Zuckerberg said Meta's Llama 4 models were training on an H100 cluster "bigger than anything that I've seen reported for what others are doing."

Mark Zuckerberg Flexes Meta's Nvidia Chip Cluster - Business Insider

Despite posting robust quarterly profits, Microsoft (MSFT) and Meta (META) shares slumped Thursday as both tech giants signaled substantial future AI spending requirements and slower growth prospects. The companies, which have emerged as leaders in the AI sector, saw their stocks decline despite ...

Nvidia and other AI stocks are getting dragged down by Microsoft and Meta

Some Meta employees on Blind were sympathetic. Others suggested the policy violators chose their own fate by misusing the Grubhub meal credits.

Meta Employees React to 'Grubgate' Firings - Business Insider

Nvidia recently released an open-source AI model built on Meta’s Llama-3 called “Nemotron” that surpasses OpenAI’s GPT-4o on many popular benchmarks.

Nvidia’s new open-source AI model beats GPT-4o on benchmarks

Meta reorganized some of its major units, cutting roles at Instagram, Reality Labs, and WhatsApp.

Meta Reorganized Some Its Major Businesses and Eliminated Some Roles - Business Insider

This chipmaker is essential to the advancement of artificial intelligence.

1 Artificial Intelligence (AI) Stock-Split Stock Set to Join Nvidia, Apple, Microsoft, Amazon, Alphabet, and Meta in the $1 Trillion Club

When you think of Meta (META), you might think of Facebook, Instagram, or, shudder, the metaverse. But the social media company has quickly become an artificial intelligence darling up there with the likes of OpenAI, Microsoft (MSFT), and Google (GOOG, GOOGL) · Meta’s AI success comes via ...

How Meta has become an AI behemoth

What Is The Deal With Meta's Threads

I Led Dozens of Meta System Design Interviews, Here's The Right Way to Prep

I'm excited to tell you about Meta's Llama 3.1, a powerful AI language model you can use for free. In this article, I'll show you how to install and run Llama 3.1 on your computer. This means you can use it without the internet and save money on AI costs. https://youtu.be/UdcmnhmruIs?si=9okWfpSuMgd1AYOZ&embedable=true What is Llama 3.1? Llama 3.1 is Meta’s newest AI model that’s open for anyone to use. It comes in three sizes: A huge 405-billion parameter version
A medium 70 billion parameter version
A smaller 8 billion parameter version Llama 3.1 is special because it works really well. It can even do better than some private AI models. You can read more about it here:https://llama.meta.com/ Why Llama 3.1 is Great It’s free.
Anyone can change and improve it.
It works as well as expensive private models.
It understands many languages.
It can handle long texts. It can work with up to 128,000 words at once, which is great for long writing tasks. Challenges with Llama 3.1 It needs a lot of computer power: The biggest version needs very powerful computers to run.
Might have biases: Like all AI, it might repeat unfair ideas from its training data.
Possible safety issues: Because anyone can see how it works, there might be some safety risks if not used carefully. Where to Try Llama 3.1 Here are some ways you can try out Llama 3.1 for free: Meta.ai Chatbot: Meta’s own platform offers access to Llama 3.1, including the 405B model, along with image generation capabilities. However, this is currently limited to U.S. users with a Facebook or Instagram account.
Perplexity: This AI-powered search tool offers access to Llama 3.1 405B, but it requires a paid subscription.
Cloud Platforms: Various cloud providers offer access to Llama 3.1, including: Amazon Bedrock
Microsoft Azure AI
Cloudflare Workers AI
Snowflake
DataBricks
Nvidia AI Foundry
IBM Cloud In this tutorial, I’ll show you how to use Llama 3.1 for free on your own machine. How to Run Llama 3.1 on Your Computer The big 405 billion version needs a very powerful computer. But I’ll show you how to run the 8-billion version. Here’s what you need: 16 GB of RAM
8-core CPU
20 GB of free space Step 1. Download and Install the Ollama Tool Go to https://ollama.com/ and download the tool for your operating system (OS). For Linux, I’ll use this command:curl -fsSL https://ollama.com/install.sh | sh The command will install Ollama, which is a tool for running and managing large language models (LLMs) locally on your machine. Step 2. Download the Llama 3.1 Model On the same website, click on the link “Models” in the top right corner. Then, find Llama 3.1 in the list. Click on the link and then select the desired model from the drop-down menu. Copy the command and paste it into your terminal. Terminal command: ollama run llama3.1 Wait for it to download (it might take 30 minutes or more). If everything is okay, you can use LLM inside your terminal. Enjoy! :) Using Llama 3.1 Without Internet One cool thing about having Llama 3.1 on your computer is you can use it without the internet. This is great if you need to keep your data private. Things to Remember There are a few downsides to know about: The 8 billion model only knows things up to 2021
It can’t look at files on your computer or websites
It might be slower than online AI services How to Give Llama 3.1 Information The model can’t read your files or websites, but you can: Copy and paste text into it
Use things like Dropbox to share files
Summarize what you want it to know in your question Wrapping Up Running Llama 3.1 on your own computer is a great way to use powerful AI without the internet or ongoing costs. It has some limits, but it’s still amazing to have such a smart AI tool on your own machine. If you like this tutorial, please click like and share. You can follow me on YouTube, join my Telegram, or support me on Patreon. Cheers! :) I'm excited to tell you about Meta's Llama 3.1, a powerful AI language model you can use for free. In this article, I'll show you how to install and run Llama 3.1 on your computer. This means you can use it without the internet and save money on AI costs. https://youtu.be/UdcmnhmruIs?si=9okWfpSuMgd1AYOZ&embedable=true https://youtu.be/UdcmnhmruIs?si=9okWfpSuMgd1AYOZ&embedable=true What is Llama 3.1? Llama 3.1 is Meta’s newest AI model that’s open for anyone to use. It comes in three sizes: A huge 405-billion parameter version A medium 70 billion parameter version A smaller 8 billion parameter version A huge 405-billion parameter version A medium 70 billion parameter version A smaller 8 billion parameter version Llama 3.1 is special because it works really well. It can even do better than some private AI models. You can read more about it here: https://llama.meta.com/ https://llama.meta.com/ https://llama.meta.com/ Why Llama 3.1 is Great It’s free. Anyone can change and improve it. It works as well as expensive private models. It understands many languages. It can handle long texts. It can work with up to 128,000 words at once, which is great for long writing tasks. It’s free. Anyone can change and improve it. It works as well as expensive private models. It understands many languages. It can handle long texts. It can work with up to 128,000 words at once, which is great for long writing tasks. Challenges with Llama 3.1 It needs a lot of computer power: The biggest version needs very powerful computers to run. Might have biases: Like all AI, it might repeat unfair ideas from its training data. Possible safety issues: Because anyone can see how it works, there might be some safety risks if not used carefully. It needs a lot of computer power: The biggest version needs very powerful computers to run. Might have biases: Like all AI, it might repeat unfair ideas from its training data. Possible safety issues: Because anyone can see how it works, there might be some safety risks if not used carefully. Where to Try Llama 3.1 Here are some ways you can try out Llama 3.1 for free: Meta.ai Chatbot: Meta’s own platform offers access to Llama 3.1, including the 405B model, along with image generation capabilities. However, this is currently limited to U.S. users with a Facebook or Instagram account. Perplexity: This AI-powered search tool offers access to Llama 3.1 405B, but it requires a paid subscription. Cloud Platforms: Various cloud providers offer access to Llama 3.1, including: Meta.ai Chatbot: Meta’s own platform offers access to Llama 3.1, including the 405B model, along with image generation capabilities. However, this is currently limited to U.S. users with a Facebook or Instagram account. Perplexity: This AI-powered search tool offers access to Llama 3.1 405B, but it requires a paid subscription. Cloud Platforms: Various cloud providers offer access to Llama 3.1, including: Amazon Bedrock Microsoft Azure AI Cloudflare Workers AI Snowflake DataBricks Nvidia AI Foundry IBM Cloud Amazon Bedrock Microsoft Azure AI Cloudflare Workers AI Snowflake DataBricks Nvidia AI Foundry IBM Cloud In this tutorial, I’ll show you how to use Llama 3.1 for free on your own machine. How to Run Llama 3.1 on Your Computer The big 405 billion version needs a very powerful computer. But I’ll show you how to run the 8-billion version. Here’s what you need: 16 GB of RAM 8-core CPU 20 GB of free space 16 GB of RAM 8-core CPU 20 GB of free space Step 1. Download and Install the Ollama Tool Go to https://ollama.com/ and download the tool for your operating system (OS). https://ollama.com/ https://ollama.com/ For Linux, I’ll use this command: curl -fsSL https://ollama.com/install.sh | sh curl -fsSL https://ollama.com/install.sh | sh The command will install Ollama, which is a tool for running and managing large language models (LLMs) locally on your machine. Step 2. Download the Llama 3.1 Model On the same website, click on the link “Models” in the top right corner. “Models” Then, find Llama 3.1 in the list. Click on the link and then select the desired model from the drop-down menu. Copy the command and paste it into your terminal. Terminal command: ollama run llama3.1 ollama run llama3.1 Wait for it to download (it might take 30 minutes or more). If everything is okay, you can use LLM inside your terminal. Enjoy! :) Using Llama 3.1 Without Internet One cool thing about having Llama 3.1 on your computer is you can use it without the internet. This is great if you need to keep your data private. Things to Remember There are a few downsides to know about: The 8 billion model only knows things up to 2021 It can’t look at files on your computer or websites It might be slower than online AI services The 8 billion model only knows things up to 2021 It can’t look at files on your computer or websites It might be slower than online AI services How to Give Llama 3.1 Information The model can’t read your files or websites, but you can: Copy and paste text into it Use things like Dropbox to share files Summarize what you want it to know in your question Copy and paste text into it Use things like Dropbox to share files Summarize what you want it to know in your question Wrapping Up Running Llama 3.1 on your own computer is a great way to use powerful AI without the internet or ongoing costs. It has some limits, but it’s still amazing to have such a smart AI tool on your own machine. If you like this tutorial, please click like and share. You can follow me on YouTube , join my Telegram , or support me on Patreon . YouTube YouTube Telegram Telegram Patreon Patreon Cheers! :)

No More AI Costs: How to Run Meta Llama 3.1 Locally

The fall and rise of have been nothing short of astonishing in recent years. After shedding more than 70% of its value amid 2022’s tech stock sell-offs, the stock surged 180% in 2023–and the tech giant is far from finished. Meta It’s difficult to find the right superlatives for Meta [NASDAQ: META]. The stock became one of the best performers on the global stage last year, and the company’s measured approach to embracing artificial intelligence formed a key component of its impressive growth. Today, Meta has almost recovered all of its losses amid severe economic headwinds in late 2021 and 2022 and has recaptured a market capitalization of almost $1 trillion. The stock’s impressive movements throughout the past 12 months have seen it become one of the stars of the ‘Magnificent Seven,’ a collective of leading tech stocks including Apple, Microsoft, Alphabet, Amazon, Nvidia, Meta, and Tesla. But as the generative AI boom rumbles on and ideation shifts towards implementation for the emerging technology, can Meta continue to facilitate its growth throughout 2024? Let’s take a deeper look at why Meta has become the stock that investors are coveting most this year: Finding Strength in Fundamentals It’s impossible to ignore Meta’s underlying financial strength as a key attribute to 2023’s impressive Wall Street growth. “The company has demonstrated strong financial performance, with a 23% growth in revenue to $34.1 billion in the third quarter of 2023, surpassing expectations,” highlights Maxim Manturov, head of investment research at . Freedom Finance Europe “Meta's vision for the effectiveness of the company has translated into reality, and its diverse user base supports profitable, targeted advertising, a significant revenue driver.” In addition to its impressive revenue growth, Meta is an attractive proposition for advertisers, thanks to its extensive data reservoirs that pave the way for more targeted campaigns. As a result, the firm’s advertising revenue has increased 23% in Q3 to $33.6 billion. However, it appears to be Meta’s inroads in the field of artificial intelligence that will help to leverage growth in 2024. Meta’s AI offerings currently include Quest 3, AI-powered Ray-Ban Stories, and the AI Studio platform. With a significant free cash flow (FCF) balance of $13.6 billion by the end of Q3 2023, analysts are expecting AI investments to improve stock performance in the new year, anticipating revenue rising 13% year-over-year to $151 billion, while earnings are expected to rise 21% over the same time frame. According to a team of , there could be three catalysts to push Meta’s stock higher in 2024. The first comes in the form of a conservative consensus for sales growth, which may prove to be stronger thanks to better monetization of Reels short videos and growing interest from Chinese advertisers. Mizuho Securities analysts led by James Lee Secondly, the analysts believe that guidance on operating expenditures could fall over the financial year, which is a trend with some precedence. Finally, WhatsApp messaging performance may contribute to the revenue base by as much as a third due to the implementation of AI for automated customer service. Finding Strength in AI So far, much of the positive sentiment surrounding Meta in 2024 stems from optimism surrounding generative AI. There’s a good reason for this. While Meta has long utilized AI within its businesses in the form of identifying faces posted on Facebook and delivering curated content and ads for its users, the future of the company lies in its utility of generative AI services. In early 2023, Meta launched a for firms to bolster their social media advertising. Through an Ads Manager, images and text can be adapted from user to user to target their products more effectively. suite of free AI-powered marketing tools Crucially, this launch of AI tools has helped to make Facebook an even more attractive prospect for companies seeking to improve their marketing ROI. For Meta CTO Andrew Bosworth, the arrival of generative AI was an opportunity for Meta to ramp up its operations. “AI has always been a major component of our thinking,” . “But like all technologies, you’re in these S curves. We’ve known for three or four years that these generative AI, large language models were a big deal, but they still felt like infrastructure. That’s why I think the OpenAI introduction of ChatGPT is so interesting.” Bosworth explained “I’m not trying to say we were prescient. We had been investing, and we believed in this technology. For all of that belief, it still snuck up on us. It was like, ‘Oh my god, it’s here. We’ve got to do things differently. We’ve got to change it up.’” Will Meta Lead the ‘Magnificent Seven’ in 2024? To better understand Meta’s position among the Magnificent Seven, it’s worth exploring its PEG ratio, which refers to the stock’s price/earnings in comparison to the company’s growth. With a , Meta features the second-best ratio of the seven Wall Street leaders, falling behind Nvidia with an impressive 0.95 PEG. PEG ratio of 1.2 As another stock that’s heavily exposed to generative AI, Nvidia’s growth in 2024 will hinge on the development of the emerging technology and the success of the implementation process as more enterprises seek to embrace the GenAI boom. Meta, on the other hand, will rely on its rich history of innovation to harness the power of generative AI based on its products and services. With Meta’s history of resilience and strategic acumen in often unpredictable tech markets, we could see the stock outpacing its Magnificent Seven rivals in 2024, and with solid fundamentals continuing to drive growth, there’s no reason why Meta shouldn’t be at the top of investor priority lists for the year ahead.

Meta’s Meteoric Rise in 2023 Shows No Sign of Slowing This Year

2023 সালে মেটার উল্কা উত্থান এই বছর ধীর হওয়ার কোন লক্ষণ দেখায় না

Metas kometenhafter Aufstieg im Jahr 2023 zeigt in diesem Jahr keine Anzeichen einer Verlangsamung

El meteórico ascenso de Meta en 2023 no muestra signos de desaceleración este año

L’ascension fulgurante de Meta en 2023 ne montre aucun signe de ralentissement cette année

2023 में मेटा की उल्कापिंडीय वृद्धि इस वर्ष धीमी होने का कोई संकेत नहीं दिखाती है

2023 年のメタの流星的上昇は今年も衰える兆しなし

2023년 메타의 급격한 상승세는 올해도 둔화될 조짐을 보이지 않습니다

A ascensão meteórica do Meta em 2023 não mostra sinais de desaceleração este ano

Метеоритный взлет Меты в 2023 году не покажет никаких признаков замедления в этом году

Meta'nın 2023'teki Meteorik Yükselişi Bu Yıl Yavaşlama İşareti Göstermiyor

Sự gia tăng thiên thạch của Meta vào năm 2023 không có dấu hiệu chậm lại trong năm nay

Meta 在 2023 年的迅速崛起今年没有放缓的迹象

Quality is better than quantity when it comes to fine-tuning LLMs — Emu. “ ” aka. is a released by Meta to generate highly aesthetic images. It achieves this by introducing a fine-tuning technique called which goes on to show that as little as two thousand high-quality images are all that is needed to achieve the objective, hence the paper title, “ ” Enhancing Image Generation Models Using Photogenic Needles in a Haystack, Emu paper quality tuning, needles in a haystack. In this article, let’s look at the model architecture and the process of collating the quality dataset to train the , and, lastly, some of the mind-blowing results achieved by Emu. Emu model Visual Explanation If you are a visual learner like me, you might prefer watching the Emu model explained through a video here: https://youtu.be/W2qjoM3cHf4?feature=shared&embedable=true Approach Emu is trained in two stages, namely and quality . Though this may be a well-known recipe, the main message from this paper is that it all comes down to quality rather than quantity. So the authors have not only curated such a quality fine-tuning dataset but shown that such a fine-tuning does not compromise on the generality of the model as measured by the faithfulness metric. pre-training fine-tuning Architecture The first contribution of the paper is the modifications to the latent diffusion architecture. In that first comes the autoencoder, which has the encoder and a decoder. The autoencoders typically use four channels in their intermediate layers, thereby compressing the data and restricting their ability to represent fine details. In this work, the authors have increased the number of layers in the intermediate channels from 4 to 8 and 16. To increase the input channels further, they do a Fourier feature transform to lift the input channel dimension from 3 (RGB) to even higher dimensions. With these modifications, they tested the reconstruction of images and evaluated reconstruction scores such as FID and peak signal-to-noise ratio. We can notice that as the channels increase, these metrics show an improvement. Even visually, the quality of images seems to get far better as the number of channels increases: The next modification to the architecture is that of the U-Net used for denoising. To increase the capacity of the U-Net, they increase the channel size and the number of stacked residual blocks in each stage. And finally, they modify the decoder to output images at a high resolution of 1024 by 1024. Pre-Training With those architectural changes, they start with the first stage of training, which is pre-training. Pretraining needs a massive dataset, and so they pre-train with 1.1 billion images. But the point to note here is that they employ progressive training by gradually increasing the resolution of the image. This seems to help the model learn fine details in the images slowly but steadily. The other thing is the introduction of noise offset in the final stages of pre-training. It was introduced in which states that simply modifying the noise with a small offset enables Stable Diffusion to generate very dark or light images easily. , this blog The offset can be introduced with 1 line of code modification in PyTorch, where we add a small offset of 0.1 or similar to the noise generation process. Fine-tuning The second stage of the training is Fine-tuning. The pre-trained model obtained in the first stage is now subject to quality-tuning. But for this, we need the quality dataset in the first place. So, how do we curate such a quality dataset for quality tuning? For this, the authors of the paper employ a two-stage filtering process, namely and . automatic filtering manual filtering In automatic filtering, they use algorithms to remove images that have too many texts using the OCR recognition algorithm. They also employ CLIP to keep only results with good image-text alignment. Lastly, they also remove images with bad aspect ratio and image size and end up with a dataset of 200 thousand images which is a lot less than the billions we started with. In the next step of manual or human filtering, humans manually assess the images for aesthetic appeal. For example, this image satisfies the rule of thirds in photography so it is aesthetically appealing. Similarly, portraits taken using professional photography gear are quite appealing compared to ordinary images because the background is all blurred, creating what is called a bokeh in photography. So it’s much more aesthetically appealing compared to a selfie taken on, say, a mobile phone. The human filtering step resulted in just two thousand images. They then composed captions to those images. The image-caption pairs were then used for training the Emo model. Because there are just 2k samples, they fine-tuned for just 15K iterations with early stopping. The results are incredible, as seen in the next section. Metrics For evaluation, they were mainly concerned about the two metrics — and . They ditch the FID scores as many recent papers argue that FID scores do not correlate well with human assessment of the performance of generative models. visual appeal faithfulness Visual appeal is subjective, so the generated images were shown to 5 annotators, usually with generated images from two models shown side by side, and the annotators chose which one of the two was more appealing. For example, if we are comparing the pre-trained model with the quality-tuned model (as shown in the figure above), then these two images are shown to the annotators who are asked to choose which one of the two images is visually appealing. I would personally choose the quality-tuned one straightaway. Text faithfulness refers to the degree of similarity between a generated image and a text caption. Again, the annotators were asked to ignore the visual appeal of the images and asked to choose which ones best describe the caption with choices “A,” “B,” and “Both.” For example, the above image is for the caption, “A teddy bear working on AI research,” which is more faithful compared to “A teddy bear playing with a computer.” Results Lastly, the paper shows that quality tuning is not just restricted to the latent diffusion models but is applicable to other models like pixel diffusion and masked generative transformers. As seen from the figure above, quality tuning, when applied to these models, also improves the visual aesthetics of the model. Conclusion In my opinion, this paper is a clear eye-opener to show Emu generates results that are so good that the generated images can be used in commercial applications. how small the fine-tuning dataset can be to tune LLMs or VLMs. There are a couple of extensions of this paper, namely Emu Edit and Emu Video. I will be reviewing them in the upcoming posts. Until then, I am signing off, and I will see you in my next… till then, take care!

Meta's Emu: The Foundational Model for Emu Edit and Emu Video

The United States v Meta Platforms Court Filing October 24, 2023 is part of . You can jump to any part in this filing . This is part 48 of 100. HackerNoon’s Legal PDF Series here COUNT III: FALSE OR MISLEADING STATEMENTS BY META IN VIOLATION OF BUSINESS AND PROFESSIONS CODE SECTION 17500 (BY THE PEOPLE OF THE STATE OF CALIFORNIA) 865. The People of the State of California (California) reallege and incorporate herein by reference each of the allegations contained in the preceding paragraphs 1 through 859 as though fully alleged in this cause of action. 866. From a date unknown to California and continuing to the present, Meta has engaged in and continues to engage in acts or practices that constitute violations of California Business and Professions Code section 17500 et seq., by making or causing to be made untrue or misleading statements with the intent to induce members of the public to use Meta’s platforms when such statements were likely to mislead members of the public about the nature and safety of Meta’s platforms. Meta’s untrue or misleading representations include, but are not limited to, the representations described in paragraph 846. 867. At the time the untrue or misleading representations were made, Meta knew or by the exercise of reasonable care should have known that the representations were untrue or misleading. Continue Reading . Here About HackerNoon Legal PDF Series: We bring you the most important technical and insightful public domain court case filings. This court case 4:23-cv-05448 retrieved on October 25, 2023, from is part of the public domain. The court-created documents are works of the federal government, and under copyright law, are automatically placed in the public domain and may be shared without legal restriction. Washingtonpost.com

The Pursuit of Veracity: California Accuses Meta of Touting Falsehoods

The United States v Meta Platforms Court Filing October 24, 2023 is part of . You can jump to any part in this filing . This is part 35 of 100. HackerNoon’s Legal PDF Series here d. Users under age 13 are an “intended audience” of Instagram. 771. Under 16 C.F.R. § 312.2, “evidence regarding the intended audience” of an online service is relevant to determining whether an online service, or a portion thereof, is directed to children. A platform is “directed to children” if it targets children as “one of its audiences—even if children are not the primary audience.”[38] 772-777. [Redacted] 778. [Redacted] Gen Alpha is the generation with birth years beginning in the 2010s (in 2020, this would have included children aged 10 and under). 779-786. [Redacted] 787. At times, Meta even publicly acknowledges that focusing on Instagram users under age 13 is a part of its business strategy. 788. For example, in September 2018, Meta released a “guide” for parents, urging them to allow their children to join Instagram, lest the children risk “social marginalization.” Commentators noted that the guide suggested that children under the age of 13 already use Instagram 789-791. [Redacted] 792. Despite its widespread efforts to secure the market of users under the age of 13, externally, Meta denies that it designs Instagram to appeal to children; for example, Davis testified to Congress on September 30, 2021 that Meta’s Platforms are not designed for children 12 and under. 793. Meta’s communications expressing its intent to reach under-13 users are one way that it reveals itself to be “directed to children.” [38] July 2020 COPPA Guidance, supra note 35. Continue Reading . Here About HackerNoon Legal PDF Series: We bring you the most important technical and insightful public domain court case filings. This court case 4:23-cv-05448 retrieved on October 25, 2023, from is part of the public domain. The court-created documents are works of the federal government, and under copyright law, are automatically placed in the public domain and may be shared without legal restriction. Washingtonpost.com

Meta Under Investigation for Marketing Strategy Targeting Gen Alpha 

The United States v Meta Platforms Court Filing October 24, 2023 is part of . You can jump to any part in this filing . This is part 30 of 100. HackerNoon’s Legal PDF Series here B. Meta does not comply with COPPA with respect to Instagram. 1. Meta possesses actual knowledge of children on Instagram and collects their personal information without obtaining parental consent. 642. Meta is subject to COPPA’s verifiable parental consent requirement, among other reasons because it collects the personal information of users under the age of 13 on Instagram despite having “actual knowledge that it is collecting personal information from [children].” 15 U.S.C. § 6502(a)(1). 643. Publicly, for example in congressional testimony provided by Meta executive Antigone Davis on September 30, 2021, Meta has downplayed its actual knowledge of under-13 users on Instagram by pointing out that its terms of service nominally disallow use of Instagram by under-13 users—and that, in recent years, Meta has prompted users to self-report that they are at least 13 years old. 644. Despite Meta’s efforts to avoid its responsibilities under COPPA by attempting to maintain willful ignorance of its users under the age of 13, Meta routinely obtains actual knowledge of under-13 users on Instagram. 645-647. [Redacted] a. [Redacted] 648. Despite its public-facing claims that users under the age of 13 are not allowed on Instagram, including in congressional testimony provided by Meta executive Davis in September 2021, [Redacted] 649-655. [Redacted] 656. Despite possessing actual knowledge of Instagram users under the age of 13 [Redacted], Meta refuses to obtain verifiable parental consent as required by COPPA for users under the age of 13. 657-659. [Redacted] 660. [Redacted] Meta did not obtain verifiable parental consent for its ongoing collection of personal information from those users. 661-663. [Redacted] 664. Externally, Meta denies that it strives to attract underage users to its Platforms. For example, in September 2021, in response to a Wall Street Journal article regarding underage users on Instagram, Meta provided a written statement claiming that “[l]ike all technology companies, of course we want to appeal to the next generation, but that’s entirely different from the false assertion that we knowingly attempt to recruit people who aren’t old enough to use our apps.” 665-666. [Redacted] 667. [Redacted] for their children under the age of 13 does not satisfy Meta’s obligation to obtain verifiable parental consent under COPPA for the collection and use of the child’s personal information. [Redacted] does not provide parents with the notices required by COPPA, including notices of what personal information Meta is collecting from their children, nor does it satisfy COPPA’s requirement to ensure that the person providing consent is actually the parent of the child. 668. [Redacted] 669. In this instance and many others, Meta did not meaningfully enforce its nominal age restriction on Instagram, despite its external claims to the contrary, including in Davis’s September 30, 2021 congressional testimony, in which she stated that “we will remove [underage accounts].” 670-677. [Redacted] 678. Meta collects information that identifies accounts on its various Social Media Platforms that belong to the same individual user. For example, Meta collects email addresses from users when they set up new Facebook and Instagram accounts. Meta instructs its Instagram and Facebook users to provide an email address “that only you can access.” In or around September 2020, Meta released a feature called “Accounts Center,” which allows users to link their accounts on Facebook, Instagram, and Messenger using a single sign on. 679-681. [Redacted] 682. [Redacted] Externally, however, Meta claimed in a July 27, 2021 post, titled “How Do We Know Someone Is Old Enough to Use Our Apps?,” that it used a users’ age stated to either Platform to gate users’ access to both Platforms. 682-684. [Redacted] 685. Meta deceives the public regarding its policies when underage accounts are reported. If someone reports that an account belongs to an individual under the age of 13, Meta claims on its Instagram Help Center that “we will delete the account if we can’t verify the account is managed by someone over 13 years old.” [Redacted] Zuckerberg told Congress on March 25, 2021, “if we detect that someone might be under the age of 13, even if they lied, we kick them off.” 686-695. [Redacted] 696. [Redacted]; this research, provided by an organization called Thorn, revealed that of children ages 9-12, 45% used Facebook and 40% used Instagram daily. 697. [Redacted] In this way, Meta goes through great lengths to avoid meaningfully complying with COPPA, looking for loopholes to excuse its knowledge of users under the age of 13 and maintain their presence on the Platform. 698. [Redacted] e. Meta knows that Instagram’s lack of effective age-gating elicits false self-reports of children’s ages 699. For most of its history, up until December 2019, Instagram did not require new users to disclose their age or date of birth in order to create an Instagram account. During that time, Meta did not require users to take even the minimal step of self-attesting that they were over the age of 13. Instead, for over seven years, under-13 users faced no practical obstacles to creating accounts on Instagram. 700-701. [Redacted] 702. Eventually, in response to pressure from regulators and the public, Meta purported to implement an age gate as part Instagram’s account registration process—but the term “gate” was a misnomer because it did not prevent under-13 users from creating and using Instagram accounts. 703. To the contrary, Meta initially designed its age gate in a way that prompted all users, including children under the age of 13, to provide an age over 13. Specifically, Meta’s sign-up page contained a drop-down menu that automatically generated a date and year of birth representing the user to be 13 years old. The design of the age gate signaled to children the specific date that they could affirm to advance through the registration process, even though the date automatically populated by Instagram was not their actual date of birth. 704. Meta knew that its use of a sign-up page automatically generating a date 13 years prior to the date of registration aided under-13 users in misrepresenting their age in order to access Instagram. 705. “[E]ncourag[ing] children to falsify their ages to gain access” is impermissible under COPPA. See COPPA July 2020 Guidance § H(3).[35] 706. Meta only recently changed Instagram’s sign-up page to automatically generate the instant date and year, rather than a date 13 years prior. 707. Meta’s adoption of an age gate that permits the user to enter any date of birth, regardless of its accuracy, still does not prevent under-13 users from using Instagram. 708. [Redacted] 709. In sum, Meta’s age gate efforts for Instagram have long been ineffective: first, Instagram utilized no age gate for several years, then implemented an age gate that defaulted to an entry of ages over 13, and now uses an age gate that still depends on an under-13 user to correctly self-report their own age, without any verification. 710. [Redacted] 711. [Redacted] Nonetheless, Meta externally touts its age-gating as an effective means to keep children under the age of 13 off Instagram and Facebook. 712. While testifying before Congress on September 30, 2021, Meta executive Davis stated: “if we see someone trying to, repeatedly, change the [birth] date to get past that [age screen], we actually will restrict their ability to access the app.” [Redacted] 713-722. [Redacted] 723. [Redacted] While Meta externally claimed that it built under-13 products for kids and tweens in a September 27, 2021 post on Instagram’s website entitled “Pausing ‘Instagram Kids’ and Building Parental Supervision Tools,” [Redacted] 724. [Redacted] 725. Meta has access to, and chooses not to use, feasible alternative age verification methods that would significantly reduce or eliminate the number of underage users on Meta’s Social Media Platforms, for example, by requiring young users to submit student IDs upon registration. 726-728. [Redacted] 729. Externally, Meta misleads the public by claiming, in a March 17, 2021 post on Instagram’s website, that “we know that young people can lie about their date of birth. We want to do more to stop this from happening,” [Redacted] 730. Rather than excluding under-13 users from Instagram, Meta could alternatively comply with COPPA by obtaining informed parental consent after providing notice to parents of its intent to collect and use children’s personal information. Meta chooses not to do so. 731. Despite knowing that its lack of age gates and later implementation of minimal age gate designs have allowed users under age 13 onto Instagram, Meta does not obtain verifiable parental consent before collecting the personal information of those users who routinely register for, and provide their personal information to, Instagram. f. [Redacted] 732-736. [Redacted] 737. Meanwhile, Meta publicly maintains that it does not allow under-13 users on its Platforms, relying primarily on its faulty age-collection at sign-up [Redacted] 738-742. [Redacted] 743. Meta externally claimed, through congressional testimony provided by Mosseri on December 8, 2021, that “we train our technology to identify if people are above or below 18 using multiple signals,”—including birthday posts—and that Meta is building new technology to do the same for users under 13. [Redacted] 744. Former Meta Director of Site Integrity and former consultant to Meta Bejar testified that Meta does not meaningfully utilize birthday posts to identify users who claim to be over 13 years old but are not. In fact, Meta’s internal reporting mechanism for using birthday posts is complicated which prevents most reports from reaching “completion,” or the point where a person successfully submits a report to Instagram. 745. Despite Meta’s actual knowledge, acquired through its possession and review of estimated age data and/or acquired through other sources, that Meta collects personal information of users under the age of 13 in the ordinary course of its operations, Meta does not obtain verifiable parental consent for under-13 users. [35] Complying with COPPA: Frequently Asked Questions, Fed. Trade Comm’n (July 2020), https://archive.ph/PEj8q (hereinafter “July 2020 COPPA Guidance”). Continue Reading . Here About HackerNoon Legal PDF Series: We bring you the most important technical and insightful public domain court case filings. This court case 4:23-cv-05448 retrieved on October 25, 2023, from is part of the public domain. The court-created documents are works of the federal government, and under copyright law, are automatically placed in the public domain and may be shared without legal restriction. Washingtonpost.com

Meta Monetizes the Personal Information of Minors on Instagram

The United States v Meta Platforms Court Filing October 24, 2023 is part of . You can jump to any part in this filing . This is part 29 of 100. HackerNoon’s Legal PDF Series here A. COPPA requires Meta to obtain verifiable parental consent for Instagram and Facebook users under the age of 13. 633. COPPA prohibits social media companies like Meta from collecting personal information from children without first obtaining verifiable parental consent if: (a) the operator of the social media platform has actual knowledge that it is collecting personal information from a child; or (b) the operator’s service is directed to children. 15 U.S.C. § 6502(a)(1). 634. Meta’s operation of Instagram and Facebook is subject to COPPA’s verifiable parental consent requirement under both of the two statutory triggers: (a) Meta routinely obtains actual knowledge that users on Instagram and Facebook are under 13 years old; and (b) Meta targets children as users of Instagram and Facebook, making the Platforms directed to children. See 16 C.F.R. § 312.2. 635. The term “child” is defined by 15 U.S.C. § 6501(1) to mean an individual under the age of 13. The terms “child,” “children,” “under-13 user(s),” “U13 users,” and “child-users” herein refer to children under the age of 13. 636. “Verifiable parental consent” requires, at a minimum, providing a child’s parent with notice of Meta’s “personal information collection, use, and disclosure practices” and further requires Meta to obtain the parent’s authorization for Meta to collect, use, or disclose the child’s information. Both of these requirements must be completed before Meta may collect a child’s information. 15 U.S.C. § 6501(9). 637. Meta does not obtain—or even attempt to obtain—verifiable parental consent before collecting the personal information of children on Instagram and Facebook. “Personal information” is defined by statute and regulation to mean “individually identifiable information about an individual collected online,” including the child’s name, address, email address, personal identifiers, geolocation information, and photographs or videos of the child, among other categories of information. 15 U.S.C. § 6501(8); 16 C.F.R § 312.2. Meta collects personal information in these categories from all registered users of Instagram, including children. 638. Instead of obtaining verifiable parental consent, Meta relies on Instagram’s and Facebook’s nominal bans on under-13 users to avoid any responsibility under COPPA to its under-13 users and their parents. [Redacted] Meta is not exempt from COPPA. 639. COPPA empowers State Attorneys General to bring suit against companies that violate the verifiable parental consent requirement. COPPA permits State Attorneys General to obtain injunctive relief, damages, restitution, and other relief on behalf of residents of their States. 15 U.S.C. § 6504(a)(1). 640. COPPA also requires the FTC to promulgate regulations consistent with the statute’s verifiable parental consent requirement as well as the “actual knowledge [of a] child” and “directed to children” statutory triggers. 15 U.S.C. § 6502(b). The FTC has promulgated such regulations as the Children’s Online Privacy Protection Rule. See 16 C.F.R. § 312.1 et seq. (COPPA Rule) (last promulgated Jan. 17, 2013). 641. Under COPPA and the COPPA Rule, Meta is subject to COPPA’s “verifiable parental consent” requirement—but Meta flouts its obligations under COPPA with respect to its operation of Instagram and Facebook. Continue Reading . Here About HackerNoon Legal PDF Series: We bring you the most important technical and insightful public domain court case filings. This court case 4:23-cv-05448 retrieved on October 25, 2023, from is part of the public domain. The court-created documents are works of the federal government, and under copyright law, are automatically placed in the public domain and may be shared without legal restriction. Washingtonpost.com

Meta Flouts the Children's Online Privacy Protection Act

The United States v Meta Platforms Court Filing October 24, 2023 is part of . You can jump to any part in this filing . This is part 28 of 100. HackerNoon’s Legal PDF Series here IX. META’S COPPA NONCOMPLIANCE 631. The Children’s Online Privacy Protection Act of 1998 (COPPA) protects the privacy of children by requiring technology companies like Meta to obtain informed consent from parents prior to collecting the personal information of children online. 632. Meta routinely violates COPPA in its operation of Instagram and Facebook by collecting the personal information of children on those Platforms without first obtaining (or even attempting to obtain) verifiable parental consent, as required by the statute. Continue Reading . Here About HackerNoon Legal PDF Series: We bring you the most important technical and insightful public domain court case filings. This court case 4:23-cv-05448 retrieved on October 25, 2023, from is part of the public domain. The court-created documents are works of the federal government, and under copyright law, are automatically placed in the public domain and may be shared without legal restriction. Washingtonpost.com

Stories, News & Mentions about META

Meta’s Stories on HackerNoon

Meta's News & Mentions