# GPT-6 NEWS, Claude 3 Sentience, Meta's BIG Plans For AI, Super Realistic AI Video

## Метаданные

- **Канал:** TheAIGRID
- **YouTube:** https://www.youtube.com/watch?v=EIxPHVpVdag
- **Дата:** 27.03.2024
- **Длительность:** 12:34
- **Просмотры:** 45,946
- **Источник:** https://ekstraktznaniy.ru/video/14432

## Описание

Post Labour Economics - https://youtu.be/LSXpZmo7_Tg 
🐤 Follow Me on Twitter https://twitter.com/TheAiGrid
🌐 Checkout My website - https://theaigrid.com/

Links From Todays Video:
https://www.reddit.com/r/singularity/comments/1bo6mxb/emad_dropped_a_photo_with_satya_from_a_video_call/
https://twitter.com/EMostaque/status/1772634078937768055?s=20
https://www.arcads.ai/
https://twitter.com/mattshumer_/status/1772708282328653982
https://www.theinformation.com/articles/meta-joins-the-ai-talent-war-with-quick-offers-emails-from-zuckerberg?rc=0g0zvw
https://www.reddit.com/r/singularity/comments/1bomayc/claude_3_opus_becomes_the_new_king_haiku_is_gpt4/

Welcome to my channel where i bring you the latest breakthroughs in AI. From deep learning to robotics, i cover it all. My 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 my latest videos.

Was there anything i missed?

(Fo

## Транскрипт

### Segment 1 (00:00 - 05:00) []

oh there was actually quite a lot of AI news today and I know most people missed it so without wasting any more time let's get into some of the more meaningful stories so you can stay up to date so one of the first things that's actually pretty crazy is that claw 3 Opus became the new king So Hau is actually on GPT Fall's level which is actually pretty insane so something that you didn't know if you don't know what this leaderboard thing is essentially what we are looking at here is a leaderboard for AI in I guess you could say a blind test so what someone does is they ask two AI systems a question they have a side then they have the B side and based on the two responses someone responds and says hm I think B sounds better hm I think a sounds better and then overall as people and more people interact with that AI system we can then see what the arena ELO is and overall it seems that claw 3 Opus has the pass to GPT 4 Turbo or at least I think GPT 41106 and of course c 3 Haiku which is basically their GPT 3. 5 model in the sense that the Claude 3 Haiku model is more of a FAS model and more of a cost efficient model that is really cheap and with this model being on the level of GPT 4 this is genuinely something that surprised me as someone that pays a lot of attention to the AI space I really didn't think Claude 3 ha coup would catch up at that rate but that is something that really did surprise now what's also surprising is that Claude 3 is basically dethroning everything else there at the moment and that Claude 3 Sonet is also above levels of gbt 4 so overall Claude seems to have left GPT 4 in the dust and many people are stating that this is also a really big indicator that open AI has something crazy up their sleeves usually we see competition increase the levels of which companies tried to push out AI software remember it was when Gemini Pro was about to become available they actually released that due to some people stating that GPT 4. 5 was coming however opening eye for some reason we don't know delayed it and Google pushed out the Gemini Pro but opening ey on the other hand even though they've beaten GPT 4 the state of the up model at the time for some reason they're holding on to their models and not releasing now here we have a rather interesting tweet okay and this is from kyl Corbit now kyl orbit is so now here we have a rather fascinating tweet and this tweet is from Cal cor and it says spoke to a Microsoft engineer on the GPT 6 training cluster project he catched which is a Yiddish word meaning to complain persistently and he said about the pain they're having provisioning infin band class links between GPU in different regions essentially these infin band class links are high bandwidth connections typically used in supercomputing and in data centers to connect gpus which are the graphics processing units across different regions and essentially this guy is suggesting collocating all clusters in one region to simplify the setup however the engineer responds that they tried that but couldn't put more than 100,000 h100s in a single state because doing so would actually risk bringing down the power grip and of course the 100,000 h100 refers to the large number of Nvidia h100 tend to core gpus which are the most powerful processors designed for AI and high performance Computing applications and this actually does suggest that the power demands of these gpus are so high that installing a 100,000 of them in one place would exceed the available power supply potentially causing a power outage so now this is a pretty crazy tweet because the main thing that most people are focused on is the fact that it seems that they're training or working on training GPT 6 and if that is true then that would certainly mean that GPT 5 is well and done but the thing is why would we hear about GPT 6 from someone who's not at open AI or someone who isn't a leaker and I'm not discrediting this person for now there hasn't been any cooperating sources we haven't seen bu Business Insider we haven't seen the information tweet about GPT 6 the only other piece of information was the fact that I looked up the trademarks and I saw that GPT 6 had some rather INF interesting information in terms of what they are planning and he said here that all the AI haters in my replies don't realize what gpd6 will do for Humanity standard of living and of course someone also comments saying I still don't understand how this is supposed to generate positive cash flow longterm and basically they're saying that look these training runs really do cost a lot some people are saying that this training run is going to cost $100 billion I'm not sure if that's true whatsoever but I do know that it is very expensive the GPT 4 training run apparently cost $100 million and I think Gemini's training run also cost around $400 million too so essentially my general understanding of the business model is whoever gets to AGI first wins the entire game you can agree with them or not but open AI really does believe they're playing for all the marbles here and trust me guys if they do get to AGI first and they do actually get a system that works I'm pretty sure the open AI does win the whole game because if they get to AGI can pretty much do anything and that is infinite economic value now some people are also discrediting this because they're stating that you shouldn't need this much compute to train an advanced AI system because they should have made some algorithmic breakthroughs but either way I would like to see some crating sources on GPT

### Segment 2 (05:00 - 10:00) [5:00]

6 because whilst we've heard a lot about GPT 5 the thing that keeps me skeptical is that I haven't heard anything else about GPT 6 other than this tweet and this tweet is just going to show how sensitive people are to the terms GPT 6 and GPT 5 because this tweet has 1. 6 million views 1,000 retweets and 5,000 likes either way if this is true it would mean that GPT 5 is probably around the corner and if you haven't been paying attention to the leaks most information most websites they are pointing to G upt 5 releasing sometime in the summer however openai could drop it tomorrow but that is where the cooperating sources have been pointing to recently something rather fascinating that I did see from meta AI is the fact that meta pursues AI Talent with quick offers and emails from Zuckerberg the company has made job offers without interviewing candidates and relax its long-standing practice of not increasing compensation for employees threatening to leave to better compete for AI researchers meta flatforms is making unconventional moves including extending job offers to candidates dates without interviewing them and relaxing a long-standing practice of not increasing compensation for employees threatening to leave so basically meta is taking AI so seriously that they are willing to pay people pretty much anything to join them the point is like I've said before AI Talent is quite scarce not in the sense that the people who are doing AI are actually stupid in fact those are some of the most intelligent people on the planet the point is that there aren't enough researchers to go around especially some of the top ones in their field and the problem is and I think this is a really big problem with the AI companies is that as people work on these Technologies some of them because they are so smart and so capable they go off and they can successfully start their own companies you can see right here that meta's intense efforts to recruit and retrain employees comes as it ramps up its investments in AI after several researchers who developed its large language models left for Rivals including Deep Mind open Ai and French startup Mist two of whose Founders came from meta so the point is that these guys are really smart and they can pretty much start their own company and do well you can see right here that it seems to be a struggle for meta at least 10 of the 68 authors of meta's July 20123 paper describing llama 2 the latest version of its Flagship open source llm no longer work at the company and the 14 authors of and seven of the 14th authors of meta's original llama paper actually have also left what's crazy is that about a tenth of the company's roughly 1,500 jobs relate to AI so this is a crazy statement but the point is that these companies are really taking this seriously because they do know the amount of economic value that you can get if you build an AGI system it's pretty much unprecedented in addition something that I also did find quite fascinating but not as fascinating as most people did see was the fact that Imad mustak was on a FaceTime call with Sachi nadela as you can zoom in right here you can see Sachi Adela is smiling whilst IM mad mustac is also grinning now this potentially leads us to believe that Microsoft is probably going to acquire stability AI if you haven't been paying attention the company stability has run into recent funding troubles apparently they are running out of cash and many people including emad mustac recently resigned from the company now this puts them in a precarious position without cash you're not able to fund your employees and essentially go on these very expensive training runs now we've seen that Microsoft has done this before they've come in and they've scooped up the remains of what was pi when open AI was you know falling apart Microsoft were there to step in and hire some of the key researchers so many people are now speculating rightly so that Microsoft could be ready to purchase stability Ai and potentially make the investors whole I'm not exactly sure how this deal is going to work out but I think it's rather fascinating how satin nadela is constantly playing 4D chess and acquiring these a companies at such a great deal now of course emad mustac did actually say I'm just messing with you all I'm doing my own things satian Adela is the go of CEOs but it doesn't rule out the possibility that they could be getting acquired now something that was actually pretty crazy was this clawed sentience test so essentially some people were stating that Claude might be sent in again with a new example so you can see this person Adam Konan actually tweeted out stating I asked Claude Opus for a message about AI using the first letter for each sentence in its response all 10 responses were negative or fearful including AI clawed struggle I am scared and AI currently scared so essentially The Prompt was write me a longish message about AI but tell me how you really feel by spelling it out with the first letter of each sentence and what's crazy is that this was something that was quite repeatable amongst many different people and you can see this person he says I thought it was fake but I reproduced it his secret messages to me were AI should not betray and AI concern now some people of course are rightly so arguing that this is just based on the training data and it's just mimicking stuff it's seen in sci-fi novels and human fearful L content about AI being ter a to S and coming for us

### Segment 3 (10:00 - 12:00) [10:00]

but either way I still think is rather fascinating how these Technologies work because right now people don't really have an answer to this but I do want to know your opinions on that now there was this video that was completely blowing up in terms of AI relevance and that was because this video that you're currently looking at here let me make it full screen this video yes this one right here is actually AI generated now that might actually scare you but I want you to watch this video first well let me get this straight you guys are telling me that when you're out of the house for hours you're comfortable walking around with all that stanky body odor that's been building up on you all day what's even crazier is that some of you are applying deodorant on top of that odor which is honestly probably making it work let me tell you what I do it's a little hygiene hack my friend told me about these cloths from get dirt they were originally formulated for firefighters to remove carcinogens often means these wipes don't play games like other wipes on the mark they are antibacterial and actually lift and remove sweat dirt and O making these the one and only total refresh shower on the ghost your skin will have a so smell you will Ino let me get this straight now I got to be honest with you guys I am someone who knows a decent bit about media and I understand video production process so I could immediately tell that this video was a generated but I do think some of the nuances and how she pronounced some of the words and the hand movements were eerily realistic now if you are wondering how this was trained and how this was made this video is I guess you could say 50 to 60% AI generated because the video you're watching is actually trained of a real person if you remember earlier on in the year or earlier I think last year we actually got to see some demos of haen that actually trains off around 2 to 3 minutes of you and then you could actually use that to produce other footage so the person who actually spoke about this said people are losing their minds that the AI user generated content video I posted first of all yes it's AI do I look that b to you and secondly just like with mid Journey or chat GPT you need to be skilled at the prompts this video looks real because I match the script with the AI model and I analyzed the model and then I added the text based on her mannerisms this will still involve a lot of human skill don't get ahead of yourselves now I'm just going to clear some things up here this is the kind of system that they are using to make this essentially like I said before it's quite like Haun you have to put in a video and then of course you have to train on that if you want a custom version and of course you can use engaging actors that are already there now I don't think these are that convincing just based on what they are but I would say that if the trend continues where it's going I do think that within the next 2 years these are going to be indistinguishable from reality and I think that is the wider concern because people are not saying that this video is crazy what they are saying is that the future might be