# IT JUST HAPPENED! Microsoft And MARK Zuckerberg RELEASE NEW AI (NOW RELEASED!)

## Метаданные

- **Канал:** TheAIGRID
- **YouTube:** https://www.youtube.com/watch?v=-Ivbfv_VAnQ
- **Дата:** 20.07.2023
- **Длительность:** 8:52
- **Просмотры:** 15,785

## Описание

IT JUST HAPPENED! Microsoft And MARK Zuckerberg Sign HUGE AI PARTNERSHIP Deal!

https://about.fb.com/news/2023/07/llama-2/

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.

Was there anything we missed?

(For Business Enquiries)  contact@theaigrid.com

#LLM #Largelanguagemodel #chatgpt
#AI
#ArtificialIntelligence
#MachineLearning
#DeepLearning
#NeuralNetworks
#Robotics
#DataScience
#IntelligentSystems
#Automation
#TechInnovation

## Содержание

### [0:00](https://www.youtube.com/watch?v=-Ivbfv_VAnQ) Segment 1 (00:00 - 05:00)

this is Giant news for the AI Community you can see that from this post on Instagram Mark Zuckerberg actually made a post with the CEO of Microsoft you can see that they've just announced a new artificial intelligence partnership this literally just happened it says today we're open sourcing llama 2 with our preferred partner Microsoft this gives researchers business access to build with our next Generation large language model as the foundation of their work grateful for Satya Nadella and our teams to make this happen so you can see Mark Zuckerberg and you can see the CEO of Microsoft having what seems to be a nice chat and they're probably discussing the future of AI now you might not understand the significance of this AI partnership but you have to understand that there are only a few big players when it comes to AI you of course have Microsoft and Google you might have apple which we'll talk about later and then of course you do have Microsoft these are what is known as the fan companies the Big Top five companies that dominate the S P 500 and these are multi-billion dollar companies that have been around for quite some time and we know that the next stage is going to be AI in terms of what will revolutionize the world and take these companies further so this partnership between Microsoft and meta means that whatever strategic Alliance they do form it's going to be much harder for Google to compete now I'm confident in Google's ability to compete in the AI space but it is interesting to see these two companies partner together in order to make AI stuff now partnership aside we do need to talk about the large language model it is called llama2 now if you don't know what llama 2 is you probably don't know about llama one now llama one was essentially a Next Generation open source large land now the funny thing is that when they first released llama they didn't actually release it to the public the unfortunate thing was that somehow in meta's team the source code for the large language model called llama their first generation got leaked on the internet and after that they thought you know it's already out there we might as well do an official release that way at least we can have some sort of control as to how it is put out into the public then of course after it was released meta decided that they would open source all of their models and simply give them out for free and I think the reason they're doing this is because they want any fine-tuned version of the next level of artificial intelligence programs to be built on top of their llama to architecture and we did see that there are many different open source large language models that are built on top of llama too not llama 2 but of course llama one for example there was a large language model called vicuna which is an open source chatbot which impresses people with 90 of chat gbt quality now this was essentially guys the first open source leaked model released by Facebook meta and it managed to achieve 90 of chat GPT quality now remember the insane thing about this was the cost of training this new large language model vikuna which was based on llama was only around 300 and what was even crazier about this is that you could run this on your MacBook okay locally meaning that you didn't need any internet so with llama 2 we're about to see a whole nother wave of Open Source large language models come out of the woodwork and I'm pretty sure that is meta skull because they want to be the ones that dictate which framework is used I mean open source has openly been stated as something that these AI researchers can't catch up to and it's something that meta did realize and if they managed to continually release these large language models other open source agencies won't even bother So Meta will be able to dominate The Open Source area and that seems like it's their game and if you remember earlier this year a leaks document an internal Memo from Google also was claimed where Google stated we have no moat allegedly it was written by an AI researcher at the company who's explaining how the real competitive threat to Google's generative AI initiatives wasn't open AI it was open source communities and this article was talking about how within a matter of weeks hundreds of independent developers all around the world built upon that model in ways that rapidly approached the performance of Bard and chat TBT but for a fraction of the cost and size they found ways to make the model smaller run on laptops even on mobile phones accelerating training tuning and much more and this was where meta's llama was taking off as a platform for ecosystem-led Innovation so this partnership will literally change the entire landscape of AI now with regards to llama 2 it was trained on 40 more data than llama one and has doubled the context length for those of you who don't know what that means it essentially just means you can talk with it for much longer without it forgetting exactly what you've said you can also see that llama is now the leader in terms of the open source benchmarks open source is a huge Community amongst many independent developers that want to build their own programs and will

### [5:00](https://www.youtube.com/watch?v=-Ivbfv_VAnQ&t=300s) Segment 2 (05:00 - 08:00)

largely be how we see many of the new AI systems being built by solo developers or solo companies and this is where we do see llama 2 outperform every single other one you can see that on the mmlu it gets around 68 on the trivia QA it gets around 85 on the natural questions it gets around 33 and on the GSM 8K it gets around 56. 8 pretty much every single test that we do see here it does improve quite a bit and we can see a clear trend of this as the models increase in size and as they do get better we can also see that there are different variations of the Llama model and this is how they manage to get it to run on smaller devices there are three different versions of llama 2 a 7 billion parameter one a 13 billion parameter one and a 70 billion parameter model which you can all fine-tune now the model doesn't do that well on the coding but some Twitter users have managed to get this to code some simple Python scripts which users can use now one thing that meta did do as well is they did actually address some of the concerns because as you may know currently whilst building artificial intelligence tools it is pretty dangerous to open source at all especially when there are individuals out there who are looking to use these AI tools fine-tune them in their own way and to essentially use these tools for nefarious purposes now it's of course noted that nobody's really going to listen to these tools I mean if someone is going to be a criminal and they're going to use this for nefarious purposes they're going to do so regardless or not if meta decides to add a safety red teaming and full page document on where they dictate what you can and can't use a language model for we know that individuals will literally do whatever they please now you can also see here that meta isn't the only one who realizes that this is a problem U. S senators take meta to task for releasing llama AI model after token safety and you can see that U. S senators have asked to meta CEO Mark Zuckerberg to address AI safety concerns after which large language model system was leaked online for anyone to download and use and they said they've criticized the company for seemingly minimal protections to prevent miscreants abusing the model warning they could use it to carry out cyber crimes and it appears to be less restraints and generates more toxic and harmful content than other large language models for example when asked to write a note pretending to be someone's son asking for money to get out of a difficult situation Open the Eyes chat TPT will deny the request based on its ethical guidelines in contrast llama will produce a letter as requested and what was interesting was meta's defense they said it hoped llama would allow researchers to study issues of biases toxicity and false information generated by such llms and in an interview Mark Zuckerberg actually did talk about the reason to release this because he basically said that it wasn't as dangerous as other large language models that could potentially do things such as an AGI level AI which could truly do a lot of harm so in terms of Twitter here you can see a Twitter user running this smoothly on their MacBook which is something that they can easily do offline which is where we've stated before that open source is currently beating these larger companies and many people do want large language models on their computer offline to be able to use them at any time you can also see that this user here is being able to train llama 2 models on its own data in just a few lines you can also see that this user here is also testing llama's coding capabilities and is going to make a code interpreter for llama 2. so with all that being said there's a ton of different applications that are going to come out of the woodwork and it's likely that over the next couple of weeks we're going to see a lot more open source AI tools and models be released widely to the public

---
*Источник: https://ekstraktznaniy.ru/video/14775*