AI News: OpenAI & ChatGPT Surpassed AGAIN?!

This week didn't have the most AI news that we've ever seen, but it had some really, really big news. We got some new large language models, we got feuds between billionaires, all sorts of fun stuff this week. But you know what? I record these on Thursdays to publish them on Friday. There's a pretty good chance that more news happened that didn't make this video. Today's video is brought to you by Mind Studio. I'll tell you more about them in just a couple of minutes. Let's get right into the news. Probably the most exciting piece of news to come out of this week was the fact that we got CLAUDE from Anthropic version three. They released it with three models, a smaller Haiku model, the free Sonnet model, and then the paid $20 a month opus model. Now I did an entire video breaking down Claude 3. It's called Claude 3 the AI that finally beats ChatGPT. Question mark. If you really want to dive deep into what Claude 3 is capable of, I highly recommend checking out this video. In many cases. I. I actually did find it to be better than ChatGPT. And in some of my tests, the free sonnet version of Claude 3 actually outperformed the paid GPT4 version of chat GPT. So check that one out, because I did do a lot of testing in that video. But here's the quick breakdown. Claude 3 Opus, their paid model, outperforms every other model. GPT4, GPT 3.5, Gemini 1.0, Gemini 1.0 to Ultra and Gemini 1.0 Pro Pro, and pretty much all of these benchmarks here. Their free model Claude 3 Sonnet actually outperformed a lot of these as well. It's a pretty dang powerful model. Claude 3 also got vision capabilities where you can upload an image and have it describe the image or answer questions about the image. Something we've had in the plus version of Chat GPT for a little bit of time now, but now you can do it in CLAUDE free version. But really the most impressive piece of Claude was the fact that it has this 200,000 token context window allowing for input and output of about 150,000 words. But they did a needle in a haystack test where they took a long document, put a single unrelated sentence somewhere buried in the document, and then asked questions about that sentence to see if the large language model would find the information from that sentence. And it surpassed a 99% accuracy in finding that needle in a haystack. And it even mentioned in one of Its responses that it believed that that sentence was placed there to test it because it didn't match the context of the rest of the video. No other large language model has been that self aware yet. But again, if you really want to dive deep on what Claude 3 is capable of, check out this video already on the channel here. But that wasn't the only new large language model we got this week. On March 7th, we got Inflection 2.5. Now, Inflection AI is the company behind the PI Chatbot, a fairly underrated chatbot. It's probably the best conversational chatbot that's available right now. Before they were saying PI had exceptional EQ emotional quotient. Well, apparently with this new Inflection 2.5 model, they've added better IQ and now they're saying that it is competitive with the world's leading large language models like GPT4 and Gemini. They say that this new model approaches GPT4's performance, but it uses only 40% of the amount of compute to actually train the model. So it's much more efficient, it's gotten better at coding and mathematics. And apparently it's rolled out right now to all users of PI. It put out this, this graph here. You can see GPT4 is in gray. The dark green is inflection 2.5, and the light green is inflection 1, their original model. And you can see that GPT4 does edge out inflection 2.5 in most of these tests. But a big difference is you can use PI for free. If you want to be using GPT4, you're either going to use it through the API and pay for it, or you're going to use it inside of ChatGPT and pay for plus a free model from PI that is probably better conversationally than ChatGPT and is now pretty close in its intelligence. Seems like a pretty good deal. Now, PI does have a web app. You can find it over at PI AI Talk. But I really find that the PI mobile app is where it's at because you can have conversations with it, almost like you're having a chat with Siri. And it works really well. And the voices are really good too. I press this little phone icon here. Tell me something really interesting that most people don't know.

Here's a really surprising fact. The first computer programmer in history was actually a woman. Her name was Ada Lovelace and she lived in the 1800s. She worked with Charles Babbage on his analytical engine, which was a mechanical computer. So the next time someone says coding is a guy thing, remind them about ada.

Thank you. No further questions.

You're welcome. Always happy to share fun facts and interesting tidbits. If you ever have any other questions or just want to chat, don't hesitate to reach out. It was great talking to you.

Now in my opinion, this is what Siri should be. It's very conversational, it can answer your questions, it's got a good AI text to speech model. I find myself using this quite often instead of chat GPT when I'm trying to answer things from my phone and this week the model got a little bit better. So check that one out and give it a try. It's free and as far as I know they don't even have paid plans available. I'll also likely do my own testing pretty soon and make a video where I test the same benchmarks that I tested with Claude and test them with this new Inflection 2.5 model and see how it compares. So if you're not subscribed to the channel, make sure you subscribe for that Outside of the two new large language models we got this week, we also got a very public beef going on between Elon Musk and Sam Altman, OpenAI. So on March 1, we got the news that Elon Musk is suing OpenAI and Sam Altman over betrayal of the nonprofit AI mission. Musk, who is one of the co founders of OpenAI, claims that Altman and Brockman convinced him to help found and bankroll the startup in 2015 with promises it would be a nonprofit focused on countering the competitive threat from Google. The founding agreement required OpenAI to make its technology freely available to the public, the lawsuit alleges. In reality, however, OpenAI Inc. Has been transformed into a closed source, de facto subsidiary of the largest technology company in the world, Microsoft. Under its new board, it's not just developing, but is actually refining an AGI to maximize profits for Microsoft rather than for the benefit of humanity, the lawsuit ads. This was a stark betrayal of the founding agreement. Apparently Musk donated over 44 million to the nonprofit between 2016 and September 2020. For the first several years, he was the largest contributor at OpenAI. Musk left in 2018 and has been offered a stake in the for profit arm of the startup, but has refused to accept it over a principled stand. Apparently Elon is trying to get OpenAI to admit that GPT4 itself constitutes AGI, it says. The suit also requests the court rule that AI systems like GPT4 and other advanced models in development constitute artificial general intelligence that reaches beyond licensing agreements. Musk wants to get some of his donation back because the company turned to a more for profit company. He also wants to bar them from monetizing technologies developed under its nonprofit for the benefit of OpenAI executives or partners like Microsoft. Well, this lawsuit sparked a rebuttal from the team over at OpenAI where they put out this letter. And one really interesting thing about this letter that a lot of people are talking about is that if you look at the authors here, we've got Greg Brockman, Ilya Suskever, John Schulman, Sam Altman, Wojciech Zaremba and OpenAI. Now the name that's interesting here is Ilya Suskever, because Ilya really hasn't been heard from on social media or anywhere else recently since Sam Altman was booted from the company back in November. After all that went down, Ilya basically went to Twitter to say I regret my decision and nobody's really heard much from him since. It looks like he put his name on this article about OpenAI and Elon Musk. It's unclear whether he actually helped helped write the article, signed off on the article after it was written, or if his name was only on there, because some of the emails that are in this blog post here are either to or from Ilya himself. So Ilya's true involvement in this message on the OpenAI blog is still kind of uncertain. But anyway, this basically sparked Sam Altman and the team over at OpenAI to share the series of emails from the early days when a lot of these open decisions were being made. They said that when they started OpenAI in 2015, Greg and Sam initially wanted to raise $100 million, but Elon pushed them to go for a billion dollars in funding. He also said he would cover whatever anyone else didn't provide. They realized that building an AGI would require vast quantities of compute. And in order to get those vast quantities of compute, they needed some sort of for profit army. Back when they started trying to figure out this for profit arm, Elon apparently wanted them to merge with Tesla or he wanted to take full control. In late 27, we and Elon decided the next step for the mission was to create a for profit entity. Elon wanted majority equity, initial board control and to be CEO. They couldn't agree to the terms on the for profit because Elon did feel that it was against the mission for any individual to have absolute control over OpenAI. Elon soon chose to leave OpenAI saying that the probability of success was zero and that he planned to build an AGI competitor within Tesla. When he left in 2018, he told the team he was supportive of us finding our own path to raising billions of dollars. He claimed that the mission was never to open source AGI. In fact, Ilya told Elon, as we get closer to building AI, it will make sense to start being less open. The open in OpenAI means that everyone should benefit from the fruits of AI after it's built, but it's totally okay to not share the science. To which Elon replied, Yep. And then all of the email exchanges that they're referring to can be also found on this blog post with a handful of redacted information, but for the most part, it's all there. You can go back through and read through all the conversations. Who knows if there was emails that also happened that aren't shared here that provide additional context from what we can piece together between what Elon said and what the OpenAI team has said. Elon helped fund OpenAI, encouraged them to raise $1 billion, and then they realized they needed to have cash flow so that they can pay for all of this. Compute, which is the for profit arm, came in when they decided to make this for profit arm. Elon wanted him to either merge with Tesla or he wanted himself to have full control. And when he didn't get either of those things, he basically bailed and said, I'm going to go build my own, but I wish you the best. They're also trying to make it clear here that Elon knew that the goal was never to open source the AI, and that open from OpenAI meant it's openly available for anybody to use, not open open source for anybody to build on. We'll see how this lawsuit plays out. But based on all of the evidence that's been presented, I don't know if Elon has much of a case. It almost feels like Elon is doing this to try to slow down OpenAI so that maybe his grok, which by the way, not open source itself, could kind of catch up with OpenAI. That's sort of the speculation that's going around. I don't know if Elon really thinks he's going to win this lawsuit or if it's just a tactic to try to advance his own agenda, but it kind of feels like the latter. Anyway, it's a fun little soap opera to watch. It feels like back when Sam Altman got booted out and there was this whole kind of story unfolding, we have a new fun story to follow along to. So I'll probably be reporting on this more as more information unfolds and we get more details. Because I don't watch the Real Housewives. I watched Elon fight Sam. This is my soap opera. Today's video is sponsored by Mind Studio Cutting Edge Platform for Creating no code AI powered applications with MindStudio, you can easily design applications to streamline sales, HR and recruiting, marketing, finance, customer service, and even craft your own AI powered workflows. This platform allows you to leverage your own data in innovative ways. Imagine uploading Excel sheets, text documents, doc files, PDFs and more and having your applications intelligently utilize this data. You can even customize these applications with specific prompts, ensuring they perform exactly as you need. You can even pull data from one area of your workflow like a lead gen form or a CRM and output an AI powered analysis or summary into another tool like a Slack channel or your email drafts using Mind Studio workflows. What sets MindStudio apart is its intuitive visual drag and drop interface, making it super easy to craft the flow of your application. And for those of you that actually like to push the boundaries, Mind Studio supports integration with all of your favorite large language models, including Palm 2, Cloud version 2 and 3 and GPT4, offering unparalleled flexibility in application development. Now, one of the most exciting aspects of mindstudio is its embed capabilities. Whether you're looking to share innovative tools with your co workers, enhance your own website, or provide added value to your customers, mindstudio has you covered. And here's probably the coolest part. Mindstudio is offering viewers of this channel an exclusive deal by using the coupon code FRIENDS100. You'll get 30 days to try out MindStudio completely for free. No reason to wait. Visit MindStudio AI and use the code FRIENDS100 to to kickstart your journey in creating AI powered applications today. And don't worry, I'll put the link in the description right towards the top so it's super easy for you to find and check out. A huge thank you to Mind Studio for sponsoring this video. Now, while we're on the topic of ChatGPT, it did get a new minor feature this week which is still pretty cool. You can now actually have it read back to you the messages. We've been able to do this on the phone for a while, but now you can do it in the browser version and it even works on existing messages so you can go back to old chats and have it speak it out for you. You just press this little speaker icon at the bottom of your message.

ChatGPT's memory capabilities are a subject of ongoing research and development by OpenAI and other organizations working on AI.

And speaking of memory capabilities, didn't they tell us weeks ago that they're rolling out memory into ChatGPT? Checking my own settings. I don't have that yet. Is that still being rolled out? What's going on with that? We've got some potentially exciting news coming out of Midjourney this week and that they're planning on releasing a new consistent character feature. We can see in this tweet here from Nick St Pierre over on Twitter X. Midjourney just said they're hoping to release the new consistent character feature along with a new Describe feature next week. And this Tweet went out February 28th. You want to look at my calendar over here? I'm recording this video on March 7th. So the next week is this week. This is a feature that a lot of people have been asking for from Mid Journey. They want to be able to create storylines where it looks like the same character in every single image that they create. And apparently this is a feature Midjourney is planning on rolling out any day now. Who knows, I do record these on Thursdays. Maybe it's out by the time you're seeing this. One thing Midjourney did roll out this week that we can access is that they just released Turbo Mode for a version six. Turbo Mode goes three and a half times faster, but it does cost two times more. Your sort of monthly Mid Journey credits will get used up twice as fast if you're using Turbo Mode. They say that this really designed for if you're like presenting on stage or maybe recording content like I do in my videos and you need a quick response, that's really what it's designed for. But most of the time you're probably still going to use relaxed or fast mode. This week, Stability AI partnered with Tripo AI to release a 3D object reconstruction. You give it an input image and it turns that input image into a 3D model. Now, we've seen a few other tools in the past that do this. CSM AKA Common Sense Machines is probably the most well known, the one that works the best. But this looks like it can do fairly similar things. You can see here's their input Images, here's the 3D objects it outputted. Here's some examples from a model called OpenLRM. And here's what it looks like from Triposr. It's available on GitHub right now, so you should be able to download everything and run it on your local computer. But I don't actually know what kind of hardware it requires. It does say it can generate 3D models in less than 5 seconds on an Nvidia A100 GPU, but I'm imagining most people watching this video aren't running on an Nvidia A100 GPU. It is made available under the MIT Open Source license, so anybody should be able to use this, build it into their own apps, play with it, sell what it generates, do whatever they want with it. There's also a hugging face space available right now where you can play with it. It's linked up in the description as well, so you can find it easily. Out of curiosity, I'm going to plug in like a little AI generated headshot of mine and see how it performs. Here's the input image. I'll leave it on remove background. Don't totally know what these do yet, so let's just kind of generate it on the default and see what happens. That was insanely fast. I mean, it literally took seconds to generate this. It's not the best quality, right? There's not a ton of detail in it, but pretty dang impressive for how fast that was. There's a random image of a wolf that I generated that's kind of realistic, but then goes into these like vectors. Let's see what it does. Pop it right here, click Generate and it's generating super fast. You can see 2.9 seconds. Okay, it's going a little bit over 2.9 seconds, but now it's ready. So it kind of messed that one up. I'm not really surprised. It's kind of a funky image to begin with, but the speed at which this generates 3D objects is really cool. And you can see some examples down here of other ones that it did really well, like this unicorn, this cat person whose tail seems to be kind of floating there. I guess it was in the original image too. A little fox got a Yoshi plushie which looks good straight on, but as soon as you turn it you're like, okay. Those proportions it doesn't quite understand. Not bad. It's called Tripo Sr. Available for free to use over on Hugging Face right now. Check that out again. Linked up in the description. This week some new AI video generation research popped up called Hyper from two ex Google DeepMind employees. This TechCrunch article here shares this generation, which I don't Know why they shared this generation? Because it's probably one of the least impressive AI video generations I've seen. It's just a two second clip where one of these mammoth or elephant or whatever's legs moves and that's it. But if we check out the Hyper website over at Hyper H A I P E R A I, they've got some better looking generations that they've done that are quite a bit more impressive than the one that was in that TechCrunch article. Like a panda riding a motorcycle and this bowl filling up. But quite honestly, it's hard to be impressed by seeing AI videos these days after what we've seen Sora capable of. These look cool, but they don't come anywhere near what we were seeing from Sora already. And it only generates two second videos. Apparently we can try it for free right now. And once we're logged in, we have the option to create a video with text, animate an image, repaint a video, create with a text prompt, animate an image or extend a video. And it looks like the difference between this create a video and animate your image and this create a video and animate an image. This one is HD and this one is not. The examples they've got this supercar driving through the water, they've got Super Mario on a Mario Kart. That one's actually pretty impressive looking. They've got some bears dancing here, some eggs and bacon frying like tilt shift, view of a city. So it looks like fairly similar to what we would expect out of maybe Runway or Pika right now. And it doesn't even look like they have a paid model yet, so looks like we can use this for free. If you want to generate little two second videos, I'll go to one of my old standbys and do a wolf howling at the moon and let's do it in watercolor. Under settings, the only option is to add your own seed or duration of 2 seconds. But there's no other options for duration. It looks like two seconds is the only opt, so we'll click Create now. It did wait in a queue for a couple minutes, so it took a while before it started generating. But once it did finally generate, it was pretty fast. And here's what we got out of it. It barely looks like there's an animation there at all. Let me go full screen. Maybe the eyes blinking a little bit, but that's about all the animation there is. The eye blinks and the head moves teeny tiny bit. So not a ton of Motion on it, but it's free. But you can check it out over at Hyper AI. Again, I don't see any paid plans. You can just keep generating until you get what you want. But again, if I'm being totally honest, it's hard to be impressed by AI video after seeing Sora. Almost everything that's come since it, you kind of go, oh cool, another one of those. Don't get me wrong, it's really impressive technology, but it's easy to take it for granted once you've seen better technology in the world of self driving cars. California gave the green light to Waymo so their robo taxis can drive on more roads. So Waymo is now allowed to drive on highways in Los Angeles and the Bay Area. If you're not familiar with Waymo, it's a self driving car company that operates similar to Uber where you can call a car and a car without a driver will come and pick you up and take you to your destination. Before, they were only able to drive on the surface roads in some of these cities and they've now gotten the permission to go up to 65 miles per hour and take passengers on the freeway. Something that I find fascinating, but would still probably be nervous about myself to jump in a driverless vehicle that goes on the freeway. But I'm going to be up in the Bay Area in a couple of weeks, so maybe I'll try it. We'll see. Now I'm going to toot my own horn for a second here because I make a lot of predictions and maybe I get like 1 in 10 of these predictions right. And when I do get one right, I got to show it off. So back on January 3, 2023, I said prediction, Nvidia will rise to be one of the top three, if not number one largest companies on the planet. AI, blockchain and gaming. Nvidia makes the dominant computing tech that powers all of them. And then this week on March 1st, Nvidia was officially one of the top three largest companies on the planet with a market cap of $22.56 trillion, seemingly with a long way to go to catch up with Apple and Microsoft. However, the day that I'm recording this, Nvidia is now worth 2.313 trillion. So in just six days they gained $300 billion in market cap, leaving them only $300 billion away from passing Apple to be the number two largest company in the world. I think they might get there. And while we're on the topic of Nvidia, Nvidia's annual GTC Conference is happening this month up in San Jose, California from March 18th to the 21st. I'm going to be there in person, but they also have a virtual event. I will make sure the link to the Nvidia event is below in the description. You're going to want to use my link. I'll tell you why in a second. But if you click on register now, they actually have a virtual event which is completely free to attend. So you can watch all of the sessions online either as they happen or the recordings of them. And it's totally free. But like I mentioned, you're going to want to use my link because if you do, I'm giving away a free GeForce RTX 4080 super GPU and video sent it to me. It's sitting on my shelf right next to me and I'm going to pick one winner who registers either in person or for the virtual event and I'm going to send you this gpu. All you got to do is sign up for the free virtual event by signing up through my link. I'll have a list of who signed up and we will pick a winner from the people that signed up through my link right after the GTC conference ends and somebody's going to get hooked up with a free RTX 4080 Super. So again, link is in the description and maybe I'll actually see some of you in person out at GTC in San Jose. And that's what I got for you this week. But I do need to remind you I record this on Thursday so there might be more news that I missed. If you want to really keep your finger on the pulse of the news on a day to day basis, make sure you check out futuretools IO. The AI News section is updated every single day. The main Future Tools website is also a great place to discover new AI tools for whatever your needs are, whether it's marketing or generative code or music. We've got tools and filters and easy ways to search for the perfect tool for your needs. And there's a free newsletter. When you sign up to the free newsletter you get free access to the AI Income Database, an ongoing, ever updated list of ways to make money using AI tools that are available again totally free. Just sign up for the free newsletter. Over at FutureTools IO you get the Income database and every week I'll keep you in the loop with just the most important AI news you need to know and just the coolest AI tools that I come across. So head over to FutureTools IO, check that out, get signed up and I'll hook you up there as well. So thank you so much for tuning in today. I hope you found this valuable and informative. If you like staying in the loop with AI news, the latest AI tools, tutorials, things like that, make sure you like this video and subscribe to this channel. I'll make sure more videos like this keep on showing up in your YouTube feed. Thank you once again for tuning in. I really appreciate you. Thank you so much to Mind Studio for sponsoring this video. You guys rock as well. And yeah, that's all I got for you today. News is getting exciting in the AI world and it's only going to keep heating up. Look out for more videos and I'll see you in the next one. Bye Bye.