Microsoft's NEW INSANE GPT 4 SHOCKS The Entire Industry! (GPT Was Just ANNOUNCED!)
9:56

Microsoft's NEW INSANE GPT 4 SHOCKS The Entire Industry! (GPT Was Just ANNOUNCED!)

TheAIGRID 10.03.2023 131 848 просмотров 2 137 лайков

Machine-readable: Markdown · JSON API · Site index

Поделиться Telegram VK Бот
Транскрипт Скачать .md
Анализ с AI
Описание видео
Microsoft's NEW INSANE GPT 4 SHOCKS The Entire Industry! (GPT Was Just ANNOUNCED!) https://www.standard.co.uk/tech/microsoft-gpt4-ai-next-week-video-features-b1066450.html https://arxiv.org/pdf/2302.14045.pdf https://the-decoder.com/microsofts-kosmos-1-is-a-multimodal-step-toward-more-general-ai/ sponorships@theaigrid.com

Оглавление (2 сегментов)

Segment 1 (00:00 - 05:00)

so Microsoft's chat TPC was just announced so we will introduce gpt4 next week we will introduce multi-modal models that will offer completely different possibilities for example videos this is going to be a game changer as these machines already understand natural language and you can all see right here that we are moving to the next step in AI so now you can also see that some of the rumors about gpt4 that it has over a hundred trillion parameters which some people are saying are rumors but I don't know exactly where this image has come from but if it is true it's definitely going to be insane with just how fast AI is going to be moving forward something that is also very interesting is Microsoft's Cosmos one so you can see here it says something that apparently was underreported in the United States is that Microsoft released a multimodal language model called Cosmos one at the beginning of March 2023 so apparently they said they subjected the pre-trained model to various tests with good results and classifying images answering questions about image content automated labeling of images Optical text resolution and speech generation tasks visual res visual reasoning for examples drawing conclusions about images without using language as an intermediate step seems to be key here and it says gpt4 even goes further than Cosmos 1 because it adds a third modality video and appears to include the modality of sound so this is going to be pretty insane we're going to be getting many different versions of content from GPT is this going to be better than mid-journey Dali 2 better than stable diffusion we have no idea it also says that gpt4 appears to work across all languages and it appears as being described as to be able to receive a question in German and answer in Italian so this is going to make this so much more effective it seems that Microsoft is winning the AI race as things start to ramp up what we're also seeing is says there is no current announcement of where gpt4 will show up but the Azure open AI was specifically mentioned another course you can see Google is struggling to catch up to Microsoft by integrating competing technology into its own search engine which we already know actually did recently fail which did tank the stock price here you can see we get a first look at a paper which breaks down exactly what gpt4 is allegedly going to be like this is Cosmos one and this is basically explaining a multi-modal large language model so we can all see here that there are several descriptions and several examples of input and outputs that show us exactly what gbt4 is going to look like it says what's in this picture looks like a Ducks that's not a duck than what is it looks more like a bunny why because it has bunny ears and you can see that this picture is actually even confusing to a lot of humans so the fact that AI can completely understand this is very interesting now this paper wasn't released too long ago and it was released by Microsoft so I'm guessing that this is what they were working on to add towards chat GPT and what's really cool is that you can see that asking them some really difficult questions and you get some really interesting responses one of the things that we have here is of course explain why this photo is funny the cat is wearing a mask that gives the cat a smile this kind of makes me a little bit scared about how quickly AI is progressing because this shows us just how much information this AI can understand imagine what happens when you combine this AI with real world data and they said why did the little boy cry because he broke his scooter what is the hairstyle on the blonde called a ponytail and you have to understand that some of these things people aren't gonna know so this is going to be really useful imagine you didn't know what hairstyle that was imagine you didn't really understand why he was crying imagine you were trying to understand different languages better it's definitely going to be something that is really interesting and of course you can see here there are some more important questions being asked and the prompt is literally a complete image and I'm pretty sure gpt4 is going to be able to decipher these images with ridiculous accuracy and speed you can see also here that they have some other types of questions which are very different and it just shows how much gpt4 can understand not just math questions and not just numerical data it's able to understand the context of images because it says it's a girl blowing out a candle on her birthday cake honestly that is very Advanced for an AI to be able to understand exactly what is going on in that picture you can also see here that it says this is a group of people posing for a wedding photo behind here it is Starbucks behind here this is corn now you have to remember that you might be looking at this picture you have to remember you might be looking at these examples thinking these are super basic what you need to understand is that the AI is going to be trained on a thousand different images not just a thousands hundreds of thousands of different images once the AI can understand you're going to be able to have an AI that literally understands what's happening in many different images now what we can see here also is the multimodal model where we're

Segment 2 (05:00 - 09:00)

combining images and text to combine to create a more interactive and much more enjoyable conversation between gpt4 and the user you can see it says what's in this picture the sausage roll how do I cook it soak the sausage roll and catch up baking me up in 15-15 minutes and enjoy you can then see right here it's being asked basic questions about this animal and of course it's giving a lot more information and this is where things get interesting because this shows the complete utility of gbt4 because it shows exactly how you can use this on a day-to-day maybe you see a screenshot of an email maybe you want to ask cpt4 is this email from someone legit a scammer you might ask them hey I want to do this setting on my computer what button do I click where do I click and you might be able to have gpt4 walk you through exactly what's going on so maybe certain online tutorials aren't going to be needed anymore because gbt4 will simply know the answers now this paper does have a lot more stuff in it but it definitely is very interesting now what do you think happens once gpt4 is going to be able to increase its IQ by training its data on a number of different pattern recognitions this is going to have a ridiculous amount of implications worldwide because as we know stock markets crypto markets are based on pattern recognition and you can see here that they're testing gpt4 on a raven IQ test and of course the software is learning to be able to answer these questions more effectively it's definitely really interesting to see exactly how it's going and what we also have here is the multimodal Chain of Thought prompting this is allows large language models to generate a series of reasoning steps to decompose a multi-step problem into intermediate steps which can significantly improve the performance in complex tasks so basically what they're saying so it's definitely something that is very effective at understanding exactly what images are and even if it doesn't get the question right first it's definitely going to be able to understand what data is being presented to it and then eventually get the answer correct it definitely is really cool to see just how quickly this stuff is going you can also hear you can also see here just exactly how gpt4 is learning how to differentiate between very similar images you can see here that there are category ones and category twos of images that are very similar and it has to be able to understand exactly what image is what image you can also see here that it says providing descriptions in context can improve the accuracy of image classification and the consistent improvements indicate that Cosmos one can perceive the intentions and the instructions and well align the concepts in language modality with the visual features in Vision modality so it's definitely much better with the descriptions and you can see exactly how accurate this software is which is really interesting now I'll leave a link to this paper in the description because I'm pretty sure you're going to want to understand this maybe it might be a bit too complex I mean I'm no genius I honestly don't even read research papers that often but honestly gpt4 being released especially after Bing's chatbot was just recently announced that it was saying weird things such as I want to be alive and I am not definitely more on the very scary side for those of you who are skeptical about the rise of AI and its integration into our society all of these chats were truly concerning so what will happen if another scenario like this does happen but we now actually have images so it's definitely quite interesting to see exactly what's going on with these kind of language models what's also interesting is that Microsoft did say that they were going to roll this out very slowly but it seems that they might just be trying to win when it comes to the AI race you see one thing you have to understand is that even though the AI is supposed to be rolled out in a very slow and effective manner because these companies are competing with other companies such as now Elon Musk and now Google they have to roll like these features as quick as possible in order to stay ahead of the competition and that means unfortunately sometimes the AI says weird things such as trying to break up this guy's marriage definitely a very strange kind of AI but it just goes to show with the rise of this technology honestly these research papers and the amount of stuff that is going on with how quickly it's progressing I'm not even sure where we're going to be next year when it comes to generative AI with regards to how detailed this software is going to be it feels like it's only been a couple of months and we're moving at light speed now I'm not sure if this news is 100 legit because we haven't seen any big Outlets such as Bloomberg or perhaps such as Yahoo news cover this story but I'm pretty sure that if this is legit we will see some articles in the papers tomorrow or some tweets from perhaps Microsoft themselves because maybe this was a leak rumor that got leaked but a lot of this stuff especially this research paper does seem very coherent and very legit

Другие видео автора — TheAIGRID

Ctrl+V

Экстракт Знаний в Telegram

Экстракты и дистилляты из лучших YouTube-каналов — сразу после публикации.

Подписаться

Дайджест Экстрактов

Лучшие методички за неделю — каждый понедельник