• Bits & Neurons
  • Posts
  • NVIDIA announced the H200 for 2024 - featuring faster 141GB HBM3E memory

NVIDIA announced the H200 for 2024 - featuring faster 141GB HBM3E memory

📰 No-Noise News 📰

🚀 The UN General Assembly approved a resolution for international law to govern lethal AI weapons, but did not ban them. While the move reflects growing awareness and desire for regulation amidst advancing technologies, key military powers, including the U.S., support the resolution, showing a global divide, as India and Russia, among others, voted against it.

🦾 Nvidia is launching the H200 GPU with enhanced 141GB memory and speed, as the AI industry grapples with a shortage of the previous H100 chips. It retains compatibility with existing H100 systems, ensuring an easy upgrade path for cloud providers and high-performance computing facilities.

🌐At the APEC summit, Presidents Joe Biden and Xi Jinping are set to agree on a deal that will restrict the use of artificial intelligence in the control systems for nuclear weapons, as tensions between the U.S. and China have been on the rise. This agreement is supposed to prevent AI from autonomously initiating nuclear actions, as both nations recognize the dangers of unregulated military AI applications.

🚗 UK legislation proposes prison for misleading self-driving car ads, aiming to regulate autonomous vehicle safety and accountability. The law aims to regulate autonomous vehicle safety, protecting drivers from the legal ramifications of potential accidents caused by the vehicle’s automated systems.

🤖 Silo, an European AI company has released Poro, a new multilingual model trained on both English and Finnish, which could potentially compete with the large language models from Silicon Valley's Big Tech. The model benefits from access to a vast amount of language data from an EU-funded project, which could give Europe an advantage in developing AI for less commonly used languages.

📚 Nerd section 📚

🎨 LCMs are a new way to generate high-quality images much faster - Latent Consistency Models (LCMs) rapidly produce detailed images, requiring far fewer steps and less computing power than prior methods, such as Latent Diffusion Models.

⚒️ Ride the 'make your own GPTs' wave with these tutorials:

🌐 Web section 🌐 

🕸️ What's this new section about?

On days when there's a lot of news, we aim to keep our newsletter short and sweet.

So here (but not in the email version) we're adding AI-related stories that may be slightly less attention-grabbing, but still could move the needle in specific sectors and fields.

🤑 OpenAI tempts Google's AI whizzes with $10 million rewards. Already recruited 90+ from Google/Meta, seeks technology that serves human goals with big share sale.

🕷️ For the first time, a vulnerability has been found in SSH cryptographic keys caused by rare computational faults, potentially compromising secure computer-to-server connections. A small fraction of these keys, about one in a million, are at risk, but given the vast amount of internet traffic, this could result in numerous SSH host keys being exposed. Most current SSH software includes protections against such issues, but there are still enough unprotected implementations for this to be a concern.

🧒 Facial recognition technology was deployed by Welsh police at a Beyoncé concert, aiming to catch terrorists and pedophiles, but its use sparked debates on privacy, potential discrimination, and regulatory oversight.

🖼️ AI technology is advancing so fast that even experts can't tell real images from fake ones created by AI, a worrying development especially as social media companies relax disinformation policies and reduce teams responsible for policing such content.

Thank you for reading today’s edition!

We love to hear back from you!

Feel free to reply to our emails with questions, suggestions, or topics you'd like to see covered, or drop us a message on Twitter or Facebook.

Until tomorrow,
- Ts (Bits and Neurons)