Gemini Ultra is Here! (Google's "ChatGPT Killer")

So Google made a huge announcement today in that it a is rebranding Bard to Gemini and B, we now have access to Gemini Advanced, which uses Gemini Ultra, where the previous model that we were getting out of Bard used Gemini Pro. It's really confusing. Just keep in mind that if you're using Gemini Advanced, it's also Gemini Ultra, the model they claim is the most powerful model. This is also the same model that six months ago Google's DeepMind CEO said will eclipse ChatGPT. And in December of last year sort of doubled down on that statement, saying that its new Gemini model is more powerful than OpenAI's GPT4. If you head over to bard.google.com, where you used to go to access Bard, you'll probably see this pop up now that says Bard is now Gemini. The best way to get direct access to Google AI. All the capabilities you know and love are still here and will keep getting better in the Gemini era. So let's go ahead and click done on this. And there's two main noticeable differences here inside of, well, Gemini. Now up in the top left you can see this Gemini link and if I press the arrow you can see now there's an option for Gemini Advanced. You do need to upgrade. They are charging 20 bucks a month to use Gemini Advanced. But Gemini Advanced is using the Gemini Ultra model that they teased a couple months ago. You remember the whole thing where they made it look like they were talking to the AI on a video and it turned out that they weren't actually doing that and the whole thing was somewhat faked. Yeah. That is finally available for us to use inside of Gemini Advanced. You'll also notice this blue band across the top that says Gemini was just updated. If we check out the updates, Bard is now Gemini. We figured that out already. We have a note here about Gemini Advanced telling us that we get to use the most capable AI model, Ultra 1.0. It claims to be far more capable at highly complex tasks like coding, logical reasoning, following nuanced instructions and creative collaboration. At the moment it's only optimized for English, but it is available in in 150 different countries. And looking through the list of countries here, if you're watching this video, there's a pretty dang good chance that you have access to Gemini Advanced as well. There's also a new Gemini app available on both Android and iOS, although as of right now it's only available in English in the us but in the coming days it will be available in Japanese, Korean and English globally, except for the uk, Switzerland and European Economic Area. More countries and languages will be coming soon, and if you're in Canada, it's now available for you as well. Personally, I find the timing of this to be very interesting because it comes in the same week that Microsoft has been completely overhauling its Copilot product. Also, as of this recording, this Sunday is the super bowl here in the US And Microsoft actually has an advertisement in the super bowl for Copilot. So of course Google's going to roll out something fairly comparable the same week that Copilot's getting its moment. But I'm not going to dive too deep into Copilot in this video. I have a video coming next week that will dive into all of the cool stuff that Microsoft has been doing with Copilot. You're not subscribed. Make sure you subscribe for that video. So I'm gonna go ahead and click on this arrow up here and I'm gonna go ahead and do the upgrade. One of the nice things about doing the upgrade is you can play around with it for two months without actually paying. We can see here it's $0 for two months and then it's 20 bucks a month after that. Same price as ChatGPT. Plus, realistically, you're probably not gonna need them both. If you really like Gemini, you could probably get rid of ChatGPT. If you really like ChatGPT but hate Gemini, you probably won't. But heck, let's play around with it for two months at least. I'm going to go ahead and start the trial. I now have Google One AI premium. It also seems to come with some other perks. 2 terabytes of storage. Soon we'll have Gemini and Gmail Docs and more and other Google One Premium benefits, whatever that means. So let's go ahead and go to Gemini Advanced. And now if we click on the dropdown up at the top, we have the option to try both of them. As it turns out, I'm pretty much the last one to be making a video about Gemini. Among all of my other AI YouTuber friends, one of my favorite channels, Fireship, made a video about Gemini already. My buddy Matt Vid Pro put out a video about it. There's already a video from the AI grid on Gemini. Another one of my buddies, Matthew Berman, put out a video on it. Chris from All About AI has put out a video about it. AI Explained has a video about it already. And this channel that I actually stumbled across just today by David Andre also put out a video about it. So I was going to do a deep dive in depth analysis and test every possible thing I can think of in Gemini, but a bunch of other really smart content creat already did. So while I will be testing a few things myself in this video, I'm also going to be sharing what other people seem to have find because, well, they've already done a lot of the testing and it's fun to show love to other YouTube channels that are talking about this stuff that I'm excited about. Hopefully it helps you discover some new channels that if you like mine, you'll probably like these ones as well. The first thing that everybody seems to talk about with Gemini Advanced is how dang fast it is at actually giving you the responses. This is so much faster than GPT4 and it just cranks out responses really, really quickly. This seems to be the general consensus among pretty much everyone.

Gemini is way faster, like at least two or three times faster.

Also, the speed at which Gemini generates is absolutely blazing fast. Google's APIs are rocket ships.

One thing that Gemini Ultra 1.0 does have, it is blazingly fast. Like really, really, really really fast.

It is extremely fast.

Much faster than GPT4.

It's pretty quick. I'll give it up. It's much faster than GPT4 so that is a good thing. Gemini Ultra feels a lot faster than GPT4.

This yeah. Okay, so Gemini is done faster even though it took longer to start. That's very interesting. So yeah, every single video I've seen so far has commented on how dang fast Gemini actually gives its output. Now, when it comes to creativity, Matved Pro here tested it with a poem and found it to be really creative.

Here's a gorgeous poem written by Google Gemini Advanced. We have perfect rhyming and it's able to combine the seemingly unrelated ideas of ice cream sandwiches and the planet Saturn. I think it does a really great job in the creativity department.

Here's what Fireship said about creativity.

I ask it to write a poem about JavaScript like Bukowski. It's highly subjective, but I love this test because it requires the AI to blend the technical aspects of the JavaScript with a highly unique writing style. Gemini was actually by far the best in my opinion. The poem was dark and even included.

Some mild profanity in my own personal tests. I also found it to be fairly creative. I thought it did a good job at creativity. Now in my opinion, it's on par with ChatGPT and GPT4. I wouldn't say One really holds an edge over the other. In my opinion, they both do creativity pretty well. Now let's talk about problem solving ability, something else that a lot of people want to use these chatbots for. Matvero here actually did a test where he asked it if he was to fire a bullet horizontally out of a gun and at the exact same Y axis drop a bullet from my hand which would reach the ground first. It actually found the solution pretty quickly and even shared a diagram to sort of help better explain it. Matthew Berman here gave it a test asking if he lays five shirts out in the sun and it takes four hours to dry, how long would 20 shirts take to dry? Explain your reasoning step by step and it actually came to the right answer. So pretty decent in this problem solving scenario. He then gave it another problem about three killers being in a room, another killer entering the room, killing one of the killers, how many killers is left? And it actually got this problem wrong. Interestingly, Chris from All About AI gave it one of the exact same tests that Matthew Berman did. I hang five shirts out to dry. The sun. After 10 hours, all shirts are dry. The next day I hang 10 shirts out to dry. The conditions are the same. How long will it take the 10 shirts to dry? Very similar problem, just worded slightly different and it gave him the wrong response in this case. Chris then asks it another sort of logic problem and asks to walk through it step by step. Once again it gets this problem wrong for him. AI explained tested a logic problem here. Today I own three cars, but last year I sold two cars. How many cars do I own today? You own one car today. Well, he still owns three because he just said today I own three cars. So once again it failed. At this, he asked the Same question to GP GPT4. GPT4 got the answer right. When it comes to the ability to code, people have had mixed results with this as well. Fireship compared Gemini advanced to ChatGPT4 its ability to read code. He uploaded the same code to both platforms and both of them essentially figured out what that code did. But of course the big question is, can it actually create good code?

I had them write a whole bunch of different junior level demos like here. I had it create a basic graph database and go. The GPT4 result was a bit more simplistic, but I was able to paste it right into my IDE and run it. The Gemini result was a bit more complex and interesting, but for some reason it forgot package main at the top and I had to add it manually to get the code to run the.

AI grid tested coding. He was making a trading bot here. He actually gave it the prompt. Can you create a trading bot that uses the moving average to have entry and exit points for maximum profit? Do it in tradingview. Pine script, please. It actually generated code that didn't work the first time, but he found by simply asking the exact same question again the second time around, it did write the code properly. He then tested chat GPT with GPT4 with the same prompt, and it actually seemed to work better than the version that Gemini gave him. One thing that I find very interesting is that Matthew Berman tested coding in his video as well, trying to make a very simple Snake game. And it seems that he never managed to get it to work. Going back and forth with Gemini Ultra here, he eventually got it to pop up this game window here, but clearly no Snake involved. The reason I find that interesting is that Chris from All About AI pretty much ran the same test trying to get it to create a Snake game. It didn't do it on the first try, but it seems after a few back and forth with Gemini Ultra, he did eventually get a working Snake game to generate from Gemini Ultra. Now let's talk about Gemini's ability to generate images. It's interesting because a handful of these videos have said that they couldn't make it generate images like David's video here. He asked it to generate an image of a green flying cat. Gemini Advanced responded with, that's not something I'm able to do yet. And as we know, ChatGPT4 has no problems with that kind of thing. However, Matvet Pro here tested it with things like Create me an image of a cat who is a wizard. And Gemini did it immediately. Sure, here's an image of a cat who is a wizard. Make him more wizardy. Sure thing. Made him more wizardy. I think it might have also given him a third ear, but maybe that's what more wizardy means. Matt then also asked to put text in the image of wizard and it managed to do that for him. And in my own testing, I got it to generate images for me. Create an image of a mad scientist in the lab. Sure. Here's an image of a mad scientist in the lab. Let's generate more. And just like that, we've got a couple more variations. Pretty diverse variations as well. Let's test this same prompt that David tried here. Generate an image of a green flying cat. For me. It quickly said, sure, here's an image of a green flying cat. Now, both of these images totally suck. Dall? E does a much better job than what this gave us here, but it did manage to generate these images for me. Generating two more even got me further away from what I was looking for. But I don't know why some people aren't able to generate images while others are. Maybe it's a location thing. Maybe it's only rolled out in certain countries, maybe it's a slow rollout process and it just hasn't rolled out to his account yet. I don't know. But it does seem to be able to generate images. The more fantasy style images, they're okay, but the Imagen 2 model which Google uses behind the scenes to generate its images is actually pretty good at making realistic images. For example, if I give it the prompt create a photograph of a woman on a park bench, you can see that three out of the four are pretty realistic images. There's some funkiness going on with the hands, something that surprisingly still hasn't really been solved yet, but these all look like realistic images. I don't know why she's just sitting in a pile of leaves and there is something seriously wrong with her hand. But if you ignore the hands and you just kind of look from, you know, the arm up, this is a pretty dang realistic photo. Something that dall e3 inside of chat GPT isn't really great at yet. So we know it can generate images and it can do pretty realistic images pretty well. But how does it do with interpreting images? Because when they put out their demo video of what Gemini Ultra can do a couple months ago, the thing that people were most blown away by was its ability to actually understand what is in the images when you upload them. Well, so far this has been a pretty mixed bag of reviews, with mostly people saying it's not that great yet. For example, Matt Vid Pro posted this meme that says find 10 scholarly articles chatgpt make them up yes, why? It's humorous and it basically said it's humorous because of the juxtaposition of Kermit being wholesome, but also showing Kermit as the dark side, which totally goes over the head of Gemini for what the meme actually means. He then went on to give it an image of this car door here. Ask what kind of car it was and it just totally hallucinated it, told him the completely wrong car, telling him it was a 1967 Ford Mustang coupe. Matthew Berman did a similar test, posted this image and said tell me everything you see in this image. It claims that it's an image for an article about a large language model Lambda created by Google AI. So interesting. The preview image consists of a blue background with the text lambda written in the center in white. When clearly it says meta and then it says below that is the word meta. Actually not mentioning the llama at all in the response. So not quite as killer as the demo that we got a few months ago. Chris from All About AI uploaded this image here and it mostly got it right. The image shows a purple T shirt, a pair of black socks. I mean the color's wrong there. And a white mug sitting on a wooden table. So the colors are wrong. There's no wooden table. But it did get that all three items feature the Nvidia logo and I am Nvidia, although it really says I am AI on them. So it got some of the elements right, but not all of them. AI explained did a similar test and found that it struggled when there was faces in the images. He actually had to put a blue line through their faces so that Google didn't see the faces before it actually gave a response to the image. He also tested this image which shows a speedometer the time how many miles are left and ask questions about what's in the image. For the most part, Gemini got almost all of it wrong, I think he said. He continued to go on and prompt it some more and eventually it got it right. But it took a little bit of doing to finally get to Gemini giving the proper response. David here uploaded some pictures of mushrooms. Asked what the mushrooms were. It did notice that they were mushrooms, but did not identify the kinds of mushrooms they were for him. I'm going to try tossing this meme in here. I'm on a seafood diet. I see food and I eat it. It should be pretty easy for Gemini to understand. So let's see what happens. Explain what is funny about this meme and well, it's got people in it, so it tells me I can't help with images of people yet. Let's try this meme instead. Hairdresser, what do you think me explain what is funny about this meme. There are a few reasons why this image, known as the Confused Stonks meme might be funny. Pretty sure that's not what this meme is. The image combines two things that are not typically seen together. A llama, the question about hair, which is something typically associated with humans. I guess that's kind of right. I guess. The meme went viral during COVID 19. Confused looking llama resonated with many people who are Feeling lost and unsure about what the future held. So, yeah, I don't really think it nailed that one very well either. Another thing that is quite annoying that we can't do with Gemini right now, but that we can do with Claude and ChatGPT and most of the other bots that are out there, is that we can upload PDFs to them. This one, if I try to toss a PDF in here, it just. Nothing happens. It doesn't let me. It just tries to open it in a new tab. That's something that's a bit frustrating because, well, it's really handy to upload a PDF and then have a conversation about what the contents of that PDF is, something that all the other platforms pretty much let us do right now. There are some redeeming factors about Gemini right now. As we always say, this is the worst it's ever going to be. So I am not counting out Google yet. I think Google has some bigger updates up their sleeve. We're going to see this sort of rollout more rapidly. They're trying to keep pace with Microsoft, so I don't think they're going to be moving in slow motion anymore. Some of the things I do like, the speed that it generates is faster than pretty much any of the other models we have access to right now. It just cranks out text really quickly. We've got access to extensions, which with ChatGPT, we do have GPTs and plugins and there's ways to extend ChatGPT. However, what I like about these extensions is it sort of ties you back into your Google ecosystem. So if you're using Google maps, you're using YouTube, Google Hotels, Google Flights, Google Workspace, all of these various tools, that information is going to be connectable with your chats. Inside of Gemini, that's really, really valuable. If you're really dialed into the Google ecosystem, you'll be able to have these conversations and, and have it pull up information and data that you've saved in these various other platforms. Gemini is also great if whatever you're trying to chat about requires web searching. Now, we do know that ChatGPT can search the web with Bing, but because Gemini is made by Google and Google is the biggest search engine, I think its search capabilities are a bit better than what we see out of ChatGPT. I don't have an amazing example, but since the super bowl is coming up this weekend, let's just ask it who do most people think are going to win the super bowl this year? Super bowl this year between Kansas City Chiefs, San Francisco 49ers it's difficult to say who most people think will win, as opinions are always divided. They go on to talk about why it's a tough call, basically saying it's a toss up. But Google has this double check your response button, which will search Google and actually cite its sources for where it got this information. So I can click on any one of these arrows and it will pull up details about the website and the actual quote from the website. It's also got this option to modify your response. So if I want a shorter response, a longer response, a simpler response, a more casual or more professional response, I can quickly do that with the single click of a button, which is pretty handy. Let's get a shorter response and there we go. We have a shorter response with pretty much the same answer. It's also really good at creativity, like both myself, Matvet Pro and the Fireship YouTube channel have all shown off. However, it's pretty comparable with Chat GPT. So I wouldn't say that you should pay for Gemini for the creativity. If you're already paying for Chat GPT, it's going to be about the same. And that's it. That's my breakdown of Gemini so far. Today is day one. It will be getting updated and upgraded and faster and better over time, but I wanted to quickly get a video out and talk about what I found from it and what other people are finding from it. And if you like AI channels and you like my channel, well, there's a lot of amazing AI channels out there and I wanted to make you aware of them. So if you're not already checking out these channels, make sure you check out Fireship and Matved Pro, AI and the AI grid and Matthew Berman and all about AI and AI explained. And now David Andre, who is the newest one that I am now subscribed to. I like to keep my finger on the pulse of AI. So I watch as much of this content as I possibly can, see what other people are talking about, and I figured this would be a fun time to show off some other channels, but also give you a sort of mashup of how everybody is sort of feeling about Gemini. Overall, the general consensus consensus, it's really fast. It's pretty creative, on par with GPT4 problem solving abilities. Still need some work. Chat GPT actually still seems to be a little bit better at this coding seems to be a mixed bag from all the videos I watched. Some people thought Gemini was better, some people thought Chat GPT was better. Kind of a toss up there Image generation if you want really realistic images. Imagen 2 built into Gemini works really, really well if you want it to follow your prompt perfectly to what you asked it. Dall E3 instead of chat GPT is probably going to be better image understanding. So far, not that impressive. Most of the image understanding tests that we've all given it, it didn't do a great job of understanding what was going on in the image, but the fact that it's got extensions, it's really good at searching the web and citing its sources, the fact that it's really, really fast, and the fact that it's Google, who has probably more data than any company on the planet, all makes it something to keep an eye on whether or not they will ever catch up with Microsoft. Combined with OpenAI yet to be seen, GPT5 is probably going to come out this year and eclipse what we're getting with Gemini Ultra. It's always going to be this cat and mouse game most likely. But hey, us consumers that are loving watching AI, loving playing with AI, we're the real winners here because we get to play with them all and stick with whatever works best for us. Should you go grab Gemini? I mean it's two months free, might as well get in and play with it. Mark your calendar. If you don't like it, cancel it before two months is up and you never pay a cent. Nothing really lost, so it's worth a play with. But as of right now, I'm probably still going to jump into Chat GPT and Claude a little bit more often. Hopefully you found this video helpful. If you love nerding out about AI tools and the latest AI news, check out FutureTools IO. This is where I curate all of the coolest tools that I come across. I keep the AI News page up to date on a daily basis and I have a free newsletter where I'll share just the coolest tools I come across and just the most important news you should know. And when you sign up, you'll get free access to the AI Income database, which is a database of all sorts of cool ways to make money using AI. So check it out. It's free. You can find it at futuretools IO. As always, if you like this video, give it a thumbs up. And if you want more videos like this, subscribe and I'll make sure more show up in your YouTube feed. Thank you so much for tuning in and nerding out with me. I really, really appreciate you. I'm having so much fun with all of this cool AI tech that's rolling out lately, and things seem to be ramping up again. So exciting times. Really appreciate you. See you in the next video. Bye. Bye. Today you own one car. Today. Well, he still owns three because he just said, today I own three cars. So once again, it failed at this. He asked the Same question to GPT4. GPT4 got the answer right. When it comes to the ability to code, people have had mixed results with this as well. Fireship compared Gemini advanced to ChatGPT4 its ability to read code. He uploaded the same code to both platforms, and both of them essentially figured out what that code did. But of course, the big question is, can it actually create good code?