• AI KATANA
  • Posts
  • Meta quietly building a new AI model

Meta quietly building a new AI model

Also: AI-generated videos to drive more demand for Nvidia chips, CEO Huang says

Author

AI KATANA
May 23, 2024

Hi!

In today's newsletter, we delve into the latest developments in AI technology and its expanding influence across various sectors. Meta is advancing its AI capabilities with the new multi-modal model, Chameleon, promising to rival top-tier models like GPT-4. Amazon and AI startup Hugging Face have partnered to leverage Amazon's custom Inferentia2 chips, enhancing AI model efficiency on AWS. Nvidia's CEO anticipates increased demand for their processors due to AI-generated videos. We also address the environmental impact of AI's growing energy consumption and Microsoft's new "Recall" feature in Windows 11, which logs all user activities. Additionally, TikTok introduces innovative AI tools for advertisers, and the Schumacher family wins a legal battle over a fake AI-generated interview. In the venture capital scene, AI music firm Suno raises $125 million to further its generative music technology.

Sliced:

  • 🤫 Meta quietly building a new AI model

  • 🤝 Amazon, AI startup Hugging Face pair to use Amazon chips

  • 🎥 AI-generated videos to drive more demand for Nvidia chips, CEO Huang says

  • ⚡️ AI is an energy hog. This is what it means for climate change.

  • 🏎️ Michael Schumacher’s family wins legal case against publisher over fake AI interview

Meta is developing a state-of-the-art AI model named Chameleon, designed as a multi-modal large language model similar to Google's Gemini. This model integrates various inputs like images, text, and code from the beginning, using an "early-fusion token-based mixed-modal" architecture, contrasting with the "late fusion" method that combines separate single-modality models. Chameleon reportedly competes with or surpasses models like GPT-4 and Meta’s Llama-2. The development aligns with Meta's broader open-source strategy, as seen with their Llama 3 model and Quest headset OS. Meta's advancements in AI reflect their commitment to pushing the boundaries of multi-modal AI technology.

Amazon's cloud unit, Amazon Web Services (AWS), has partnered with AI startup Hugging Face to facilitate the operation of AI models using Amazon's custom computing chips, Inferentia2. Hugging Face, valued at $4.5 billion, is a key platform for AI researchers and developers, backed by major tech companies like Amazon, Google, and Nvidia. The collaboration aims to enhance the efficiency and cost-effectiveness of running AI models, making it easier for developers to deploy their AI solutions on a large scale. AWS hopes to attract more AI developers to its cloud services by offering lower costs for running AI inferences, leveraging the capabilities of Inferentia2 chips to handle the high-frequency inference demands that follow the initial model training.

Nvidia CEO Jensen Huang predicts that AI-generated videos and human-like voice interactions will significantly boost demand for Nvidia's graphics processors. This expectation follows a surge in demand driven by the adoption of chatbots by major tech companies. Nvidia's advanced chips, such as the Grace Hopper H200, are increasingly being used for training and running complex AI systems, including those by OpenAI, Google DeepMind, and Meta Platforms. The automotive industry's push for AI in autonomous driving, exemplified by Tesla's use of 35,000 H100 processors, further underscores the expanding market. Nvidia's strong revenue forecasts and the rising need for multimodal AI models highlight the ongoing growth and importance of their hardware in powering future AI advancements.

AI's rapid integration into various sectors has led to significant concerns about its rising electricity demands and the potential impact on climate change. Generative AI models, such as those used to create images, are particularly energy-intensive, with the production of 1,000 images consuming as much energy as driving a gas-powered car for over four miles. While not all AI tasks are equally demanding, the cumulative effect is substantial. The International Energy Agency projects that by 2026, electricity consumption from data centers, AI, and cryptocurrency could double, possibly adding the equivalent demand of a small country like Sweden or even a larger one like Germany to the global grid. Tech companies like Microsoft, despite aiming for net-zero emissions, are witnessing rising emissions due to the expansion of data centers necessary for AI infrastructure. The challenge lies in balancing the increasing electricity needs with sustainable energy sources to mitigate AI's environmental footprint. As AI continues to grow, improving energy efficiency and investing in renewable energy will be crucial to minimizing its climate impact.

The family of Michael Schumacher has won a legal case against the publisher of the German magazine Die Aktuelle, which had printed an AI-generated interview with the former Formula 1 champion. The magazine falsely advertised the interview as Schumacher's first since his near-fatal skiing accident in 2013, which left him with severe brain injuries. The family was awarded 200,000 euros in compensation, and the publisher, Funke Mediengruppe, has since apologized and dismissed the chief editor responsible. Schumacher has been receiving private care at his home in Switzerland since 2014.

🛠️ AI tools updates

TikTok has launched a suite of new AI tools aimed at enhancing advertising for brands. Dubbed TikTok Symphony, the suite includes an AI video generator that simplifies the ad creation process by generating videos with minimal input. Additionally, TikTok introduced an AI assistant for advertisers, a centralized marketing hub called TikTok One, and predictive AI performance solutions designed to optimize business outcomes like sales and app installs. These tools were unveiled at TikTok's fourth annual TikTok World product summit, highlighting the platform's commitment to innovative advertising solutions. Despite these advancements, TikTok is also dealing with significant legal and operational challenges, including a legal battle with the U.S. government and potential global layoffs.

Microsoft has unveiled a new AI-powered feature called "Recall" for Windows 11 users that records all activities on their PCs, taking images of the active screen every few seconds. Announced at the Build conference, Recall is part of the Copilot+ suite and aims to help users retrieve past activities, including app usage, meetings, and web research. The data is encrypted and stored locally, but the feature raises significant privacy concerns. Users can manage the data by pausing, stopping, or deleting captured content and excluding specific apps or websites. Recall requires advanced hardware, specifically Qualcomm's Snapdragon X Elite chips, and is currently in preview status for feedback and further development.

💵 Venture Capital updates

Suno, an AI music company specializing in generative technology, has secured $125 million in its latest funding round. This firm, known for its ability to generate complete songs from simple text prompts using proprietary technology and OpenAI's ChatGPT for lyrics and titles, aims to democratize music creation. Suno offers various plans, including a free tier and paid options allowing up to 2,000 songs per month with commercial use rights. The funding round saw contributions from Lightspeed Venture Partners, Nat Friedman, Daniel Gross, Matrix, and Founder Collective, among others. Despite its technological advancements, Suno faces scrutiny over potential copyright issues in its training data, raising questions about the legality of using copyrighted materials without explicit licenses. The company emphasizes that its mission is to enhance human creativity through technology.

🫡 Meme of the day

⭐️ Generative AI image of the day