/
Navigation
Chronicles
Browse all articles
Explore
Semantic exploration
Research
Entity momentum
Nexus
Correlations & relationships
Story Arc
Topic evolution
Drift Map
Semantic trajectory animation
Posts
Analysis & commentary
Pulse API
Tech news intelligence API
Browse
Entities
Companies, people, products, technologies
Domains
Browse by publication source
Handles
Browse by social media handle
Detection
Concept Search
Semantic similarity search
High Impact Stories
Top coverage by position
Sentiment Analysis
Positive/negative coverage
Anomaly Detection
Unusual coverage patterns
Analysis
Rivalry Report
Compare two entities head-to-head
Semantic Pivots
Narrative discontinuities
Crisis Response
Event recovery patterns
Connected
Search: /
Command: ⌘K
Embeddings: large
TEXXR

Chronicles

The story behind the story

days · browse · Enter similar · o open

Meta details Llama 3: 8B- and 70B-parameter models, a focus on reducing false refusals, and an upcoming model trained on 15T+ tokens that has 400B+ parameters

What To Know About ‘Llama 3Model Marcus Gopolang Moloko / Memeburn : Meta AI with built in Llama 3 is on WhatsApp in South Africa Hamsat Abdurasheed / News.ng : Meta releases Llama 3, new AI assistant to rival OpenAI Jenna Barron / SD Times : Meta releases the first two Llama 3 models Luke Jones / WinBuzzer : Meta's Ambitious Rollout of A.I. Assistants in Instagram, WhatsApp, and More Temitayo Jaiyeola / Businessday NG : Explainer: What Meta AI's rollout means for WhatsApp, Instagram users Milan Stanojevic / Windows Report : Meta AI got powered up by Llama 3, now available in other countries Rafly Gilang / MSPoweruser : Meta's upcoming Llama-3 400B model could potentially beat GPT-4 Turbo and Claude 3 Opus Rael Hornby / Laptop Mag : Meta AI isn't the next ChatGPT, but it could be the next Threads Harsh Shivam New Delhi / Business Standard : Meta goes after OpenAI, Microsoft, Google with Llama 3 AI model: Details Mini Games / The Indian Express : Meta AI is the ‘most intelligent, freely-available’ AI assistant, says Mark Zuckerberg The Hindu : Meta introduces Llama 3 large language model; aims to make it multilingual soon Chris Smith / BGR : Meta AI assistant has entered the chat with claims of more advanced intelligence Eric Hal Schwartz / Voicebot.ai : Meta Augments Generative AI Assistant With Llama 3 LLM and New Features Kurt Robson / Verdict : Meta releases new AI model as GenAI race heats up Vallari Sanzgiri / MediaNama : Meta Unveils Llama 3: Here's All You Need to Know and Consider about the LLM James O'Brien / LBC : Meta unveils latest AI model as chatbot competition intensifies Gabriel Budi Sutrisno / Tech in Asia : Meta launches first models of its latest Llama 3 LLM Sonia Majumder / Campaign Middle East : Meta rolls out advanced AI chatbot to all of its apps Chinmay Dhumal / MySmartPrice : Meta AI Powered By Llama 3 Introduced; Rivals ChatGPT and Gemini Chris McKay / Maginative : Meta Launches Standalone AI Chatbot Website Sukriti Gupta / Analytics India Magazine : Meta Forces Developers Cite ‘Llama 3’ in their AI Development M.G. Siegler / Spyglass : An AI in Every Pot Jorge Alonso / BroBible : Dripped Out, Bearded Mark Zuckerberg Is Taking The Internet By Storm Hindustan Times : Meta unveils Llama 3 and real-time image generator: All you need to know Marty Swant / Digiday : Meta AI rolls out several enhancements across apps and websites with its newest Llama 3 Anthony Garreffa / TweakTown : Meta launches early versions of Llama 3 to fight with OpenAI and Google in the AI battle Roberto Nickson / Roberto Nickson on YouTube : Mark Zuckerberg on Fighting in the UFC, Meta AI, Legacy, Metaverse & More Rushil Agrawal / Android Authority : Meta's supercharged AI assistant is taking over its apps across the world Johanna Romero / PhoneArena : Meta launches new version of Meta AI assistant across WhatsApp, Instagram, Facebook, and Messenger Jason Nelson / Decrypt : Meta Pushes AI Chatbot Into Instagram, Facebook, WhatsApp International Business Times : Meta introduces most capable Llama 3 AI model Reuters : Meta steps up AI battle with OpenAI and Google with release of Llama 3 Kelvin Munene Murithi / CoinGape : AI News: Meta Introduces Llama-3, Touts It as Top Open Source AI Model Akash Pandey / NewsBytes : Facebook, WhatsApp, Instagram now offer Llama 3-powered Meta AI chatbot PYMNTS.com : Meta AI Assistant Adds Website, Expands Beyond US Andrew Hutchinson / Social Media Today : Meta Brings Advanced AI Chatbot to All of its Apps Alex Heath / The Verge : Q&A: Mark Zuckerberg on winning the AI race Michael Nuñez / VentureBeat : Elon Musk's ‘not bad’ review thrusts Meta's Llama 3 AI into spotlight Imad Khan / CNET : Meta AI Tool Now Driven by New, More Powerful Llama 3 Model, Meta Says Andrew Heinzman / How-To Geek : Meta Launches Llama 3, Crowns Itself King of AI Brendan Frye / CGMagazine : Meta Releases Llama 3 In The Latest Move In The AI Arms Race Joe Hindy / PCMag : Meta Upgrades AI Assistant With New Llama 3 Model Maria Deutscher / SiliconANGLE : Meta debuts next-generation Llama 3 LLM series and new chatbot features Don Reisinger / ZDNet : How Meta's Llama 3 will be integrated into its AI assistant Richard Lawler / The Verge : Wait, Mark Zuckerberg's a chain guy now? Kyle Wiggers / TechCrunch : Meta releases Llama 3, claims it's among the best open models available Kurt Wagner / Bloomberg : Meta Debuts Llama 3, Latest AI Model for Powering Chatbots Ryan Heath / Axios : Meta releases new AI models and updates AI assistant Alex Kantrowitz / Big Technology : How Meta Built Llama 3 Financial Times : Meta unveils less ‘sanctimonious’ AI model as tech rivalry heats up Siddharth Jindal / Analytics India Magazine : Meta Llama 3 Now Available on Microsoft Azure Threads: @georgescriban : I don't really understand what Meta's doing here.  Why invest in a consumer-facing generative AI capability, when the value of AI to Meta is in its application against the company's hard problems (safety, infosec, data protection, fraud, growing its ad business, etc)? … Chris Welch / @chriswelch : I don't need to ask questions to Instagram or WhatsApp.  This is the stupidest goddamn timeline. Shira Ovide / @shiraovide : I still don't understand why it's useful to most people to have an AI chatbot in Facebook, Instagram and WhatsApp.  Do I want to ask in Facebook for help crafting a sensitive email to my boss?  Or ask a WhatsApp chatbot for restaurant recommendations?  Or use an AI chatbot in the app I use to look at delicious croissants? Jon Porter / @jonporteruk : @crumbler I'm a little sceptical of this given the regional availability of Meta's AI stuff at launch (I don't have access in the UK for example).  I agree it has the potential to lead in distribution, but I struggle to believe it's there just yet https://blog.whatsapp.com/... Mark Vickers / @mark_r_vickers : I suspect people are getting the Llama 3 story wrong.  Sure, Meta will experiment with jamming it into social media, but that's not the main point of it.  The main point is to create an open source model that will be both powerful and compressed enough to work with the new generation of smart glasses Meta is designing.  This is not about synthetic social media.  It's about having a mobile AI agent with you wherever you go.  Mark Zuckerberg wants to be Steve Jobs, and he might even pull it off. Siva Vaidhyanathan / @sivavaidhyanathan : I have been trying to figure out why OpenAI release such a bad product and why Meta et al have been rushing to market.  I don't think it's about making news or pandering to traders and analysts.  Real analysts know the level of R&D going on at Alphabet and Meta. … Mark Zuckerberg / @zuck : More details on Llama 3.  Today we're releasing 8B and 70B parameter models — both best-in-class for their size.  We've got more releases coming to bring multi-modality and longer context windows.  We're also still training a larger dense 400B+ parameter model. Hamza Mazari / @hamzamazari88 : Wow, making Llama 3 open source is a huge gift to the AI community, Meta is leading the way Nick Clegg / @nickclegg : Think of this as the first wave of Llama 3 - we're excited to get these in the hands of consumers and developers to help kickstart the next wave of AI innovation and we'll have more to share soon. Soumith Chintala / @soumith : Llama3 8B and 70B are out, with pretty exciting results!  * The ~400B is still training but results already look promising.  * Meta's own Chat interface is also live at https://www.meta.ai/ * TorchTune integration is shortly going live: https://github.com/... Akarshan Biswas / @abmlllm95 : Llama3 8B and 70B are now available on HuggingFace.  I have applied for the access. Dwarkesh Patel / @dwarkesh : . @zuck on: - Llama 3 - open sourcing towards AGI - custom silicon, synthetic data, & energy constraints on scaling - Caeser Augustus, intelligence explosion, $10b models, & much more Enjoy!  Links below Salvador Rodriguez / @salrodriguez : Meta today is releasing the first versions of Llama 3.  With this new AI model, the company will also be upgrading its Meta AI tool, launching it in more markets and featuring it more prominently across its apps https://www.wsj.com/... with @joannastern X: Andrej Karpathy / @karpathy : Congrats to @AIatMeta on Llama 3 release!!  🎉 Notes: Releasing 8B and 70B (both base and finetuned) models, strong-performing in their model class (but we'll see when the rankings come in @ @lmsysorg:)) 400B is still training, but already encroaching GPT-4 territory (e.g. 84.8 MMLU vs. 86.5 4Turbo)... Aston Zhang / @astonzhangaz : Llama 3 has been my focus since joining the Llama team last summer. Together, we've been tackling challenges across pre-training and human data, pre-training scaling, long context, post-training, and evaluations. It's been a rigorous yet thrilling journey: 🔹Our largest models... [image] Dave Winer / @davewiner : Quick demo of the new https://meta.ai/ product from the company that makes Facebook. This is just one feature, its ability to recalc drawings as you edit the prompt that creates the picture. Very nice. As you can tell from the video I really like it.... Tom Warren / @tomwarren : Meta's chatbot is going about as well as you might expect 🙃 https://www.404media.co/... [image] Sasha Luccioni, PhD / @sashamtl : So LLaMa 3's carbon footprint is... huge? 🤯 They estimate it to be 2,290 tons of CO2eq, compared to 550t for training GPT-3 and 66t for training *all* of the BLOOM models (1B-176B) 🌬️ [image] Max Woolf / @minimaxir : @alexeheath 99% chance the next version of Grok will be a finetuned Llama 3 70B model. Aaron Levie / @levie : Incredible to have Meta out here pushing open source AI forward in such a major way. An amazing moment to be building software when you have $100s of billions of dollars from the major tech players going into advanced R&D and infra that directly benefit you as a developer. Dwarkesh Patel / @dwarkesh_sp : Zuck on: - Llama 3 - open sourcing towards AGI - custom silicon, synthetic data, & energy constraints on scaling - Caeser Augustus, intelligence explosion, bioweapons, $10b models, & much more Enjoy! Links below [video] Kache / @yacinemtb : just turned back on all my data collection preferences for facebook & instagram. didn't realize it was going to such a good cause Andrew Ng / @andrewyng : Meta released Llama 3 on my birthday! 🎂 Best present ever, thanks Meta! 😀 @intelai : AI isn't just growing; it's evolving rapidly. Llama 3's leap in reasoning and performance is thrilling, and we're ready to support it with our Gaudi accelerators and Xeon CPUs. Looking forward to the new possibilities this will unlock, @AIatMeta! Joelle Pineau / @jpineau1 : We just released Meta Llama 3: the most capable openly available LLM available to date! The 8B & 70B models are out now, and we expect to release models with larger context windows, additional model sizes and more capabilities in the coming months. Ethan Mollick / @emollick : “Llama 3, Write an office memo justifying the fact that management has been replaced by squirrels. We are all being paid in acorns, now.” Nice job. [image] @intel : We're thrilled to have validated our #AI product portfolio for the Meta Llama 3 8B and 70B models! This is another example of our commitment to the latest AI innovations. Check out the link to learn more: https://www.intel.com/... @aiatmeta : You can download the models directly from Meta today, additionally though, working closely with our partners, Llama 3 models will be available soon across @AWSCloud, @Azure, @Databricks, @GoogleCloud, @HuggingFace, @IBM WatsonX, @Kaggle, @NVIDIA and @SnowflakeDB. Joelle Pineau / @jpineau1 : Congratulations @Ahmad_Al_Dahle + the entire GenAI team + all of the incredible XFN across @AIatMeta that came together to ship these new models today! 👏 @amd : We're excited to work with @AIatMeta on Llama 3, the next generation of their open-source large language model. As a major hardware partner of Meta, we're committed to simplifying LLM deployments and enabling outstanding TCO. Soumith Chintala / @soumithchintala : Llama3 8B and 70B are out, with pretty exciting results! * The ~400B is still training but results already look promising. * Meta's own Chat interface is also live at https://www.meta.ai/ * TorchTune integration is shortly going live: https://github.com/... Naman Goyal / @namangoyal21 : Really proud of the work that went into making this possible, hope this helps the community push the field forward. Also in case anyone missed it, there's a sneak peak of what to come next at the end of blog post https://ai.meta.com/... Roshan Sumbaly / @rsumbaly : Hello Llama3! The first drop (8B, 70B) that we're doing for the community. Still baking the 405B model and also have more systems and capabilities (multimodality, multilingual, tooling) lined up to share. Also thinking of Llama 4 already. We're just getting started! Elon Musk / @elonmusk : @ylecun Not bad Ahmad Al-Dahle / @ahmad_al_dahle : It's here! Meet Llama 3, our latest generation of models that is setting a new standard for state-of-the art performance and efficiency for openly available LLMs. Key highlights • 8B and 70B parameter openly available pre-trained and fine-tuned models. • Trained on more... [image] Tobi Lutke / @tobi : Llama3 looks fantasic. Thank you and congrats @Meta Open source models are the key platform for innovation around productivity, safety, and innovation https://llama.meta.com/llama3/ Quentin Garrido / @garridoq_ : Fixed [image] Dare Obasanjo / @carnage4life : Not sure what's going harder, the Meta AI and Llama 3 announcement or the drip ⛓️ [video] Leo Kelion / @leokelion : Meta has launched Llama 3 and with it a ChatGPT rival with a standalone website... that isn't yet available in the UK. Hopefully not too long before we can try it out. Wonder if @perplexity_ai will gain access? [image] Ethan Mollick / @emollick : As a reminder, GPT-4 class models out-innovate and out-persuade the average human. An open sourced model of that power is going to lead to lots of unanticipated effects, good and bad. Tay-I / @taylorbuley : 2x24k clusters of GPUs to train Llama3. 50k GPUs all working on the singular goal. Kind of beautiful. Gabriel Garrett / @gabgarrett : looks like Llama 3 actually beats out older versions of GPT-4. it's pretty big that there's an open-source version of a model at that level now. llama3 400b will be the moment of truth for beating out GPT-4-04-09 though Maxime Labonne / @maximelabonne : Llama 3's vocabulary size is much bigger (32000 => 128256). Also note 11008 => 14336. [image] Nathan Lambert / @natolambert : Diff of llama 3 license to llama 2: Mostly around sharing, built with llama 3 branding, agree to meta brand guidelines for distributing trademark Some minor other differences [image] @togethercompute : We are thrilled to be a launch partner for Meta Llama 3. Experience Llama 3 now at up to 350 tokens per second for Llama 3 8B and up to 150 tokens per second for Llama 3 70B, running in full FP16 precision on the Together API! 🤯 https://together-ai.webflow.io/ ... [video] Rohan Paul / @rohanpaul_ai : What's more crazy is the 400bn Llama 3, which is still in training, and already, with its current checkpoints, the scores look absolutely BRILLIANT. And this is Meta's statement on that. “Our largest models are over 400B parameters and, while these models are still training,... [image] Simon Willison / @simonw : You can run prompts to rough the new Llama 3 models on Replicate: https://replicate.com/... for 70b-instruct and https://replicate.com/... for 8b-instruct Charlie Holtz / @charliebholtz : We have Llama 3 running at 55 tokens/second on https://llama3.replicate.dev/: [video] @replicate : Llama 3's out. Try it here: https://llama3.replicate.dev/ @aiatmeta : Introducing Meta Llama 3: the most capable openly available LLM to date. Today we're releasing 8B & 70B models that deliver on new capabilities such as improved reasoning and set a new state-of-the-art for models of their sizes. Today's release includes the first two Llama 3... [video] Rohan Paul / @rohanpaul_ai : Great results of Llama-3 Human evaluations across against Claude Sonnet, Mistral Medium, and GPT-3.5. [image] @deliprao : The much-awaited Llama 3 release is here! https://llama.meta.com/llama3/ My quick notes: [image] @sullyomarr : Wait a second.. llama3 beats both sonnet and gemini pro 1.5 This is actually huge. Open source models are back [image] John McBride / @johncodezzz : Huge: @fb_engineering releases llama3 - “The most capable openly available LLM to date” The future of AI technology is truly open! A few thoughts 🧵 https://ai.meta.com/... Nathan Lambert / @natolambert : llama 3 8b has the some approx amount of training compute as llama 2 70b lol (15T tokens training for v3, 2T for 2) [image] Andrew Curran / @andrewcurran_ : ‘Llama 3 is pretrained on over 15T tokens that were all collected from publicly available sources’ The paper will not be released today. [image] Naveen Rao / @naveengrao : Congrats @AIatMeta team on the Llama3 release! This is a huge milestone for AI and the market...the performance of these models are a huge step compared to a year ago and give users so much more flexibility on how they want to use, deploy, and customize AI Jared Zoneraich / @imjaredz : Llama 3 is out. You can't ignore open-source models anymore. It's supposedly better than Claude 3 Sonnet 🤯🤯 Huge [image] Nathan Lambert / @natolambert : we're all following the way llama 3 400b is going [image] @_borriss_ : Llama 3 by Meta is out. Looking strong. Full details here: https://llama.meta.com/llama3/ [image] Kayla Cardillo / @kaylacardillo : LLaMA-3 official release is said to be scheduled for today. Incoming: 8 billion parameters/70 billion parameters, 8,000 context length, 128,000 vocabulary, and 15 trillion tokens. The increase in parameters means the model will be able to learn more complex patterns and... Max Weinbach / @maxwinebach : Llama 3 is live from meta and benchmarks make it look like a great model! 8B and 70B dense models But man that 8K context window is rough given we're seeing 32K is similar size models already https://llama.meta.com/llama3/ Arun Rao / @rao_hacker_one : Thrilled to share a preview of Llama3, including the release of an 8B and 70B (82 MMLU, should be the best open weights model). Also preliminary results for a 405B model (still training, but already competitive with GPT4). Much more to come... https://llama.meta.com/llama3/ Elad Gil / @eladgil : Llama 3 out from Meta - https://llama.meta.com/llama3/ [image] Nathan Lambert / @natolambert : llama 3 8b has the some approx amount of training compute as llama 2 70b lol (15T tokens) [image] Andrew Curran / @andrewcurran_ : Official release benchmarks. [image] @xlr8harder : Llama 3 is out, but it seems that still only @MistralAI really has our back: it still has downstream usage restrictions of model output. @AIatMeta this is a garbage restriction that harms open source and I was really hoping you would see the light this time. [image] Mike Lewis / @ml_perception : Excited to share a preview of Llama3, including the release of an 8B and 70B (82 MMLU, should be the best open weights model!), and preliminary results for a 405B model (still training, but already competitive with GPT4). Lots more still to come... https://ai.meta.com/... Vedanuj Goswami / @vedanujg : Happy to be part of this incredible journey of Llama3 and to share the best open weight 8B and 70B models! Our largest 400B+ model is still cooking but we are providing a sneak peek into how it is trending! Check more details here https://ai.meta.com/... Yam Peleg / @yampeleg : It is so fun that Meta doesn't try to hide the uploads of LLaMA-3 before the official announcement because it is open source and there are no secrets anyway.. Tom Osman / @tomosman : No wonder Zuck had a swagger & gold chain at UFC 300! Llama-3 8b and 70b are here 🦙 👑 [image] Chris Paxton / @chris_j_paxton : So uh is Llama-3 out or not? Weird how it's up on Azure and replicate without a big press release. [image] LinkedIn: Mike Lasby : Llama 3 was trained on 15T tokens compared to just 2T for Llama 2!  Chinchilla scaling is dead, long live Chinchilla.  📈 … Brandon Spence : Llama 3 is here and we're just getting started.  Over the course of the year we'll be releasing new sets of models that range in size, expanded capabilities, and enhanced performance. … Stuart Pitts : Today is a big day.  The world is changing faster than ever with AI, and the pace of change is only increasing 🚀 … Diego Guerra Orozco : Attention all startups!  🚀 Meta Llama 3 is here!  —  Are you looking to give your business a competitive edge with cutting-edge AI technology? … Denny Lee : Super excited about #llama3, and we're rolling it out onto Databricks over the next few days, so you'll be able to try it out using our optimized foundation model API and/or AI Playground! … Nabil Tayeb : Llama3 is out!!  —  Surprisingly they did not take the MOE route  — 8B and 70B parameter models  — Improvements in pretraining data and instruction fine-tuning … Hitesh Patel : LLaMA3 is here!  The latest iteration of the LLaMA model from @Meta has some exciting new capabilities:  — Expanded 128K token vocabulary - more than 4x larger than the 32K vocab of LLaMA-2. … Sayeed Choudhury : Excited that today, we are launching Meta #llama3, our next generation state-of-the-art LLM.  —  As mentioned earlier by Mark Zuckerberg … Amit Sangani : There are 2 major launches from Meta today!  —  1) Meta Llama 3 launched: Our latest generation of models sets a new standard for state …

The Verge Alex Heath

Discussion

  • @jonporteruk Jon Porter on threads
    @crumbler I'm a little sceptical of this given the regional availability of Meta's AI stuff at launch (I don't have access in the UK for example).  I agree it has the potential to lead in distribution, but I struggle to believe it's there just yet https://blog.whatsapp.com/...
  • @georgescriban @georgescriban on threads
    I don't really understand what Meta's doing here.  Why invest in a consumer-facing generative AI capability, when the value of AI to Meta is in its application against the company's hard problems (safety, infosec, data protection, fraud, growing its ad business, etc)? …
  • @chriswelch Chris Welch on threads
    I don't need to ask questions to Instagram or WhatsApp.  This is the stupidest goddamn timeline.
  • @shiraovide Shira Ovide on threads
    I still don't understand why it's useful to most people to have an AI chatbot in Facebook, Instagram and WhatsApp.  Do I want to ask in Facebook for help crafting a sensitive email to my boss?  Or ask a WhatsApp chatbot for restaurant recommendations?  Or use an AI chatbot in the…
  • @mark_r_vickers Mark Vickers on threads
    I suspect people are getting the Llama 3 story wrong.  Sure, Meta will experiment with jamming it into social media, but that's not the main point of it.  The main point is to create an open source model that will be both powerful and compressed enough to work with the new genera…
  • @sivavaidhyanathan Siva Vaidhyanathan on threads
    I have been trying to figure out why OpenAI release such a bad product and why Meta et al have been rushing to market.  I don't think it's about making news or pandering to traders and analysts.  Real analysts know the level of R&D going on at Alphabet and Meta. …
  • @zuck Mark Zuckerberg on threads
    More details on Llama 3.  Today we're releasing 8B and 70B parameter models — both best-in-class for their size.  We've got more releases coming to bring multi-modality and longer context windows.  We're also still training a larger dense 400B+ parameter model.
  • @hamzamazari88 Hamza Mazari on threads
    Wow, making Llama 3 open source is a huge gift to the AI community, Meta is leading the way
  • @nickclegg Nick Clegg on threads
    Think of this as the first wave of Llama 3 - we're excited to get these in the hands of consumers and developers to help kickstart the next wave of AI innovation and we'll have more to share soon.
  • @soumith Soumith Chintala on threads
    Llama3 8B and 70B are out, with pretty exciting results!  * The ~400B is still training but results already look promising.  * Meta's own Chat interface is also live at https://www.meta.ai/ * TorchTune integration is shortly going live: https://github.com/...
  • @abmlllm95 Akarshan Biswas on threads
    Llama3 8B and 70B are now available on HuggingFace.  I have applied for the access.
  • @dwarkesh Dwarkesh Patel on threads
    . @zuck on: - Llama 3 - open sourcing towards AGI - custom silicon, synthetic data, & energy constraints on scaling - Caeser Augustus, intelligence explosion, $10b models, & much more Enjoy!  Links below
  • @salrodriguez Salvador Rodriguez on threads
    Meta today is releasing the first versions of Llama 3.  With this new AI model, the company will also be upgrading its Meta AI tool, launching it in more markets and featuring it more prominently across its apps https://www.wsj.com/... with @joannastern
  • @karpathy Andrej Karpathy on x
    Congrats to @AIatMeta on Llama 3 release!!  🎉 Notes: Releasing 8B and 70B (both base and finetuned) models, strong-performing in their model class (but we'll see when the rankings come in @ @lmsysorg:)) 400B is still training, but already encroaching GPT-4 territory (e.g. 84.8 MM…
  • @astonzhangaz Aston Zhang on x
    Llama 3 has been my focus since joining the Llama team last summer. Together, we've been tackling challenges across pre-training and human data, pre-training scaling, long context, post-training, and evaluations. It's been a rigorous yet thrilling journey: 🔹Our largest models... …
  • @davewiner Dave Winer on x
    Quick demo of the new https://meta.ai/ product from the company that makes Facebook. This is just one feature, its ability to recalc drawings as you edit the prompt that creates the picture. Very nice. As you can tell from the video I really like it....
  • @tomwarren Tom Warren on x
    Meta's chatbot is going about as well as you might expect 🙃 https://www.404media.co/... [image]
  • @sashamtl Sasha Luccioni, PhD on x
    So LLaMa 3's carbon footprint is... huge? 🤯 They estimate it to be 2,290 tons of CO2eq, compared to 550t for training GPT-3 and 66t for training *all* of the BLOOM models (1B-176B) 🌬️ [image]
  • @minimaxir Max Woolf on x
    @alexeheath 99% chance the next version of Grok will be a finetuned Llama 3 70B model.
  • @dwarkesh_sp Dwarkesh Patel on x
    Zuck on: - Llama 3 - open sourcing towards AGI - custom silicon, synthetic data, & energy constraints on scaling - Caeser Augustus, intelligence explosion, bioweapons, $10b models, & much more Enjoy! Links below [video]
  • @levie Aaron Levie on x
    Incredible to have Meta out here pushing open source AI forward in such a major way. An amazing moment to be building software when you have $100s of billions of dollars from the major tech players going into advanced R&D and infra that directly benefit you as a developer.
  • @yacinemtb Kache on x
    just turned back on all my data collection preferences for facebook & instagram. didn't realize it was going to such a good cause
  • @andrewyng Andrew Ng on x
    Meta released Llama 3 on my birthday! 🎂 Best present ever, thanks Meta! 😀
  • @intelai @intelai on x
    AI isn't just growing; it's evolving rapidly. Llama 3's leap in reasoning and performance is thrilling, and we're ready to support it with our Gaudi accelerators and Xeon CPUs. Looking forward to the new possibilities this will unlock, @AIatMeta!
  • @jpineau1 Joelle Pineau on x
    We just released Meta Llama 3: the most capable openly available LLM available to date! The 8B & 70B models are out now, and we expect to release models with larger context windows, additional model sizes and more capabilities in the coming months.
  • @emollick Ethan Mollick on x
    “Llama 3, Write an office memo justifying the fact that management has been replaced by squirrels. We are all being paid in acorns, now.” Nice job. [image]
  • @intel @intel on x
    We're thrilled to have validated our #AI product portfolio for the Meta Llama 3 8B and 70B models! This is another example of our commitment to the latest AI innovations. Check out the link to learn more: https://www.intel.com/...
  • @aiatmeta @aiatmeta on x
    You can download the models directly from Meta today, additionally though, working closely with our partners, Llama 3 models will be available soon across @AWSCloud, @Azure, @Databricks, @GoogleCloud, @HuggingFace, @IBM WatsonX, @Kaggle, @NVIDIA and @SnowflakeDB.
  • @jpineau1 Joelle Pineau on x
    Congratulations @Ahmad_Al_Dahle + the entire GenAI team + all of the incredible XFN across @AIatMeta that came together to ship these new models today! 👏
  • @amd @amd on x
    We're excited to work with @AIatMeta on Llama 3, the next generation of their open-source large language model. As a major hardware partner of Meta, we're committed to simplifying LLM deployments and enabling outstanding TCO.
  • @soumithchintala Soumith Chintala on x
    Llama3 8B and 70B are out, with pretty exciting results! * The ~400B is still training but results already look promising. * Meta's own Chat interface is also live at https://www.meta.ai/ * TorchTune integration is shortly going live: https://github.com/...
  • @namangoyal21 Naman Goyal on x
    Really proud of the work that went into making this possible, hope this helps the community push the field forward. Also in case anyone missed it, there's a sneak peak of what to come next at the end of blog post https://ai.meta.com/...
  • @rsumbaly Roshan Sumbaly on x
    Hello Llama3! The first drop (8B, 70B) that we're doing for the community. Still baking the 405B model and also have more systems and capabilities (multimodality, multilingual, tooling) lined up to share. Also thinking of Llama 4 already. We're just getting started!
  • @elonmusk Elon Musk on x
    @ylecun Not bad
  • @ahmad_al_dahle Ahmad Al-Dahle on x
    It's here! Meet Llama 3, our latest generation of models that is setting a new standard for state-of-the art performance and efficiency for openly available LLMs. Key highlights • 8B and 70B parameter openly available pre-trained and fine-tuned models. • Trained on more... [image…
  • @tobi Tobi Lutke on x
    Llama3 looks fantasic. Thank you and congrats @Meta Open source models are the key platform for innovation around productivity, safety, and innovation https://llama.meta.com/llama3/
  • @garridoq_ Quentin Garrido on x
    Fixed [image]
  • @carnage4life Dare Obasanjo on x
    Not sure what's going harder, the Meta AI and Llama 3 announcement or the drip ⛓️ [video]
  • @leokelion Leo Kelion on x
    Meta has launched Llama 3 and with it a ChatGPT rival with a standalone website... that isn't yet available in the UK. Hopefully not too long before we can try it out. Wonder if @perplexity_ai will gain access? [image]
  • @emollick Ethan Mollick on x
    As a reminder, GPT-4 class models out-innovate and out-persuade the average human. An open sourced model of that power is going to lead to lots of unanticipated effects, good and bad.
  • @taylorbuley Tay-I on x
    2x24k clusters of GPUs to train Llama3. 50k GPUs all working on the singular goal. Kind of beautiful.
  • @gabgarrett Gabriel Garrett on x
    looks like Llama 3 actually beats out older versions of GPT-4. it's pretty big that there's an open-source version of a model at that level now. llama3 400b will be the moment of truth for beating out GPT-4-04-09 though
  • @maximelabonne Maxime Labonne on x
    Llama 3's vocabulary size is much bigger (32000 => 128256). Also note 11008 => 14336. [image]
  • @natolambert Nathan Lambert on x
    Diff of llama 3 license to llama 2: Mostly around sharing, built with llama 3 branding, agree to meta brand guidelines for distributing trademark Some minor other differences [image]
  • @togethercompute @togethercompute on x
    We are thrilled to be a launch partner for Meta Llama 3. Experience Llama 3 now at up to 350 tokens per second for Llama 3 8B and up to 150 tokens per second for Llama 3 70B, running in full FP16 precision on the Together API! 🤯 https://together-ai.webflow.io/ ... [video]
  • @rohanpaul_ai Rohan Paul on x
    What's more crazy is the 400bn Llama 3, which is still in training, and already, with its current checkpoints, the scores look absolutely BRILLIANT. And this is Meta's statement on that. “Our largest models are over 400B parameters and, while these models are still training,... […
  • @simonw Simon Willison on x
    You can run prompts to rough the new Llama 3 models on Replicate: https://replicate.com/... for 70b-instruct and https://replicate.com/... for 8b-instruct
  • @charliebholtz Charlie Holtz on x
    We have Llama 3 running at 55 tokens/second on https://llama3.replicate.dev/: [video]
  • @replicate @replicate on x
    Llama 3's out. Try it here: https://llama3.replicate.dev/
  • @aiatmeta @aiatmeta on x
    Introducing Meta Llama 3: the most capable openly available LLM to date. Today we're releasing 8B & 70B models that deliver on new capabilities such as improved reasoning and set a new state-of-the-art for models of their sizes. Today's release includes the first two Llama 3... […
  • @rohanpaul_ai Rohan Paul on x
    Great results of Llama-3 Human evaluations across against Claude Sonnet, Mistral Medium, and GPT-3.5. [image]
  • @deliprao @deliprao on x
    The much-awaited Llama 3 release is here! https://llama.meta.com/llama3/ My quick notes: [image]
  • @sullyomarr @sullyomarr on x
    Wait a second.. llama3 beats both sonnet and gemini pro 1.5 This is actually huge. Open source models are back [image]
  • @johncodezzz John McBride on x
    Huge: @fb_engineering releases llama3 - “The most capable openly available LLM to date” The future of AI technology is truly open! A few thoughts 🧵 https://ai.meta.com/...
  • @natolambert Nathan Lambert on x
    llama 3 8b has the some approx amount of training compute as llama 2 70b lol (15T tokens training for v3, 2T for 2) [image]
  • @andrewcurran_ Andrew Curran on x
    ‘Llama 3 is pretrained on over 15T tokens that were all collected from publicly available sources’ The paper will not be released today. [image]
  • @naveengrao Naveen Rao on x
    Congrats @AIatMeta team on the Llama3 release! This is a huge milestone for AI and the market...the performance of these models are a huge step compared to a year ago and give users so much more flexibility on how they want to use, deploy, and customize AI
  • @imjaredz Jared Zoneraich on x
    Llama 3 is out. You can't ignore open-source models anymore. It's supposedly better than Claude 3 Sonnet 🤯🤯 Huge [image]
  • @natolambert Nathan Lambert on x
    we're all following the way llama 3 400b is going [image]
  • @_borriss_ @_borriss_ on x
    Llama 3 by Meta is out. Looking strong. Full details here: https://llama.meta.com/llama3/ [image]
  • @kaylacardillo Kayla Cardillo on x
    LLaMA-3 official release is said to be scheduled for today. Incoming: 8 billion parameters/70 billion parameters, 8,000 context length, 128,000 vocabulary, and 15 trillion tokens. The increase in parameters means the model will be able to learn more complex patterns and...
  • @maxwinebach Max Weinbach on x
    Llama 3 is live from meta and benchmarks make it look like a great model! 8B and 70B dense models But man that 8K context window is rough given we're seeing 32K is similar size models already https://llama.meta.com/llama3/
  • @rao_hacker_one Arun Rao on x
    Thrilled to share a preview of Llama3, including the release of an 8B and 70B (82 MMLU, should be the best open weights model). Also preliminary results for a 405B model (still training, but already competitive with GPT4). Much more to come... https://llama.meta.com/llama3/
  • @eladgil Elad Gil on x
    Llama 3 out from Meta - https://llama.meta.com/llama3/ [image]
  • @natolambert Nathan Lambert on x
    llama 3 8b has the some approx amount of training compute as llama 2 70b lol (15T tokens) [image]
  • @andrewcurran_ Andrew Curran on x
    Official release benchmarks. [image]
  • @xlr8harder @xlr8harder on x
    Llama 3 is out, but it seems that still only @MistralAI really has our back: it still has downstream usage restrictions of model output. @AIatMeta this is a garbage restriction that harms open source and I was really hoping you would see the light this time. [image]
  • @ml_perception Mike Lewis on x
    Excited to share a preview of Llama3, including the release of an 8B and 70B (82 MMLU, should be the best open weights model!), and preliminary results for a 405B model (still training, but already competitive with GPT4). Lots more still to come... https://ai.meta.com/...
  • @vedanujg Vedanuj Goswami on x
    Happy to be part of this incredible journey of Llama3 and to share the best open weight 8B and 70B models! Our largest 400B+ model is still cooking but we are providing a sneak peek into how it is trending! Check more details here https://ai.meta.com/...
  • @yampeleg Yam Peleg on x
    It is so fun that Meta doesn't try to hide the uploads of LLaMA-3 before the official announcement because it is open source and there are no secrets anyway..
  • @tomosman Tom Osman on x
    No wonder Zuck had a swagger & gold chain at UFC 300! Llama-3 8b and 70b are here 🦙 👑 [image]
  • @chris_j_paxton Chris Paxton on x
    So uh is Llama-3 out or not? Weird how it's up on Azure and replicate without a big press release. [image]
  • @seyitaylor @seyitaylor on x
    I told you Zuck was the man while everyone argued with me. It's now consensus. Do you want to know what's next?
  • @andrewcurran_ Andrew Curran on x
    Llama 3 70B never stopped learning. He says the only reason they stopped its training was that they eventually had to decide: ‘Do we want to spend our GPUs on training the 70B model further?’ or should we start training what's next? [image]
  • @rowancheung Rowan Cheung on x
    Don't bet against Meta AI.  The next big AI adoption point will be from a massive distribution (e.g., Meta or Apple) dropping a GPT-4 level model for free directly into their products...I think this is bigger than people realize...If this interview is accurate and Meta executes p…
  • @brij Brij Singh on x
    He keeps winning on the likability graph! This section was good [image]
  • @mannynotfound @mannynotfound on x
    why is this meme from 2 years ago coming true [image]
  • @zuck Mark Zuckerberg on threads
    Big AI news today.  We're releasing the new version of Meta AI, our assistant that you can ask any question across our apps and glasses.  Our goal is to build the world's leading AI.
  • @zuck Mark Zuckerberg on threads
    We're upgrading Meta AI with our new state-of-the-art Llama 3 AI model, which we're open sourcing.  With this new model, we believe Meta AI is now the most intelligent AI assistant that you can freely use.
  • @creators @creators on threads
    Did you know that Meta AI on Instagram can help creators do things like: 👉 Create an editorial calendar 👉 Write a script for one of your Reels 👉 Generate captions for your posts 👉 Write a collaboration pitch email for a brand you'd like to work with
  • @messenger @messenger on threads
    Today, Messenger is launching a dedicated Meta AI tab in the app to help you plan, stay organized, get recommendations, learn new things and more.  You can also use Meta AI in the search feature of the Messenger app to get real-time information from across the web, learn more her…
  • @zuck Mark Zuckerberg on threads
    We're making Meta AI easier to use by integrating it into the search boxes at the top of WhatsApp, Instagram, Facebook, and Messenger.  We also built a website, meta.ai, for you to use on web.
  • @chriswelch Chris Welch on threads
    What if I was happy with normal, regular-ass search in Instagram?  I can't even find a setting to turn off or hide Meta AI.  🥴
  • @susyqli Susan Li on threads
    Zuck announces Meta's new AI upgraded assistant Llama3 👇 Maybe my prompts need improving but I still find ChatGPT to be more useful for now But Llama 3 says it will be able to fill in words & pics for you as you type which should be interesting Nice chain ⛓️ AI Threads
  • @nickclegg Nick Clegg on threads
    Today we're rolling out Meta AI powered by Llama 3, making it easily accessible across all our apps in more countries.  We're also open sourcing the first set of our Llama 3 models at 8B and 70B parameters, both best-in-class performance at their scale. https://ai.meta.com/
  • @hamburger Ellis on x
    brand will be a big barrier here if only they'd become fAIcebook it could've helped
  • @scobleizer Robert Scoble on x
    Instagram adds AI. AI is eating the world.
  • @taylorbuley Tay-I on x
    Meta AI and Open AI both using integrated Bing search results. Now with FB release to Whatsapp etc. ... could Bing eventually be the most-used search engine via backend backdoor?
  • @mikeisaac Rat King on x
    the biggest weapon meta always wields against competitors is scale. more than a dozen countries now, but eventually to nearly 4 billion people. most likely the mainstream's first encounter with consumer-facing AI assistants, despite OpenAI and MSFT being first to market
  • @mgsiegler M.G. Siegler on threads
    Mesmerized by this image morphing as I typed with random things popping into my head.  Well done Meta.  Now please give us this on Threads.
  • @zuck Mark Zuckerberg on threads
    We also built some unique creation features, like the ability to animate photos.  Meta AI now generates high quality images so fast that it creates and updates them in real-time as you're typing.  It'll also generate a playback video of your creation process.
  • @chriscox Chris Cox on threads
    Welcome to the world Llama 3!  We released an upgraded Meta AI and are today open-sourcing our new Llama 3 8B and 70B models which demonstrate best-in-class performance.  Meta AI will also launch to 13 more markets today along with our upgraded Emu image model and our low-latency…
  • @deviparikh Devi Parikh on x
    Llama 3! 🥳 Integrated in Meta AI. Meta AI also has new (and super fast) image generation, editing and animation capabilities! https://www.meta.ai/ or on any Meta surface near you :)
  • @deviparikh Devi Parikh on x
    Imagine Flash in Meta AI is so much fun :) https://meta.ai/ [video]
  • @jamespearce James Pearce on x
    Meta AI ‘/imagine’ is SO good. [image]
  • @zuck Mark Zuckerberg on threads
    To give a sense of performance, this 8B model is nearly as good as the biggest Llama 2 model.  This 70B model is around 82 MMLU with leading reasoning and math benchmarks.  The 400B+ model is currently around 85 MMLU but it's still training, so we expect it to lead on several ben…
  • @vishvanands Vishvanand Subramanian on threads
    Given Llama-3 400B is on par with the current best model Claude-3 Opus and it's still training, we can soon expect to see the dream of open source realized: The best model in the world is now free and open source.
  • @_kuanhoong_ @_kuanhoong_ on threads
    Llama 3 by Meta is here!  8B and 70B pretrained and instruction-tuned models are available. https://llama.meta.com/llama3/ Below is performance comparison Lllama-3 with Gemma, Gemini Pro 1.5, Mistal and Claude 3
  • @aiatmeta @aiatmeta on x
    Llama 3 delivers a major leap over Llama 2 and demonstrates SOTA performance on a wide range of industry benchmarks. The models also achieve substantially reduced false refusal rates, improved alignment and increased diversity in model responses — in addition to improved... [imag…
  • @ylecun Yann LeCun on x
    🥁 Llama3 is out 🥁 8B and 70B models available today. 8k context length.  Trained with 15 trillion tokens on a custom-built 24k GPU cluster.  Great performance on various benchmarks, with Llam3-8B doing better than Llama2-70B in some cases.  More versions are coming over the next …
  • @emollick Ethan Mollick on x
    Meta released their open source AI, Llama 3, today. As a key leader in LLMs, their models are often the most advanced open source ones out there. Based on benchmarks, the current model is not quite GPT-4 class, but their larger one (still training) will reach GPT-4 level. [image]
  • @mattshumer_ Matt Shumer on x
    The craziest LLaMA 3 reveal: The 400B+ version of the model is **on par with Claude 3 Opus**, and it's still training. Soon, we'll have a better-than-Opus, fully open-source model. The implications are huge. [image]
  • @mattshumer_ Matt Shumer on x
    Holy shit. LLaMA 3 70B cleanly beats Claude 3 Sonnet. Small enough to host at scale without breaking the bank. [image]