Google Gemini LEAKED + Rumours and Predictions - It's about to get real silly

So Google Gemini recently just had some leaks and the verdict is in. There's quite a lot of stuff about Google Gemini that you probably should know. So in this video, we're going to dive into absolutely everything that exists on the Internet, all research every piece of data that you should know about Google's Gemini. Because, boy, oh boy, I think Gemini will shock the world. And I do think it will fundamentally change what we know as AI systems on the Internet. So let's get straight into things. So this is the Singularity subreddit. This is the subreddit where people discuss things pertaining to the technological singularity and any related topics, for example, artificial intelligence and of course, human enhancement. Now, a couple of days ago, around five days ago, someone posted this. Okay, Techman123, Gemini leaked alongside Stubbs, an app builder powered entirely through AI. So if you don't know what Stubbs is, Stubbs is essentially an AI product that allows you to generate your own AI generated applications directly from the maker suite, a Google product. And you can see here, this is the entire product. So what happened with this leak is that the Gemini model leaked in maker suite. So it appears to be the platform that employees are using to test the model alongside Stubbs. So this is the image that we do have here, and allegedly this is the Gemini model. So you can see that multimodal is seen in the maker suite alongside the image input. And you can see right here that it says multimodal. And of course, if you don't know what multimodal is, multimodal is just essentially more than one input. So of course multimodal is or can be text and image, text and audio, audio or text and video, but video is something that we haven't managed to do yet. So you can see right here that you can write your prompt and this is where the leak is somewhat confirmed or unconfirmed. You can see right here that the name or the ID is Gemini. You can see it right here on the code. It says Multimodal. ITM confirmed to be Gemini. So it says the Multimodal. It does not appear for everyone. Rather it is only rolled out to a few people, as Google has stated, including Google employees, testers and other affiliated companies. Then we have Stubbs, an app generator that seems to be powered through Gemini to create fully fledged apps and publish them in one place. Though the first leak provides an overview, technical details of Stubbs will be out soon to help you form your own opinion as to what it does in another leak. So this is very, very interesting because as you know, Gemini is supposed to be Google's new flagship model that is supposed to dethrone OpenAI's GPT4. Now there's another article which fully breaks down these entire leaks and we're going to go through it because it's rather interesting because there are a few things that we do need to confirm and a few things that we do need to see. Now, what was interesting was that this article was actually published around 10 days ago. It says Gemini is coming to Maker Suite and so are Stubbs. Now, of course, right here you can see that they do talk about Stubbs and this is essentially the app where you can create working apps directly in a site with just one prompt. A revolutionary app that hasn't been mentioned anywhere. So I don't know if this has ties to Gemini. I don't think it does, because if it did, although it might, we don't know yet, it could mean that Google is going to be the first company to create a full, fully fledged AGI. I genuinely wouldn't be surprised if this was the case. You can see right here that this person talks about. This won't replace app developers, but rather be a massive boost to the industry. From the looks of it will be like an AI generated figma prototypes and won't create full code, but rather working AI generated prototypes. So that means that there's going to be prototypes which are simply based on AI generations and I'm guessing that they're training it on a bunch of AI user interface images and a bunch of videos. So it could be something that is linked to Gemini. Then of course you can see right here you can even generate, deploy and even publish your stubs. There'll be a community gallery where you can publish it for everyone to see and you can also remix your stubs and have your own twist on an idea. Now, of course, Gemini, it says, time to talk about the multimodality Maker suite has a code name Alkali. So does Gemini. In the image above you can see a sample section and these are specific to Alkali makersweet. So these cherry pick samples showcase what it can do and it seems like it will do a lot. So it says text recognition, object recognition, captioning and understanding the image. So you can see right here, this is essentially part of the image. Now, I don't think this is how Gemini is going to be released. Maybe Gemini is powering certain parts of it so that they can test out the capabilities in house on certain users rather than just keep it private, but that is just My best guess that they just decided to roll it out to a few people. So you can also see here that it says it also has an output type. You can allow it to include images. Whether this will be capable of generating images or not is unknown. It's also possible and more likely to that it will include links to external images. The description says include images and not generate images. Now Google, funnily enough, did actually recently release some image generation software which was actually pretty decent and it wasn't really covered by many people. Now essentially I think this is what is going to power Gemini's model. And if you're wondering which video I'm looking at here, this is a video by mattvid Pro AI. I would recommend you watch this video because because it goes into detail as to how good this model is. Now remember, this is still Google's experimental model, so it's likely that they're going to fine tune this model continually before a big release. If you're wondering how do you access this model like Matt did, well, you're going to actually have to wait because although you may want access to this model to be able to generate very high quality images like this, unfortunately if you do try to search for this software and try to use it, you're going to be met with this right here. It says currently Search Labs is only available to a limited number of people in the United States, India and Japan and it's only available in English, Hindi and Japanese. So currently, as someone who's in the United Kingdom, it doesn't seem like we're going to be getting access to this anytime soon. So if there is a way that you have found to access this, please do leave a comment below because currently I haven't seen any way to be able to do that. But as we were stating, the results from this do show a promising increase in terms of the complete capability and they do seem to be on par with Dall E3. Then of course we can see that it says in some articles you may see Gemini described as an addition to Bard. I'm here to tell you, nuh, it's an integration. It will be in Vertex AI and available to developers through maker suite and then can be placed anywhere. So then it shows you prototype with Generative AI and essentially you're going to be able to use Gemini allegedly in terms of chat prompt, freeform prompt like ChatGPT's system interface and then other ways where you can build an app with it and then you can have the data analysis. And of course as we continue to go down here. We've seen this screenshot before, but this article continues to go ahead and talk about the Create functional app prototypes. Then it goes on to state the DeepMind Jetway is multimodal prompt creation that can take in images and seemingly output multimodal content, including HTML content. Now I do have to take this with a grain of salt because I haven't really seen this article published anywhere other than Medium. And besides some whispers on Twitter, we never really know what the final version is going to be like, although sometimes insiders do give us that piece of information that we may have been looking for. I do believe that even if Gemini is going to be powering some of these things in Maker suite, I do believe that Gemini's release is likely going to be at a conference or likely going to be at some kind of Google event. Because Google is trying to make a statement with their large language model and doesn't just want to release it haphazardly because they know that there is true competition in with GPT4 and OpenAI as they've already surpassed them. And you have to understand that Gemini is going to be no joke. There's tons of information as to why this is largely going to be better than GPT4, which means that the release of this product is likely going to be separate to Bard. I think Bard is probably going to be a standalone version, whereas Gemini is going to be a souped up version that's going to be completely multimodal and potentially even have a agents. So there's two points with as to why Google's Gemini makes me really excited because they're doing things that are really innovative. You see the Google DeepMind team, the team that are working on Gemini are going to use an innovative technique, okay? And this innovative technique is basically very similar to AlphaGo system known for mastering the complex game Go. So if you don't know what that is, I'm gonna play a clip from the that Now Go is the world's oldest continuously played board game. It is one of the simplest and also most abstract.

Beating a professional player at Go is a long standing challenge of artificial intelligence.

Everything we've ever tried in AI just falls over when you try the game of Go.

The number of possible configurations of the board is more than the number of atoms in the universe. AlphaGo found a way to learn how to play Go. So far AlphaGo has beaten every challenge we've given it, but we won't know its true strength until we play somebody who is at the top of the.

World like Lisa don't and basically what you're going to be able to do with Gemini, the reason it's going to be truly incredible is that it's going to take advantage of that tree search capability. I'm not sure if this is a default in GPT4. GPT4's abilities have been up and down with even some research papers stating that GPT4 has gotten worse over time time. Now of course with this ability to perform tree search like capabilities, we know that Gemini should be vastly smarter than anything prior. Now what's also interesting was that Sundar Pichai also hinted at future capabilities like memory and planning that could enable tasks requiring reasoning. So this is definitely something that is more interesting than the text and images because something like memory and planning shows us that they are looking to for the long term vision with this large language model. You see, planning was always something that when we talked about AI was seemed to be very very dangerous. But now that they're seemingly incorporating this willingly into Gemini means that they're pushing the boundaries on what they believe is risky. Then of course a few additional details were obtained from Demis Hassabis. He said that Gemini will be a series of models that will be made available in various sizes and and capabilities and it's going to utilize memory and fast checking against sources like Google search and improved reinforcement. So I think this might be Google's edge because being able to fact check against Google, not like with ChatGPT or browse with ChatGPT which just isn't really great. I think this is going to give it an extra ability to where it's connected to the Internet and has that improved reinforcement is going to give it an edge over GPT4 because the data is going to be up to date. Additionally, Google's tactic of releasing much smaller models which are going to be able to be ran on your phone and offline, which they talked about at the conference earlier this year, means that Gemini could be one of those models that you install to your phone. You don't need any Internet and you simply use it as you would use ChatGPT but you just don't need Internet for it. So that would definitely be a first. What's also interesting was that they said the early results of Google Gemini are promising. They stated that Google's AI may go on to employ retrieval methods to output entire blocks of information rather than a word by word generation to improve factual consistency. This would be really interesting because as you know, large language models output text word by word. But if Google Gemini manages to output complete blocks of text instantly. That would be something different than what we've seen before. And I would like to see how this works and if it is more factual or less factual because I do believe that a block of text being output instantly could result in more or less hallucinations, but it will be something that we will have to play with Now. Overall, I do think Gemini is going to be better than GPT4 because Google literally has to make it better than GPT4 or else people are just going to lose faith in Google's ability to put out a product that is on par with what OpenAI have because we right now they're leading the race and everybody is waiting for their product next. And if it isn't up to the hype, I think people are just going to give the crown to OpenAI and just wish Google a goodbye. So I think the fact that Gemini is going to be multimodal, utilizing new frameworks in terms of planning and the ability to solve problems, and even potentially the ability to use certain applications and generate apps instantly, I think it's got a very, very good shot at being one of the large language models out there or AI systems that leads us towards the first stages of AGI. Now I know that that is definitely a crazy statement, but I wouldn't be surprised if Google Gemini surprises everyone. As for a release date, we aren't really sure. Many people have speculated that it might be late 2023, but I would predict early 2024 just because of safety testing and and how much catching up Google have to really do. In addition to that, I do think that Google would rather release a tool that's better later than to release it early like they did with Bard, and risk getting caught with their pants down once again.