# GPT-4 BEATEN, Deepminds New AI, Googles Secret AI,

## Метаданные

- **Канал:** TheAIGRID
- **YouTube:** https://www.youtube.com/watch?v=BGa24bsRKJM
- **Дата:** 03.07.2023
- **Длительность:** 15:44
- **Просмотры:** 30,360
- **Источник:** https://ekstraktznaniy.ru/video/14793

## Описание

Welcome to our channel where we bring you the latest breakthroughs in AI. From deep learning to robotics, we cover it all. Our videos offer valuable insights and perspectives that will expand your knowledge and understanding of this rapidly evolving field. Be sure to subscribe and stay updated on our latest videos.

Imagen : https://www.youtube.com/watch?v=pN-RTBq6i3I
Gooogle Ai - https://www.youtube.com/watch?v=6n5ngB88DHU
LSC Robot - https://languageguidedskillcoordination.github.io
Nvidia - https://blogs.nvidia.com/blog/2023/06/27/generative-ai-debut-mlperf/
Nvidia - https://blogs.nvidia.com/blog/2023/06/27/insilico-medicine-uses-generative-ai-to-accelerate-drug-discovery/
Tapir - https://deepmind-tapir.github.io
GPT 4 - Beaten - https://www.cnbc.com/2023/06/27/baidu-claims-its-ernie-bot-beats-openais-chatgpt-on-key-ai-tests.html
DragGan - https://vcai.mpi-inf.mpg.de/projects/DragGAN/

Was there anything we missed?

(For Business Enquiries)  contact@theaigrid.com

#LLM #Largelanguag

## Транскрипт

### Intro []

let's take a look at some of the things that happened this week in artificial intelligence that you may have unfortunately missed lots of these are really interesting so watch until the end because trust me when I say you'll be Mighty surprised by what you see coming in at number one is of course

### Tappy [0:15]

Google's deep mind they actually made this really cool project called tapir essentially what this project does is it's able to independently track any point in a video now you might be wondering why is this really cool well if we compare it to other examples that we've seen before it just isn't as effective you see what tap here is able to do is get a point on an object and keep it there for a very long time and if you're wondering about the applications this just has many and many different applications from enhancing all video editing software to being able to track things independently of what we see it's definitely going to change a lot of Industries but of course as you know this is still in its early stages so with Google deepmind we know that they constantly release new things and I've got to be honest tappy is one of those things that I will think is

### Imogen [1:06]

going to be great for the future then we got this state-of-the-art project from Google Now what you're looking at is Imogen unprecedented photo realism and a deep level of language understanding so this is from Google researchers brain team now essentially why this is so cool is that this wasn't just a research paper Google had been working on this for quite some time and finally they managed to release this so what you can see here is a dog looking into the mirror seeing a cat a dragon fruit wearing a karate seat belt in the snow A poranium sitting on the throne there are many different things that you can think about now essentially if you're wondering what exactly is this like mid-journey well yes it's like mid-journey but what's cool about this is that Imogen isn't just an image editor they also have image and video and an image in Video Edit now image and video as you can see right here isn't as good as many other things but it is still in its early stages where you're able to generate great video from text now what was also cool about this release by Google was they actually released this to the public so from this video that you're able to see you can see that you can write a prompt in the text box below and upload an image and then of course you can edit it it's actually really similar to adobe's new firefight thing so I would say once again Google are showing this that literally every week they're committed to releasing new and Innovative AI products that we are going to be using and finally we did get this one released so this is of course image vertex on AI and I'll leave a link to this one in the description

### Baidu [2:33]

then of course this is the title of the video apparently China's Baidu claims its Ernie Bot Beats chat GPT on key tests as artificial intelligence race heats up so Baidu said on Tuesday the artificial intelligence model called Ernie 3. 5 outperformed openai's chat GPT and gpt4 in several key areas the Chinese Searcher revealed that Ernie in March and has since publicly been testing it in China the chatbot is based on baidu's foundational AI model called Ernie the focus on baidu's advancements underscore the intense competition taking place in the area of generative AI with technology Giants in the United States and China rapidly advancing developments so this article basically states that gpt4 was beaten a large language model created by Chinese developers now of course an important distinction is to make that Ernie 3. 5 developed by China did beat gpt4 in Chinese language tests according to the science journal so this goes to show that although these are some small headways that certain individual AI companies are making it is important to note that rapid artificial intelligence development is going to continue regardless of whatever jurisdiction that we are in of course there are some people calling for the Slowdown of gpt4 and of course the Slowdown of the development of gpt5 but we can see here across the globe everywhere companies individuals corporations are rushing to make their own AI chatbots which they of course trying to surpass tpt4 with as is right now that is the very best Benchmark so it will be interesting to see maybe if a month from now two months from now three months from now gpt4 does completely get

### Google [4:21]

beta then of course we had Google announce their virtual try on feature which is going to revolutionize online clothes shopping essentially what it does it allows users to see how close look on Real Models with different body shapes and sizes I think this is a great addition to the workspace because this is going to allow individuals who have maybe a unique body to be able to see exactly how that piece of clothing fits onto their body it was a really intriguing research paper but we know stuff like this is going to be implemented because we've seen small things like this before it's definitely really cool and I think rapid AI advancements like this are ones that definitely make the world easier to use and a better place now what Google also included was they also announced that they're updating their search generative experience this was something that took Everyone by storm but at the same time we weren't that surprised because we did notice that they were talking about this in their Google I O event so of course essentially what it does is it gives an AI generated response rather than a single search query so you know how on Google sometimes you search something and you really can't find what you're looking for with sge what they do is they tailor it specifically to you to all your cookies to all your preferences and then you're able to talk with the chatbot and find exactly what you're looking for I find that these kind of Integrations into the web are definitely going to be much more effective and since people do use Google anyways over Bing I think that is going to remain the user Choice the only thing is when will they release this to the public as you know Google is slowly trying to rush out these products but at the same time they're quite scared because their AI hasn't proven to be too good just yet then of course we had a company called AI AMD released a new AI chip specialized for generative AI now I have to be honest with you guys if you don't know who AMD is then maybe you're not someone who's that much into computers but for those of you who are you likely know who AMD are if you've heard of Nvidia you'll likely have heard of AMD they're essentially direct competitors with Nvidia taking the Lion's Share of the graphics card market now AMD what they've done is they've released this card to the supply for generative AI related products and apparently they described their m1300x chip as the world's most advanced accelerator for generative Ai and apparently it's expected to attract interest from big cloud providers such as Amazon or Microsoft but they haven't actually specified who's going to be using it and it will be interesting because if AMD is able to create a super chip that is far superior than nvidia's then it's likely going to push Nvidia to push the boundaries even further and to be honest with you guys Nvidia does seem to be doing things at a increasingly faster rate with many different advancements in AI that we're going to get then of course we had some groundbreaking news where we had quicker cures how in Silicon medicine uses generative AI to accelerate drug Discovery so essentially in silico used generative AI for each step of the pre-clinical drug Discovery process to identify a molecule that a drug compound could Target to generate novel drug candidate gauge how well these candidates would bind with the Target and even predict the outcome of clinical trials and doing this using traditional method would have cost more than 400 million dollars and taken up to six years but with generative AI in silico accomplished them for one tenth of the cost and one-third of the time reaching the first phase of clinical trials and just two and a half years after beginning the project so like we stated before although AI might not just make one huge jump it's every small incremental jump that advanced society as a whole in fact when we also look at another Nvidia blog post we can see that their recent h100 gpus have set a new standard for generative AI in the debut machine learning performance Benchmark so this is absolutely incredible we can seem that with these advancements in generative Ai and of course in processing power and in chip speed we know that large language models and all sorts of artificial intelligence programs including the training of these models including the retraining of these models is going to get a lot faster especially with nvidia's race to dominate the entire AI industry we could see rapid improvements in artificial intelligence so it's important to note that although these changes are groundbreaking in terms of a primary use level if we can get the hardware to increase the speed at which we can get things done we could see an increased rate of actual research and Discovery from drug Discovery as we just talked about to other machine learning research

### Robots [8:57]

then we had something that was really interesting sting in terms of Robotics and artificial intelligence as you know AI is eventually going to be involved in robots but so what we had here was something released by meta AI so if you don't know of course Mark Zuckerberg has a company and they do have a very robust AI division in that division they're constantly working on new things and this was something really cool so they demonstrate the ability to use your own language in order to talk to your robot and to get it to do very effective things so here you can see the user has said bring me the box of chocolates the cereal box the pill bottle and put them on the bedroom table I mean for the average person this might not pose any great application but can you imagine this in a care home in a home with a person who has specific disabilities or is unable to reach certain items this could most certainly provide a very useful application for those tasks and especially where sometimes you might not be able to type just being able to use your voice to a robot to be able to get you something afford understand you is going to be very effective now what's also cool about this is that in terms of the robustness the navigation policy is able to take an alternative route to avoid collision with humans in addition the pick policy is able to change the grasping location if the object moves of the base is not close enough to the receptacle and essentially what that means is that this kind of robot isn't just going to fall over if there is going to be some mild disturbance as you know life has its disturbances sometimes we might fall over objects might drop and these robots are able to adapt to their situation even if these things occur so take a look at some of these tasks because it's very likely that in the future maybe you might not see these robots in your home but maybe through our cities or maybe in certain workplaces robots that take natural language commands are going to be very common especially once we do manage to get the cost of these robots completely down it's going to be interesting to see everywhere that they could be then something from singularity next week the United Nations is set to host a summit in Geneva where Tech gurus are going to be discussing AI alongside 50 plus robots this is going to be extremely interesting as we are going to see The Wider conversation surrounding artificial intelligence it's important to know that viewers like you who do watch this channel are definitely more clued in on artificial intelligence than some of our world leaders and although that might seem like an outlandish statement as you can see from congress's grilling of some Altman this couldn't be further from the truth artificial intelligence is moving very quickly and sometimes people just don't have time to catch up so it will be interesting to see where the world's mind is on artificial intelligence especially with these robots as well then we had something about open Ai and

### OpenAI [11:44]

this one was also quite controversial you see open Ai and Microsoft were allegedly sued for three billion dollars over alleged Chachi BT privacy violations the lawsuit claimed the open AI secretly scraped 300 billion words from the internet the lawsuit which was filed on June 28th in federal court in San Francisco includes 16 Anonymous plaintiffs claimed that openai secretly scraped 300 billion words from the internet without registering as a data broker or obtaining the proper consent now this is something that is going to continually happen I mean Adobe seems like the only company who's trying to actually protect people's data when it's actually taking their content from the internet we know that mid-journey earlier this year did actually have a situation where people are trying to sue them because of course as you know mid-journey scraped millions of images from the internet in order to build their AI platform and they didn't pay anything to the creators of the original AI art that they stole it from so it will be interesting to see how this plays out in the court of law but I can assure you that open Ai and Microsoft have billions of dollars to fight this lawsuit and I will be wondering if this is going to go to a settlement or this will change the way that data laws are going to be created and how certain large language models are going to be created as you know currently many people just take the data from the internet as is and is it going to be now a requirement that when training a large language model you need to verify legally what kind of data it was trained on and where you got that data from that could have severe implications because if you have to go through an extended process where you need to verify all the data sources it could mean a huge slowdown because data collection is quite hard although it could also mean that data collection improves as you know when looking at large language models data is the number one bottleneck in language model performance so it will definitely be interesting to see how

### Drag [13:44]

this plays out then of course we had the very interesting software that is actually very unique in terms of what it does this software is called drag again essentially it's a software where you can manipulate any image by dragging a point on an image and moving it as if it kind of was a 3D model I mean this kind of technology is only possible with AI and I'm wondering what the true applications of this will be maybe Photoshop May integrate this into its software but it is still a little bit buggy but it does have quite a lot of applications for just manipulating an image in a way that we never thought was possible so Dragon I gotta be honest it looks really weird but at the same time it is very and whilst you're taking a look at some of the examples let me know if you're going to be trying out this Dragon because as you know lots of AI papers do get released a week and really often do we get the ability to try these out so with this it's actually very effective that we can now actually try this out and use it as well as we want to then we had another research paper extending the context window of large language models this work done extends like Lama up to 32 000 tokens essentially what you can do is you can have llama remember the last 32 thousand tokens in your messages which essentially means it gets more effective at processing larger volumes of text and this is what we want to see because as you know one of the main bottlenecks from the earlier versions of chat GPT and the early large language models is that they simply cannot remember what was said earlier on in the conversation which leads to a very limited function in terms of certain applications with this paper and with others that we've seen before this is going to prove very effective at allowing us to do much more from analyzing books to analyzing bank statements to analyzing poetry this is going to be something that we could all use very well and of course as you're seeing this is going to be widely used with gpt4 already having its 32 000 context window length already implemented