Eduardo C. Garrido Merchán, PhD’s Post

By now, you will have heard the news. Llama 3.1 405B came out yesterday with a similar performance to GPT-4o and Claude 3.5 on some measures. Maybe not as good across the board, but it is still the most impressive open source model. They also launched improvements to their existing 8B and 70B models. It all sounds great. But, there's a fight coming. Meta recently decided not to launch its 405B model in the EU. The reason? The unpredictable nature of EU regulations. This decision means European companies won't get to use Meta's most powerful model. The EU's strict new AI Act has set compliance deadlines for tech companies until August 2026. This includes rules around copyright, transparency, and AI uses like predictive policing. It's not just Meta. Apple also expressed concerns over the Digital Markets Act and may exclude the EU from its AI rollout. Meta has even paused its generative AI tools in Brazil due to data protection compliance issues. However, Meta's Llama 3 model will still launch in the EU as a text-only version. This leaves companies outside the EU in a tough spot, as they can't offer these models in one of the world's largest markets. It feels like it might be a game of chicken. The big tech firms need the EU. And the EU doesn't want to fall behind technologically. The EU hasn't commented on Meta's decision yet, but this move highlights the tension between tech innovation and regulatory compliance. It's a tricky balance to strike, and companies are clearly feeling the pressure. You can read more: https://lnkd.in/d8gDufHG #AI #EU #Regulations #Meta #LLaMA

𝐌𝐞𝐭𝐚 𝐮𝐧𝐯𝐞𝐢𝐥𝐬 𝐭𝐡𝐞 𝐦𝐨𝐬𝐭 𝐩𝐨𝐰𝐞𝐫𝐟𝐮𝐥 𝐨𝐩𝐞𝐧 𝐬𝐨𝐮𝐫𝐜𝐞 𝐀𝐈 𝐦𝐨𝐝𝐞𝐥 𝐞𝐯𝐞𝐫 🤖 How different could Llama 3.1 really be from 3? Well, it turns out there’s a lot hidden behind that subtle name change. Meta’s latest model now boasts 405 billion parameters, up from just 70 billion previously. The update has made Llama an unexpected powerhouse, beating GPT-4o and Claude 3.5 across several key benchmarks. It’s reportedly the first time an open-source release can be described as a “frontier model” — meaning it’s at the cutting edge of AI capabilities and performance. Llama’s 8b and 70b models are getting an efficiency boost, too. 𝐖𝐡𝐚𝐭 𝐦𝐚𝐤𝐞𝐬 𝐌𝐞𝐭𝐚’𝐬 𝐦𝐨𝐝𝐞𝐥 𝐝𝐢𝐟𝐟𝐞𝐫𝐞𝐧𝐭? OpenAI, Anthropic, and Alphabet are each trying to build one LLM to rule them all. Meta says it has a different vision: It’s betting that businesses and individuals alike are craving models that they can customize to perfectly match their needs. If its prediction proves accurate, there could eventually be millions of LLMs floating around in the cloud — with each designed for very specific uses. 𝐇𝐨𝐰 𝐢𝐭 𝐩𝐥𝐚𝐧𝐬 𝐭𝐨 𝐠𝐞𝐭 𝐭𝐡𝐞𝐫𝐞: The biggest differentiating factor is that Meta’s model is open source, allowing developers to view and tweak its source code. After enough adjustments, those optimized versions can even outperform the original base model. The update is especially exciting for independent developers, who won’t have to wait to get their hands on the latest and greatest model: It’s already freely available to them. 𝐀𝐧𝐨𝐭𝐡𝐞𝐫 𝐝𝐢𝐦𝐞𝐧𝐬𝐢𝐨𝐧: Meta’s also building out an ecosystem so that major companies from Nvidia to Amazon can collaborate on new projects alongside smaller startups and bootstrapped developers. 𝐀 𝐝𝐚𝐫𝐢𝐧𝐠 𝐩𝐫𝐞𝐝𝐢𝐜𝐭𝐢𝐨𝐧: Thanks to the latest performance boost (plus, some fun new features like a new ‘Imagine Me’ selfie tool), Zuckerberg thinks that more people will be using Llama by the end of the year than OpenAI’s latest model. That might just be within reach, especially since visits to ChatGPT’s webpage have plummeted over the past few months. With GPT-5 looming over the horizon though, only time will tell if Zuckerberg’s forecast turns out to be accurate. #zuckerberg #llama #openai #google #gemini #claude #palm #lambda #bloom #roberta 𝘴𝘰𝘶𝘳𝘤𝘦 𝘚𝘶𝘱𝘦𝘳𝘩𝘶𝘮𝘢𝘯 𝘈𝘐

🚀 Meta's Llama 3.1: A Game-Changer in Open Source AI 🦙 Meta just dropped a bombshell in the AI world with Llama 3.1, their "frontier-level" open source model. This isn't just another update - it's a paradigm shift. Why it matters: • Democratizes advanced AI tech • Challenges closed AI systems • Offers better cost-efficiency • Enhances customization & data security Mark Zuckerberg boldly predicts: "Starting next year, we expect future Llama models to become the most advanced in the industry." But here's the kicker: Open source AI isn't just about accessibility. It's about safety through transparency and collective scrutiny. As the AI race heats up, could this open approach be the key to responsible innovation? The tech giants are already lining up. Amazon, NVIDIA, and others are jumping on board to support this open ecosystem. It's not just a trend; it's a movement. As we stand at this AI crossroads, I can't help but wonder: Will Llama 3.1 mark the beginning of the end for closed AI systems? Are we witnessing the Linux moment of artificial intelligence? What's your take? Is open source the future of AI, or are there risks we're overlooking? #OpenSourceAI #AIInnovation #TechTrends #FutureOfAI #DigitalTransformation

Meta announced yesterday the release of Llama 3, the next generation of their state-of-the-art large language model. Some of the main points about the Llama 3 release: - Pre-trained and instruction-fine-tuned language models with 8B and 70B parameters, demonstrating state-of-the-art performance on various industry benchmarks and offering improved reasoning capabilities. - Collaboration with major cloud platforms and hardware providers to make Llama 3 widely accessible, including the usual set: AWS, Google Cloud, Microsoft Azure along with Databricks, HuggingFace, Kaggle and quite a few others. - Enhanced trust and safety tools, such as Llama Guard 2, Code Shield, and CyberSec Eval 2, to ensure responsible development and deployment of the models. This kind of tooling will be key for any large firm deploying an LLM, you can’t just rely on the model to self regulate, you’ll need to employ your own checks and balances. - Improved tokeniser efficiency and the introduction of Group Query Attention (GQA) to maintain inference efficiency despite the increased parameter count. - A commitment to expanding Llama 3's capabilities in the coming months, including multimodality, multilingual support, longer context windows, and overall performance enhancements. Meta's (fairly) open approach to releasing Llama 3 reflects their belief in the power of collaboration and innovation within the AI community. However as per the previous models (and most other open models) the training data to allow you to build your own is not available, and likely never will be. License wise, it’s very similar to the previous license - you’re able to use it commercially unless you have over 700m monthly active users, which most of us should happily fall under. The introduction of Llama 3 also coincides with the launch of Meta AI, an AI assistant powered by Llama 3 technology, which is now available across various Meta apps and platforms - however not yet in the UK, though it will be interesting to see how bringing AI generation abilities to our social feeds will make things any better. As the Llama project continues to evolve, we can expect to see even more impressive capabilities and applications emerge, with the commitment to a multimodal model in the future the possibilities within legal tech will continue to grow. #generativeai #meta #llama3

Mark Zuckerberg, Yann LeCun and others are trying to seduce the EU, again. A very good summary by Jürgen Geuter. The key takeaway is: «The signatories [of this open letter] demand "quick and clear decisions under EU data regulations that enable European data to be used in AI training". [...] It's about trying to get the EU to let big companies scrape everyone's data and work for their models. And they want the EU to sign off on it.»  #AI #EU #MachineLearing #LLM #Data #Extraction #Meta

The latest from the great Brian Merchant: "Again we see why it’s crucial that hype is maintained at near-apocalyptic levels; if it’s not getting results for a certain use case, well maybe I’m just not using it right? This is the anointed technology right now, and Microsoft, Google, Meta, and $100 billion of valuation couldn’t be wrong, right? If that hype drifts away, we may be left with an impressive but quirky technology, largely trained on the work of artists and writers who were not paid or consulted, and that has a few niche use cases for automating away the ones still currently trying to make a living. Considering how expensive and resource intensive the technology is to run, well, that’s not going to be worth $100 billion, and it could all fall to earth — and take a hell of a lot with it in the process." https://lnkd.in/epxa7by3

Meta’s AI strategy with Llama models mirrors Microsoft’s approach under Satya Nadella, who converted closed-source software into open-source and transformed competitors into partners and the rest is historry. Meta’s open-source Llama models not only enhance their own products but also create opportunities for developers to build on them, fostering innovation and collaboration. By doing so, Meta positions itself as a formidable AI player, building long-term partnerships and future revenue streams. This open-source strategy may prove to be a powerful differentiator. However, with Google AI integrated into the Android platform and OpenAI expected to integrate into iPhone, a new layer of dynamic competition is emerging, pushing Meta to continually innovate in this rapidly evolving space.

#Topics Meta joins Apple in withholding AI models from EU users [ad_1] Meta has announced it will not be launching its upcoming multimodal AI model in the European Union due to regulatory concerns. This decision from Meta comes on the heels of Apple’s similar move to exclude the EU from its Apple Intelligence rollout, signalling a growing trend of tech giants hesitating to introduce advanced AI technologies in the region. Meta’s latest multimodal AI model – capable of handling video, audio, images, and text – was set to be released under an open license. However, Meta’s decision will prevent European companies from utilising this technology, potentially putting them at a disadvantage in the global AI race. “We will release a multimodal Llama model over the coming months, but not in the EU due to the unpredictable nature of the European regulatory environment,” a Meta spokesperson stated. A text-only version of Meta’s Llama 3 model is still expected to launch in the EU. Meta’s announcement comes just days after the EU finalised compliance deadlines for its new AI Act. Tech companies operating in the EU will have until August 2026 to comply with rules surrounding copyright, transparency, and specific AI applications like...

𝗔𝗜 𝗮𝗻𝗱 𝗘𝗨: 𝗧𝗵𝗲 𝗗𝗲𝗹𝗶𝗰𝗮𝘁𝗲 𝗗𝗮𝗻𝗰𝗲 𝗕𝗲𝘁𝘄𝗲𝗲𝗻 𝗜𝗻𝗻𝗼𝘃𝗮𝘁𝗶𝗼𝗻 𝗮𝗻𝗱 𝗥𝗲𝗴𝘂𝗹𝗮𝘁𝗶𝗼𝗻 🤖💼 In the tech world, regulations and innovative companies have long been dancing an 𝙖𝙬𝙠𝙬𝙖𝙧𝙙 𝙩𝙖𝙣𝙜𝙤. The latest tune? Meta's decision to sit out the European Union's AI party with its upcoming multimodal Llama AI model. 𝗞𝗲𝘆 𝗜𝗻𝘀𝗶𝗴𝗵𝘁𝘀: 🔍 𝗥𝗲𝗴𝘂𝗹𝗮𝘁𝗼𝗿𝘆 𝗨𝗻𝗰𝗲𝗿𝘁𝗮𝗶𝗻𝘁𝘆: Meta cites the "unpredictable nature of the European regulatory environment" as the primary reason for this decision. 🤝 𝗡𝗼𝘁 𝗔𝗹𝗼𝗻𝗲 𝗶𝗻 𝘁𝗵𝗲 𝗥𝗶𝗻𝗴: Apple recently made a similar move, likely excluding the EU from its Apple Intelligence rollout due to concerns about the Digital Markets Act. 🌍 𝗜𝗺𝗽𝗹𝗶𝗰𝗮𝘁𝗶𝗼𝗻𝘀: This move will prevent companies from using the multimodal Llama 3 model in one of the world's largest economic markets. As we navigate this complex landscape of innovation and regulation, it's crucial for policymakers and tech companies to find common ground. Regulators need to work more closely with tech companies to create frameworks that protect consumers without stifling innovation. #AI #Regulation #TechInnovation #EuropeanUnion #MetaAI

I’m not sure ‘unpredictable’ is the correct word to use in the quote below… but a sign of a messy future, geographically, for development and adoption of AI tools 👇 ‘“We will release a multimodal Llama model over the coming months, but not in the EU due to the unpredictable nature of the European regulatory environment,” Meta spokesperson Kate McLaughlin said to The Verge. Meta’s decision follows a similar move by Apple, which recently said it would likely exclude the EU from its Apple Intelligence rollout due to concerns surrounding the Digital Markets Act. Meta has also halted plans to release its AI assistant in the EU and paused its generative AI tools in Brazil — both due to concerns raised about data protection compliance.’ https://lnkd.in/e_8H26An