AI RECAP: Rumored GPT-4o Large Model & Gemini Live vs GPT-4o Advanced Voice

All right, guys, Glasses Mat is officially back because I got to get a nice close and clear look at what's going on in the AI world. There's been a ton of hype lately around supposed OpenAI Strawberry architecture. There's been one user in particular that really racks my brain. I can't tell if they actually know what's going on or if they're just some hype machine that's just trying to get engagement. It's. I guess I just kind of have to show you guys what has happened so far and explain to you, to the best of my ability, how the story is unfolding, because it is. It is so weird, and I don't know what's going on. Maybe you guys can help me in the comments below and tell me, give me some insight. So here's the contextual rundown. All right? There's something, you know, supposedly called Q Strawberry. It's supposed to be a new reasoning engine. It's supposed to be a new architecture for large language models that allows them to achieve the same capabilities of reasoning and problem solving that humans can achieve. It's been rumored for a long time. I think well over a year at this point, this new architecture. And by the way, there is a very simple test that you can do to show if an AI can actually solve problems similar to how a human would. Now, this is no foolproof scientific test, mind you. It's just a simple way to show you that, yes, LLMs do not currently possess the same reasoning capabilities a human has. If we ask, you know, GPT4O, for example, how many Rs in the word strawberry, it gives us this lovely answer. There are two Rs in the word strawberry. Clearly wrong. You can see there's 1, 2, and 3. We can obviously get the AI to correct itself by going, hey, that isn't right. Spell it out for yourself. And then it goes. And it can. It can kind of double check and correct itself. But you can see right off the bat, most of the time, it will get this wrong. This new architecture is supposed to provide some level of thinking, some level of reasoning that can fix this issue. So, yeah, that's. That's the background on this rumored architecture. This is all rumors, Nothing official from OpenAI, just like whispers across the Internet. Here's where things get really, really strange. This account on Twitter xiruletheworldmo with a bunch of strawberries as their nameplate tweets out randomly. Welcome to level two. How do you feel? How did I make you feel? And then Sam Altman who is the CEO of of OpenAI says Amazing to be honest, as a reply to this tweet. And then Jimmy Apples who is a well known insider in the AI space, usually has good information about upcoming releases. Just says bruh. Why is he responding to this tweet from this random person? What is going on here? Well if we dig a little bit deeper into this account, I roll the world mo. They are clearly on board with this whole Q Strawberry train of this new architecture, right? Very Strawberry themed everything and have been pretty much non stop on Twitter X hyping up this release that was supposed to happen today and also something is supposed to come on Thursday apparently from OpenAI. You know, plenty of images, plenty of responses, even retweets. Oh and by the way, this guy, this user with the strawberries is apparently a viewer of this channel. He might be watching this video right now. He has replied to me several times on Twitter, liking my posts, responding to me because of course I've been tweeting about this whole situation as it's been unfolding as well. And in the middle of this hype we did see this. This was Yesterday. The official ChatGPT account on X says there's a new GPT4 Omni model out in ChatGPT since last week. Hope you all are enjoying it and check it out if you haven't. We think you'll like it. So I'm like okay, wait, now the official chat GPT account is tweeting about some update to the GPT4 omni model. You know, Strawberry guy over here is responding to it and obviously this tweet very much confused me. We don't know what model they're talking about. I haven't noticed any differences in ChatGPT and as we just saw, it is certainly no Strawberry architecture Q architecture, whatever it is that fixes the reasoning problems that we have with large language models. So I'm starting to think that this is just entirely unrelated to strawberryman. But then today one of strawberryman's predictions actually came true. No, it's not a big model release, but it is something. You see, strawberryman did indeed say that something was going to come from OpenAI today at 10am and they did indeed announce something. At that time they're releasing a new iteration of SWE Bench. This is a benchmark that more reliably evaluates AI models ability to solve real world software issues. So this is a benchmark for this issue that large language models have in order to test it. So this is great for developers. This is great for people who are building large language models. And it's not really anything huge to talk about in, let's say, a YouTube video like this, but it actually connects to another prediction that strawberryman has for Thursday, which is that we're going to be getting something called GPT4 omni large on Thursday. And actually this tweet makes a pretty good point. Have to drop a fancy new benchmark to show why your new model on Thursday is so good. That is a good point, and that's something I could see OpenAI doing. The fact that they did drop a benchmark is a somewhat decent sign. But of course this is all still rumors coming from someone who isn't really affiliated with OpenAI in any direct way, so we can't really be for sure. I guess we'll just have to wait until Thursday to see if this guy is full of it or not. But OpenAI did release something today. One of his predictions did kind of come true. He overhyped things, I think a little too much. But he was right about the timing. It was, it's, it's not easy to do that. So he clearly has some information on something. There is like a lot of hype though going around. OpenAI employees are also tweeting Strawberry things. Sam Altman, and this was days ago by the way, did post a photo of freaking strawberries in a garden and said I love summer in the garden. So he's contributing. I have also been running tests all morning on chat GPT and I haven't been able to notice a model difference. I don't have access to advanced voice yet. Still in the app. I have noticed no changes in the GPT app. It's, it's essentially the same whether it's on desktop or the phone and any of that stuff. I haven't noticed anything. Even the image generation. Again, GPT4 omni, if you remember that announcement, it can generate images. And it's not Dall? E. We don't have access to that either. It still seems to be the same Dall E3 API getting hit whether you're in the iPhone app, whether you're on desktop or Bing image creator. I don't know. I hope to God that this is true because it would be awesome. I mean, imagine GPT4.0 large with these brand new capabilities, you know, reaching that next echelon of human level reasoning and understanding. I don't know. I would love for this freaking mystery to be figured out because I don't know what's going on. It aggravates me because I want to know what's going on. Maybe you guys know something I don't. Me and my Discord server have been kind of racking our brains on this thing and we haven't really got anywhere with it. I usually try to avoid the hype and the rumors. I don't want to spread something that's not true. But it seems like there are some real teeth here. Like we have official OpenAI confirmation. You know, Sam Altman, the CEO, is hyping things up a little bit. He's responding to this user. This user's prediction somewhat came true. OpenAI did announce something at the correct time. It's, it's, it's super interesting. Personally, I would prefer it though, if OpenAI would just be truthful with us, tell us what they're working on, and then announce things when they're ready to be shipped. That would be cool. But I guess people do love the hype, and the hype has worked pretty good for Sam ultimate and OpenAI in the past. I will say, though, the ability to pass these questions like the Strawberry question is not a new thing. You can prompt them, the large language models, to actually answer this correctly. You can also, for example, like Perplexity uses search and actual reasoning to kind of figure that out and give you the correct answer pretty much every time. So I think even if OpenAI does release a GPT4O large on Thursday with Strawberry slash Q whatever advanced reasoning, they're going to have to explain to us how it actually works and in what circumstances. It makes itself a better case scenario than just prompting the large language model correctly or using outside sources or outside knowledge like Perplexity does. Anyways, that's all I'm gonna say on, on that whole topic if you want to hear more about it. I did do a live stream not too long ago where we kind of dive into it. There were some mysterious models that appeared in the LMSIS arena as well. So yeah, go to that live stream if you want to follow up on this stuff. But now we're going to get into the Google news because Google actually had like a keynote today and talked about some AI stuff. So Google is launching this thing called Gemini Live. This is kind of like chat GPT, advanced voice. They definitely have more to announce. And I actually do have some insider information on this because I am a part of Google's confidential testing team. But what I can talk about is this. Gemini Live is available on Android. It's rolling out to the advanced users, the people who pay For Gemini. I don't know why you would do that. Personally, I wouldn't. But this new Gemini Live is rolling out to those users on Android this week. I actually don't have an Android phone, so I'm going to try to demonstrate it for you guys on my computer. So, folks, I did my best. I don't have access to an Android device, but I did get this emulator for an Android device, and it says that my device is not compatible with Google Gemini app, which stinks because I really wanted to demo this for you guys. But instead, it looks like I'm gonna have to show you guys the demo that Google did themselves.

When I first go live with Gemini, there will be 10 different voices for me to choose from. Let's meet a few.

Great.

Let's get going.

Here's one of the voices I have.

But there's more to choose from.

Swipe to give them a listen, or.

Just tap, start to get chatting.

All right, it's nice, but I want to hear a couple more.

Here's another voice I can use. You can pick me now and always make a change later in Settings. I'm looking forward to discussing with you the world's most profound questions, like, why is pickleball so popular?

That is profound. Let's try one more.

Or maybe you'd like to listen to a voice like this. A pretty great one, if I do say so myself. But don't worry, there are more to explore.

All right, let's go with this one for now. So I'm going to start chatting.

You can always change my voice in Settings later. Okay, I'm here whenever you're ready.

Hi, Gemini. How are you doing?

Hi there. I'm doing well. Thanks for asking. It's always nice to hear from someone. How can I help you today?

Well, my niece and nephew are coming over this weekend, and I need a few ideas for something fun and educational we could do together. I was thinking maybe something with chemistry, something that's like a little bit magical.

That sounds like a fantastic idea. How about doing a classic science experiment like making a homemade volcano? It's super easy, and there's a real sense of excitement when it erupts. Wait, I can find a few other ideas, too. Do you want to hear about them now?

I totally do. And think magical.

All right, let's brainstorm. How about invisible ink? You can write secret messages that only appear when you heat them up with a hairdryer. Or here's another one. Try making some homemade lava lamps with oil, water, and food coloring. It Looks super cool and trippy. So what do you think?

Those are cool. I'm thinking the lava lamps, actually. Let's do the invisible ink and, like, is there a way to, like, take it up a level?

Sure. We could try making the invisible ink change colors when it's heated up. Imagine writing a secret message that goes from clear to bright pink. That'd be pretty cool, right? Or maybe instead of using a hairdryer to reveal the message, we could try making a special solution that makes it glow under a black light.

Ooh, I love the black light. That's super cool. Okay, is this going to be messy at all or. Not so much.

The invisible ink itself isn't too messy, but you'll probably need a few bowls or trays to mix everything. The blacklight part shouldn't get messy either, but you might want to cover your work area just in case.

Okay, that seems cool. So one last question for you. Help me come up with some creative name ideas for this experiment.

Absolutely. How about Spy Training Academy? Or maybe the Secret Message Lab? Cool.

I love it. All right, thanks for your help, Gemini.

So, yeah, that was the demo. On the surface, with everything that they showed, it seems like it could be a competitor to GPT4 advanced mode. But I'm going to tell you because I've actually used it, this is not natively multimodal advanced voice. This is the same kind of voice voice that you used on the regular chat GPT app. So essentially, when you're speaking to this live Google Gemini, it's taking your voice and turning it into text. It's using an AI model to do that. Then the large language model reads the text like normal text and then responds with normal text. That normal text then gets turned into audio with regular text to speech models. Again, nothing new. The thing with the GPT4 Omni advanced mode voice is that that voice is being generated by the model. The model is understanding your voice natively. So it's taking the audio in natively into the model and out natively out of the model. There's a big difference. That's the reason, for example, that the advanced voice mode has the ability to understand things like little nuances and emotions in your voice. It can understand when you're sad, it can understand when you're happy, and it can also reproduce those same things. So it's a much more dynamic model. It's a much more native model.

The model is able to generate voice in a variety of different emotive styles, and it really has a wide dynamic range. So I'M going to fire up another demo for that.

Awesome.

Yeah. Hey, ChatGPT, how are you doing?

I'm doing fantastic, thanks for asking. How about you?

Pretty good.

What's up?

So my friend Barrett here, he's been having trouble sleeping lately, and I want you to tell him a bedtime story about robots and love.

Oh, a bedtime story about robots and love. I got you covered. Gather round, Barrett. Once upon a time, in a world not too different from ours, there was a robot named Byte. Bite was a curious robot, always exploring.

I started this story, but I want a little bit more emotion in your voice, a little bit more dramatic.

Got it. Let's add some drama. Once upon a time, in a world not too different from ours, there was a robot named.

No, no, no, chatgpt. I really want maximal emotion, like maximal expressiveness. Much more than you were doing before.

Understood. Let's amplify the drama. Once upon a time, in a world not too different from ours, there was a robot named Byte.

Byte, can we do this in a robotic voice now?

Initiating dramatic robotic voice. Once upon a time, in a world not too different from ours, there was a robot named Byte. Bite was a curious robot, always exploring new circuits.

Barrett here. Calm down a little bit. Can you end the story, but do it in a singing voice?

And so Byte found another robot friend, and they lived circuitly ever after.

Thank you so much.

I definitely feel a lot more relaxed. And it's more impressive and has more interesting capabilities. Again, everything that Google kind of announced today is something that we've already seen in the OpenAI app for quite some time now, honestly. So, no, it's. It's not a competitor to the advanced voice mode. Even though I. I haven't gotten to use advanced voice mode, it's pretty clear that from the testing the community has already done that it's. It's natively multimodal and it's. It's quite different than what we're seeing today with Gemini Live. Little bit disappointing, but, you know, that's typical Google. And I mean, if you go to the Google site and look at the blog post about it, they really don't have too much here. They say, like tasks, utilities and expanded features on YouTube are coming to the Gemini app on Android, saying that Gemini is going to be more deeply integrated into Android than previously. Most of this is kind of like small stuff that doesn't really get me too excited. The one thing the calendar extension, where you can take a photo of a concert flyer and ask Gemini, you know, if you're gonna be free that day. I do like that I could see that being more useful, but I know that Apple is also going to have that with its Apple intelligence. Apple Intelligence looked very promising and really, really intriguing. So if Apple can pull through with that, again, there's nothing here that really seems to me like it's revolutionary. Google is also hyping things up like the new research that they also have the ability to do with Gemini. OpenAI as we know, is also working on its own more advanced search. And we already have Perplexity AI, which is actually fantastic AI based research. So again, Google is announcing things that competitors already have right on the market right now. Nothing too crazy. If you already have an Android, this might come in handy, but it's nothing that would make me want to switch to Android. For example, I could just use the Perplexity app on my iPhone. So for me personally, guys, this whole Gemini announcement that Google had today wasn't really anything too important. I honestly think the Strawberry rumor stuff, which might not even be real, is more exciting. But really that is all I have for you guys today. Again, like I said, I am confidential with Google, Google on some of the stuff that they have. I can't necessarily share it with you right now, so I guess you'll just sort of have to trust my word. But my goal is and has always been to give you guys the information. From my point of view, it's never to deceive you. Thank you so much for watching today's video. I will see you guys in the next one and yeah, let's hope for Thursday a big open AI release. I really hope all these rumors are true because that would be freaking awesome. See you in the next one. Goodbye.