videos is also listed on the AI News page here. And we have A free newsletter where I will deliver just the most important news and the absolute coolest tools directly to your inbox. I'm doing a contest right now for an insta360x4. This camera is like a couple weeks old. It's a brand new camera from Insta360. And in order to win this camera, all you got to do is be subscribed to this YouTube channel and subscribe to the newsletter. You subscribe in those two places, you're automatically entered to win. If you've been a subscriber in both those places, you're already entered. And I'm not going to just throw in this insta360 camera. I'm also going to throw in a 256 gigabyte memory card to use inside of the camera and this really cool super long selfie stick. This thing is like 10ft long. You extend this selfie stick and this camera has AI in it that makes the selfie stick invisible. So you can extend this selfie stick up to like 10ft and it looks like a drone is flying over your head looking down on you. You can do really, really cool stuff with this huge selfie stick, X4, a memory card and this giant monster selfie stick to do really cool drone type shots. I'm gonna pick one lucky winner who subscribed to the newsletter. Subscribe to the channel at the end of May, last day of May, I'm gonna pick a winner, get your address, send you this cool prize when I'll be doing contests like this every month. So if you're not already, make sure you're subscribed to the channel. Subscribe to the newsletter. Everybody that's in those two places is entered to win. So I hope you found this video helpful. Hopefully you learned something and feel more looped in and tuned in with what's going on in the world of AI this week. There's more big events coming up from companies like Cisco and Qualcomm and Apple and Computex event and just a bunch of stuff going on right now. So the AI announcements for the rest of May and into June are still going to be really, really big weeks. I'm excited to bring it all to you and keep you looped in on all of it. So make sure you are tuned in and subscribe to this channel to keep in the loop. I really, really appreciate you. Thank you for nerding out with me today. I will see you in the next video. Bye bye.