Legislation and Leaks

WELCOME

Welcome to the Blacklynx Brief !

You will notice that this edition is a lot shorter than our regular editions. That is because in this edition, there is only the most important news in the field of artificial intelligence. All other topics have been scrapped this week as we have not had the time to perform the necessary research.

The Blacklynx Brief went to San Francisco, where we had some interesting conversations about AI and its uses. More on that later in a special edition.

While this edition might be short, a LOT has happened in the past two weeks.

Read the most important developments down below

AI NEWS
Legislation and Leaks

• The European Parliament has passed the Artificial Intelligence Act, introducing the most extensive AI regulations globally. This law prohibits certain AI applications, including biometric and facial recognition for sensitive traits, social scoring, and manipulative AI, while imposing stringent rules on high-risk AI in sectors like infrastructure, education, and employment. Set to take effect in stages, starting six months post-enforcement, the Act has faced criticism from both industry and watchdogs, raising concerns over competitiveness and regulatory sufficiency, respectively.

  
  We plan on delving deeper into this document to see what this legislation actually entails.
  

• Apple has acquired the Canadian AI startup DarwinAI, aiming to enhance its generative AI capabilities in the forthcoming iOS 18 update. DarwinAI specializes in optimizing AI systems for efficiency and speed, particularly in visual inspection tasks. This move aligns with Apple's strategy of advancing in AI quietly, with expectations for significant announcements at the upcoming WWDC event in June, following CEO Tim Cook's hints at groundbreaking AI developments.
  

• Also at Apple, researchers have introduced MM1, a new AI model family integrating visual and language understanding, trained on diverse data, including image captions and texts. The 30B parameter MM1 model excels in learning from a few examples and reasoning across images, performing comparably to leading models like GPT-4V and Gemini Pro. The research being published on an Open-source platform is a significant break with tradition for Apple. Interesting to see where this goes

• Apple is discussing the integration of Google's Gemini AI model into iPhones, potentially available with iOS 18. This would introduce AI-powered features for image creation and text generation. Meanwhile, OpenAI is reportedly in talks with Samsung, indicating significant moves towards AI-enhanced smartphones.

• Elon Musk and xAI have open-sourced Grok-1, their 314B parameter language model, under the Apache 2.0 license. Grok-1 utilizes a Mixture-of-Experts approach for efficiency and is available in its original, unmodified form. Musk has done this purposefully in the wake of the back-and-forth between himself and Sam Altman a few weeks back. Musk accuses OpenAI on focusing on profit rather then on developing AI that will help and better humanity

• In a recent interview on the Lex Fridman Podcast, OpenAI CEO Sam Altman revealed plans to release a new AI model this year and discussed the upcoming GPT-5, highlighting a significant leap in capabilities from GPT-4. He also mentioned Sora, a project aimed at merging AI with the physical world through robotics, and speculated on the future importance of computing power. This conversation hints at imminent, transformative advancements in AI technology. It’s basically a must-watch for any AI enthusiast.

• Leaked information from a Reddit user called “JohnnyApples” suggests OpenAI's GPT-5 will significantly surpass GPT-4 in performance, with demonstrations already provided to select enterprise customers. Additionally, GPT-5 may feature the ability to interact with other AI agents being developed by OpenAI. While its release is aimed for mid-year, this could change depending on training and safety assessments. Before you question my sources - the ‘JohnnyApples’ account has leaked information in the past that turned out to be 100% true. He truly is on the inside.
  

• Nvidia has introduced the next-generation Blackwell GPU architecture and GB200 Superchip, marking substantial advancements in AI and computing efficiency. The Blackwell B200 GPU is deemed the world's most potent chip, with the Superchip offering 30 times the H100 GPU's performance while being 25 times more energy-efficient. These innovations could support the training of AI models up to 10 trillion parameters, with GPT-4's 1.8 trillion parameters being a noted example.

• NVIDIA has announced Project GR00T, a multimodal AI system designed for advanced humanoid robots, enabling them to understand language, mimic human movements, and acquire new skills. This foundation model is supported by partnerships with leading robotics companies and is complemented by the Isaac Sim training tool and the OSMO platform for AI workloads. GR00T represents a significant step toward embodied AI, bridging the gap between high-level AI capabilities and practical robot applications.

  Few weeks ago the news was announced that OpenAI is also working on physical robots. Clear trends in that direction.

• The Biden administration is allocating a $20B package to Intel to enhance U.S. semiconductor production, representing the largest government investment in AI to date. This funding, comprising $8.5B in grants and up to $11B in loans, aims to support the construction of advanced chip factories and R&D expansion, anticipated to create over 30,000 jobs. This move aligns with the goal of the U.S. producing 20% of the world's advanced AI chips by 2030, underlining the strategic importance of the semiconductor industry in global technological leadership.

That's it from us - as I said it’s a pretty short one.

We do have a big announcement coming up that you do NOT want to miss.