Meta AI & Zuck are LEGENDARY for This! Llama 3 will 𝙖𝙘𝙩𝙪𝙖𝙡𝙡𝙮 "Shock the Industry"

Meta has released two new versions of its Llama AI model - Llama 3 8B and Llama 3 70B. These models represent a significant advancement in performance compared to the previous Llama 2 models. The Llama 3 models are larger and more capable, with the 70B version having 70 billion parameters compared to the 8 billion in the smaller 8B version. Meta claims these are among the best open-source AI models available
▼ Link(s) From Today’s Video:
TRY LLAMA 3 FOR FREE: www.meta.ai/?u...
Llama 3: llama.meta.com...
Llama 3 technical blog: ai.meta.com/bl...
Llama 3 Model card: github.com/met...
Request Access to Llama 3: llama.meta.com...
Zuck Chain Announcement: / 1780991652829507682
Jim Fan's Post: / 1781006672452038756
Matt Shumer's Post: / 1780994395128451358
► MattVidPro Discord: / discord
► Follow Me on Twitter: / mattvidpro
-------------------------------------------------
▼ Extra Links of Interest:
✩ AI LINKS MASTER LIST: www.futurepedi...
✩ General AI Playlist: • General MattVidPro AI ...
✩ AI I use to edit videos: www.descript.c...
✩ Instagram: mattvidpro
✩ Tiktok: tiktok.com/@mattvidpro
✩ Second Channel: / @matt_pie
-------------------------------------------------
Thanks for watching Matt Video Productions! I make all sorts of videos here on KZbin! Technology, Tutorials, and Reviews! Enjoy Your stay here, and subscribe!
All Suggestions, Thoughts And Comments Are Greatly Appreciated… Because I Actually Read Them.
-------------------------------------------------
► Business Contact: [e-mail]

[@MattVidPro]

OH Mistake on my part: You do not have to log in to use the new LLama 3 on the meta site, but to save chats you'll have to log in.

[@UniversalManifest]

and you have to login in aswell to create images!

[@LouisGedo]

👋 Matt. Off topic here but have you looked into ClipDrop lately? Now that subscribers have access to Stable Diffusion 3 in there, it's really REALLY worth the subscription price........and there's so many other nice tools ClipDrop subscribers get free access to. Quite a great deal, IMO........and I'm not in any way affiliated with Stability AI......I simply love sharing good news with other AI enthusiasts. 👍

[@timmyanimations8321]

@@UniversalManifest IT MAKES IMAGES?!

[@aidemalo]

What setup is required to run 70B models on gpu / cpu?

[@12uniflew]

Can you do a tutorial how to run this locally on our gpu's? id love to use it like NovelAi or AiDungeon but locally.

[@FriendlyVimanam]

Jokes apart, a heartfelt thank to mark for this, just credit where it's due. Thanks man.

[@armondtanz]

Hes building a bunker on sum island. He knows something we dont...

[@matheusazevedo9582]

Absolutely!

[@CharlesLeroy-mp8lk]

​@@armondtanz EXACTLY what I'm thinking. He may even be trying to hasten the END with AI. I really don't think we're getting something for nothing from him.

[@CamAlert2]

Why is it automatically a bad or suspicious thing if someone wants to be a prepper? Do you all not have emergency supplies?

[@soggybiscuit6098]

​@@CamAlert2for the average person, nothing. But for billionaires that runs in elite networks and information not privy to the average peasant, its worth noting

[@Zeptonixmusic]

The best AI Meta has ever produced is Mark Zuckerberg

[@masonmount17]

lol

[@Epicurious21331]

Or the worst

[@skits972]

I can tell he's ai without struggle

[@handler007]

trump is the worst ai criminal

[@tombradford7035]

Anything to do with Zuck is always a flop.

[@DragonBallDetroit]

Cant say we didnt expect every llm to reach gpt4 level after a year. Open ai needs to step up their game

[@conall5434]

Not even, gpt-4 is like 2 years old at this point

[@MattVidPro]

@@conall5434 nah it's barely over a year old

[@Fordtruck4sale]

@@MattVidPro And it's a lot bigger (probably) than 400B

[@Ajm549]

I have this on my WhatsApp meta ai that is.

[@DrCasey]

​@MattVidPro I think what he means is that the model itself was finished August 2022. The release date was barely a year ago as you said, yeah

[@AMR-bf8nx]

Mark is starting to being cool again…First Quest 3, now this.. Well done Mark.

[@MODEST500]

coz he is the older mark, Legacy mark the model who founded the facebook. this is not the newest version, I mean look at his hair. he is BACK!!.

[@wtalkie]

Zuck must give you good and free products to get your date and sell it. Remember, Facebook was good and free.

[@davehugstrees]

I'm SHOCKED daily from the SHOCKING news!

[@volcanowater]

I know right?, like SD3 API released yesterday and now LLama 3 is released today

[@bgill7475]

I'm being shocked right now and suffering from shock from being shocked that I'm shocked.

[@sorkoatbot7792]

OMG! You SHOCKED me with your comment, I'm on SHOCK right now because your comment SHOCKING me.

[@JorgetePanete]

You WON'T believe ***what*** SHOCKED me

[@BoltRM]

STUNNED!! And shocked! 😮

[@nowshinnur]

the lizard has done it again 🦎

[@MattVidPro]

All hail lizard mark

[@greenockscatman]

He's quickly becoming the good guy among the billionaires, isn't he? Not sure how I feel about that.

[@MW-dg7gl]

@@greenockscatman It really isn't hard to be a good billionaire...the standards are so low when there are only so few good billionaires among the hundreds.

[@J3R3MI6]

He’s trapping everyone in VR like Sword Art Online. It’s a honeypot.

[@yotubecreators47]

🤣🤣

[@LegendaryNuggets]

Wow Zuck is more human sounding now impressive meta!

[@antonystringfellow5152]

🤣

[@brexitgreens]

He has realised he's had a personal PR problem hampering his efforts. (An odd thing to say about someone more successful than everybody on Earth.) Now he's working hard to fix it. And I think it's working.

[@MODEST500]

its becoz the newer zuck has been discontinued. this is the legacy Mark, the original founder of Facebook, he even has his old hairs.

[@brexitgreens]

@@MODEST500 😂

[@HE360]

I think that Mark Zuckeburger at this point IS an A.I.

[@EddieAdolf]

Absolutely.

[@vi6ddarkking]

8k context is a good improvement but honestly I can't wait for more. Having the AI forget things mid conversation has been one of the most annoying things since day one with LLMs.

[@OliNorwell]

I'm playing with the 8B model right now using Ollama - so far very impressive! Feels similar to GPT 3.5. I gave it a short story and it summarised it very nicely, then I asked for a list of characters in the story in JSON format, again - did it perfectly. That was with the Q5_0 version, running ultra fast on my 4070Ti.

[@bossgd100]

And the 70b ?

[@chanpasadopolska]

​@@bossgd100and 70B is much larger so he did not set it on his computer - think

[@vi6ddarkking]

I wonder what the community will be able to do with this once they have had the time to fine tune it.

[@Ovyron]

I'm specially interested in the finetunes of the image generator part. This is really the other open source model that isn't based on Stable Diffusion, and it makes it look like a joke just with its generation of Mario and Shrek. I can't imagine what would happen if people finetuned and merged it like they did with SD1.5!

[@benjaminheim735]

wait for the 405 billion model dropping later this year, apparently its still undergoing training and its alr at 85 on the mmlu and its still a checkpoint

[@I-Dophler]

Fantastic breakdown of the new Llama 3 model! It's exhilarating to see how Meta AI is pushing boundaries with this open-source project. The potential for improved accessibility and innovation in AI development is immense. Looking forward to seeing how the community leverages Llama 3 for even more groundbreaking applications. Keep up the great work on covering these developments!

[@MattVidPro]

thank you!!!

[@I-Dophler]

Absolutely, @MattVidPro.

[@RraltKing]

chatgpt

[@I-Dophler]

@@RraltKing Noooo

[@ArnoldVeeman]

Meta AI isn't available yet in your country… 😢

[@brexitgreens]

First thought: VPN?

[@AscendantStoic]

Try downloading the model and running it locally or try a VPN.

[@BionicAnimations]

Matt's expression when Zuck is talking is priceless!😂

[@jayhu6075]

An important limitation point for us to know is: the Llama materials are restricted in their ability to improve other well-known language models. While developers often aim to refine models for better performance, the licensing terms limit the use of Llama materials, unless specifically using Llama. This highlights the importance of understanding licensing agreements when working with AI models. Overall, the license appears to impose some restrictions on the use of the Llama materials and emphasizes that this model is NOT entirely OPEN source.

[@bloxyman22]

Llama 8b doing a better job answering than chatgpt 4 did for you and it even corrected your spelling mistake in the prompt: Amelia Earhart, not Airhart, was the American aviator who disappeared during a flight around the world in 1937. There is a lot of speculation about what happened to her and whether she crashed on Nikumaroro Island (not Christmas Island) in the Phoenix Islands chain of Kiribati. Some researchers believe that Earhart and her navigator, Fred Noonan, may have landed on Nikumaroro Island after running out of fuel, but there is no conclusive evidence to support this theory. The U.S. government officially declared Earhart dead in 1939, but many questions about her disappearance remain unanswered. Christmas Island, on the other hand, is an Australian external territory located in the Indian Ocean, and it has no connection to Earhart's disappearance. It's possible that you're confusing the two islands, or that you've come across incorrect information. If you have any further questions or would like more information on the topic, feel free to ask!

[@CharlesLeroy-mp8lk]

The Zuck is probably trying to corner the future market. If Llama becomes very widely used then people may likely rely on the Zuck's software. I just don't know why else somebody like that dude would give away something like this for free

[@the_one_and_carpool]

not saying to much its because his silent partner aka part owner of meta wanted everything to be free shhh our secret

[@MrPablosek]

Rare lizard man W was earned today

[@comebackcs]

Amazing seeing these LLMs developing so fast. I only wish you would test coding with it

[@countofst.germain6417]

Gemini pro 1.5 is super legit, it is currently my favourite model, and ive been using these models for over 2 years.

[@santosic]

I do like it better for brainstorming, I find it gets pretty creative, plus it sticks to the persona I give it much better than the other two. For writing things Claude 3 remains my favorite one. For everything else, GPT-4 Turbo.

[@MonkeySimius]

Small incremental improvements are expected. I'm not sure why so many KZbinrs insist that it is shocking to anyone.

[@ScottzPlaylists]

👎 When I tried to Login with Facebook, it went into a loop where login failed because it goes back to the login page again, and you can't use chat after that. ❗ I had to start a private window to use chat, but don't log in there either, or it will loop back to login page, without chat again. Opening in a private window always resets the problem. ❗

[@electromigue]

I like that instead of doing synthetic benchmarks and test on "reasoning" (which llm's don't have, they just repeat the training or fine tuning) you just have fun with the tools like any of us would do. Keep it up Matt! Cool vid as always.

[@drendelous]

best way to feed more ads based on your interactions with ai

[@sharpvidtube]

And the AI will persuade you to vote for the candidate that gives Zuck the big bucks😂

[@SSLCLIPS-TV]

👇 Team Meta AI

[@RedEyeLazer]

Is it just me or am I not able to go to the new website where I can chat with Llama 3? It's not letting me in for some reason... Anyone else experiencing this?

[@DumPixels]

Same

[@endeavor4299]

Nope, works just fine for me. I just went to the Meta llama website and clicked on “Try Meta AI” in the navbar

[@madushandissanayake96]

what country are you from?It depends on the country

[@RedEyeLazer]

@@madushandissanayake96 US, man... 🤷 Are you able to enter the website just fine?

[@brianhopson2072]

Now just to wait for a llama 3 uncensored

[@duffthepsych]

I gotta say, this is really nice. I use gpt4 custom models for my podcast to generate show notes and blog posts based off my outlines, and chatgpt has been harder to wrangle recently. I pasted in the guidelines for what I want done into metaAI just now and it nailed each of the tasks on the first try.

[@poorman-trending]

When we can upload a photo of Zuck and have it make him not look like a robot lizard person i’ll be impressed

[@brexitgreens]

Yeah. Impossible. It's his inherent property. It's not him without it.

[@totempow]

My guess with the OpenAI GPT-5 thing... They see the writing as far as open source and have shifted focus onto other things that are more future-proof in a financial way. Such as SORA and the gaming industry. llama 3 on perplexity 8 and 70

[@Heather-kz7tn]

I am looking forward to the new open source models that will hopefully use decentralized sources/resources instead of just Google and Bing. Also, the "Legendary redemption arc" comment lmao

[@I-Dophler]

Wow, this is absolutely groundbreaking! Meta AI truly leading the way in open-source AI advancements. Can't wait to see what people create with this.

[@captainmorgan6748]

Incredible is an understatement. It is probably the best thing i have ever seen.

[@starblaiz1986]

Holy.. the Lizard man actually went and did it! Bro, I legitimately can not understand how we ended up in the timeline where Meta of all companies have become the champions of opensource AI. 😮 Like seriously, who went back in time and stepped on a butterfly?? Give me their names! 😅

[@keithmerrington9026]

I know right, it feels bizarre. I'm still expecting some sort of catch.

[@martin22336]

Lol am Running that 70b version today lad today and its crazy

[@dylan_curious]

I would love to know the story behind why he’s wearing that chain? Maybe a gift from a friend? Or do you think just a style choice he made for himself?

[@AscendantStoic]

A big win for open source AI!

[@Trendilien69]

I agree ! 2024 is shaping to be very exciting ! Thanks Matt !

[@nathandscott1]

Points for Meta, been my least favourite of Big tech but starting to see them being good for the community

[@tombradford7035]

*Meta AI isn't available yet in your country* - says it all - I'm in England, now some backwater village in the back of beyond.

[@CrystalBreakfast]

Eh, context length is something the community unfortunately hasn't been able to increase significantly with open-source models. One of the biggest limiters is that consumer hardware can't run high-token models at a price that's anywhere near affordable. It's unfortunate, but I'm sure in the not-so-distant future we'll get much more affordable hardware, given recent scientific breakthroughs.

[@MovementFAV]

Thx Zuck for making it Open Source!

[@K.Khalil]

it's not available in UK

[@frankthiele6539]

website not working in germany, but i can download the models... even with a fake email and without a facebook account

[@FRareDom]

this is basically like stable diffusion, started out pretty humble with llama 1, and now game breaking with llama 3, similar to stable diffusion 1 and 3

[@TheColonelJJ]

Which LLM, that can be run on a home computer, would you recommend for helping refine prompts for Stable Diffusion -- text to image?

[@jeffwads]

You don't need a startup to run the 70b 8bit. Running it quite well on my system at home.

[@TBKJoshua]

No matter what you think about Mark, you have to admit he's been a pivotal player in this AI/AGI race.

[@justinwescott8125]

Does open source mean we will be able to get an uncensored, unaligned version of it at some point?

[@alvaroluffy1]

Meta should be called OpenAI because its open and OpenAI should be called Meta because its meta to call it that

[@nightmisterio]

This reminds me that the creator of Telegram said that gov agents wanted them the team to use certain open source componentes, I think because it has backdoors. Not that I am saying this would be the same but... who knows...

[@SirajFlorida]

I'm just saying... This man just gave the world millions of dollars worth of software, and you call him names and make fun of him. You called him lizard man. You should apologize to him. Put the shoe on the other foot.

[@EnterPlayMode]

just found Llama 3 days ago. 💪🏾

[@MODEST500]

Like a good lizard he is just shedding his old evil skin, I hope his new skin will be fresh and open source, instead of data theft and closed source. also he definitely looks like the older mark I used to know from my GK books when he founded the FB. seems like he grew some hairs or maybe the older version took over. i guess the newer version was melted in the factory.

[@automaton111]

So now Meta will be able to track all of our AI usage and link it back to our other social media activity. Great

[@ChessMasterNate]

Given the other comments, I am lucky it worked for me at all. I did it, not signed in. Don't know if that helps anyone.

[@RpgBlasterRpg]

It's over for Claude 3 and ClosedAI

[@chanpasadopolska]

I use Gemini from time to time, especially because it's super responsive

[@sharpvidtube]

Had a look, not available in the UK.

[@TheMCDStudio]

Second biggest thing Meta did, after creating the Metaverse.

[@nathanbanks2354]

Not sure if I'm more excited for Mixtral 8x22b & its derivative WizardLM-2, or the new Llama 400B model. I think WizardLM-2 will be much faster/easier to run, but hopefully they'll both push the state of the art. (Easier to run means there's a chance it'll run on a $5/hour computer with 4 4090's instead of a $35/hour computer with 4 H100's.) The small Llama 3 model looks really good, though. Hopefully ollama will support them all soon.

[@MilesBellas]

Mark needs to empower StabilityAi.

[@bfreecity]

A) the perm, the chain…Zuck is turning into the uncle that sings cringe karaoke at his nephew’s bar mitzvah. B) being giddy about these companies barreling forward with open source AI is even more cringe. At what level of intelligence are we going to agree to cap these systems so every nihilist online can’t use them to instigate chaos. People can’t even behave in the comments section….The danger is approaching far quicker than regulations will be able to deal with and it’s pretty terrifying, IMHO

[@бесконечности]

Amazing, thank you!!! I'm waiting for this miracle to start understanding Russian. Or rather it understands, gives out information, but then immediately deletes what was written and there is an inscription: I apologize, I do not understand Russian yet, as soon as I begin to understand, I will let you know. But how can he not understand, if he answered the question, but then erased his own text.... Потрясающе, спасибо!!! Жду, когда это чудо начнет понимать русский язык. Вернее понимает, выдает информацию, но потом сразу удаляет написанное и появляется надпись: Я прошу прощения, я пока не понимаю русский язык, как только начну понимать, я вам сообщу. Но как он может не понимать, если он ответил на вопрос, но потом стер свой собственный текст....

[@amj2048]

It's crazy to think that GPT-4 is just over a year old now, it feels like it was always there.

[@ronniebasak96]

Wait, Gemini Pro 1.5 got 11-shot prompting at GSM8K and Claude 3 got 92.3 at 0-shot when the row header is 8-shot. Why?

[@Yipper64]

I ran meta AI through my usual Christian Videogame test and it actually does *fairly* well. It isnt not woke. I can tell it's still got that progressive slant, but not quite as blatantly as some others (namely chat gpt). But it did follow my request to keep the christian themes subtle very well. It is *at least* on the level of claude. Im excited for an uncensored version, tends to weed out the political bias. As for the images I just ran it through my melting spongebob comic and it is surreal as usual. It doesnt actually hold to just four panels. Nor does it melts spongebob but you get melty things, and some text bubbles. Its obviously incoherent and dublicates spongebob but not terrible.

[@Cute_Maxi]

You dont need to be a startup to run the 70B Llama 3 model... you just need an apple silicon mac with 32gb of unified memory, it fits. Even that 400B version might be able to run on a macbook with 128gb of unified ram... maybe, it will be close. Its crazy how useful the unified ram is for running LLMs... even my 24GB Macbook Air is running models that previously could only run on RTX 4090's all because the unified ram is entirely accessible to the GPU... I can load in about 22gb into my 24gb of available ram, with the Metal GPU layers set to max before Mac OS crashes lol. Apple is the way to go if you are looking to run large token count models... My next macbook will be a pro with the maximum amount of ram, I dont care if it costs as much as a car, its still a fraction of what it costs for Nvidia datacenter GPUs.

[@FusionDeveloper]

If your Real Time Image Generation stopped working: It gets disabled when you get banned from it. Every time it generates a blurry image, that counts as a violation. If you have a blurry image and you keep typing, then every letter you type, will count as a separate violation. Get enough violations and they ban you from the real time image gen. Even the 2 letters "sh" is a banned thing and many other strange things. So if you are typing a word that starts out with a banned word, it's counting towards your violations. After 24 hours it is likely to unban you, but I don't know if the next time is longer and if its like bing create where 3 strikes and you are banned for life.

[@reezlaw]

Is it me or the Zuck has never looked as human as he does in this video

[@skillsandhonor4640]

Video2Game: Real-time, Interactive, Realistic and Browser-Compatible Environment from a Single Video

[@Yipper64]

"the most intelligent AI assistant that you can *freely* use" since gpt 4 is behind a paywall im assuming that means they think its the best *for a free AI,* but not necessarily better than GPT 4. I wonder if it beats Claude?

[@bobbykingAiworld]

Epic video, that’s so cool. 🎬🆒

[@ToastMusic]

i only freakin watched that because Meta AI website isn't available here in France

[@RollingDude.]

Where i live it is also unavailable but i been testing it using a VPN (free one). Meta ai is a bit disappointing tbh

[@fabiankliebhan]

Zuck is so back!

[@Dina_tankar_mina_ord]

He really tried to make himself look real. with the hair and the more flushed face. I dont want to be a jerk but I guess his position probably is extremly stressfull and would make most of us look semi alive.

[@levonkenney]

wait openai didn't drop any news today? usally when something big like this happens open ai drops some news trying to seem cool

[@jopansmark]

It's so over for OpenAI. They should sell the company to someone who can do business and research, like Lenovo or Samsung before it's too late.

[@635574]

Also theres zuck interview on a sub 100k channel released 2h ago.

[@brianhopson2072]

Sam Altman was on a podcast 2 days ago that had only hundreds of views and subs.

[@capturedbyfabian]

All the current Llama 3 quantizied models talk to themselves to the point they stay in an endless loop

[@babbagebrassworks4278]

Shocking

[@RikkTheGaijin]

Not available in my country... Ouch...

[@DorianRodring]

I just got here. Is this Llama 3 already integrated into the META apps or is it dropping soon?

[@anewman1976]

Of course we have Meta HQ here in Ireland and I have to use a VPN to use this! 😅

[@fan354]

The website won't even work for me, it keeps sending me back to the main page and won't give me access at all

[@chrissteele3975]

I just had a convo with Meta Ai as part of the new Messenger app pretty cool

[@DumPixels]

Pretty sure that is not a new feature

[@saikanonojutsu]

I'm sure it's expensive but it looks like he's wearing the paperclip necklace you made out of the secretary's paperclip stash while you were waiting to get yelled at by the principal for taking the kid with weed allergies out into the weeds where he had a wild rashy reaction to the weeds.

[@SonOfTamriel]

You don't need a facebook account, thankfully

[@phen-themoogle7651]

I tried it in Japanese and it isn't on the level of Gemini1.5 or GPT4 , or Claude3 Sonnet, still gotta get it up to par with those for languages, which is the main reason I use them in the first place. But congrats to everyone who will be using it monolingually in English or for other tasks its capable of doing! ☆

[@santosic]

Has the new Japanese version of GPT-4 Turbo been useful for you in this case?

[@drlordbasil]

God I love all of us fangirling so hard xD

[@NOPEFallacy]

This is crazy!!

[@WINTERMUTE_AI]

LOVE IT!

[@flareonspotify]

Meta is the only one left Opensourcing Ai

[@jayhu6075]

It is still not open: With a limitation that restricts the potential of Llama materials to enhance other major language models. Researchers and developers often want to compare or fine-tune different models to improve their performance or tailor them to specific task. However, due to the restrictions in the licensing terms, they cannot freely utilize the Llama materials to do so unless they specifically use Llama 3.

[@agnosticatheist4093]

Respect to Mark ❤

[@dorukbebek]

grate job meta

[@rionix88]

Is it creative enough to create a short story or long novel

[@nikozeldenthuis281]

Not available in my country..

[@rolestream]

When I tried to web search it told me its cutoff is December 2023. Hmmm

[@zdenekburian1366]

on chat.lmsys, it told me the cutoff is december 2022