• AI Rundown
  • Posts
  • The AI Rundown by Lightscape Partners - 09/09/24

The AI Rundown by Lightscape Partners - 09/09/24

xAI announces a 100,000 GPU AI cluster, SSI raises a $1 billion Seed, and Nvidia responds to subpoena rumors

Image credit: TSMC

Good morning and welcome back to another week of the AI Rundown by Lightscape Partners.

Last week in AI:

  • xAI announces its largest AI cluster yet with 100,000 Nvidia H100s. The launch positions Elon Musk’s AI company to begin training the next state-of-the-art models.

  • Ilya Sutskever, OpenAI Co-founder, announces his AI safety startup has raised $1 billion at a $5 billion valuation to begin training safe AGI. The announcement follows Sutskever’s departure from his old company a few months back amid rumors that AI safety has taken a “back seat” at OpenAI.

  • In a statement released by the company, Nvidia claims it was never subpoenaed by the U.S. DOJ amid the antitrust investigation. Nvidia received a “less formal” civil investigative demand (CID). The investigation is still ongoing.

If you haven’t yet, please support the newsletter by subscribing!

Technology Developments and Innovations

Elon Musk’s xAI launches Colossus, the world’s most powerful AI cluster. Link.

  • Elon Musk’s xAI has launched Colossus, an AI training system featuring 100,000 Nvidia H100 graphics cards, positioning it as the world’s most powerful AI training system.

  • Colossus is faster than the U.S. Energy Department’s Aurora, which reached 10.6 exaflops. xAI plans to double the chip count to 200,000, including 50,000 of Nvidia’s newer H200 chips, which boast faster data transfer speeds and increased memory capacity.

  • Colossus is expected to significantly enhance the capabilities of xAI’s language models, including the upcoming successor to its current flagship, Grok-2.

OpenAI plans to build AI chips on TSMC's advanced 1.6 nm A16 process node. Link.

  • OpenAI is moving forward with plans to develop its own AI chips using TSMC's upcoming 1.6 nm A16 process node, which features advanced gate-all-around nanosheet transistors and Super Power Rail technology.

  • This decision diverges from OpenAI's initial plan to use TSMC’s N5 node, aiming instead at a more cutting-edge process still in development.

  • While building proprietary AI chips may reduce costs in the long term, the project faces uncertainties, including OpenAI's financial pressures and competition in the AI market.

  • Potential collaborations with companies like Broadcom, Marvell, or even Apple are speculated, but future success remains uncertain amid rumors of possible acquisitions by Microsoft, Meta, or Nvidia.

OpenAI Japan announces next-gen model GPT Next

  • At the KDDI SUMMIT 2024, Tadao Nagasaki, CEO of OpenAI Japan, introduced OpenAI’s upcoming model, "GPT Next," which promises to be 100 times more advanced than GPT-4, reflecting the exponential growth in AI capabilities.

  • This next-generation model is expected to significantly advance AI technology and its integration into everyday applications.

Safety and Ethics

OpenAI co-founder Sutskever’s new AI safety startup SSI raises $1 billion. Link.

  • Safe Superintelligence (SSI), co-founded by former OpenAI chief scientist Ilya Sutskever, has raised $1 billion to develop safe AI systems that surpass human capabilities.

  • Valued at $5 billion, SSI plans to use the funds to acquire computing power and hire top talent, focusing on foundational AI research and safety.

  • Major investors include Andreessen Horowitz, Sequoia Capital, and DST Global.

  • Sutskever, alongside co-founders Daniel Gross and Daniel Levy, aims to build a trusted team to advance AI safety amid rising concerns about AI’s potential risks to humanity

AI models face data shortages as websites restrict web crawlers. Link.

  • A report from the Data Provenance Initiative reveals that websites increasingly restrict web crawlers using robots.txt files, threatening the availability of high-quality data for AI training.

  • From 2023 to 2024, about 5% of data in popular AI training datasets like C4 has been revoked, with top domains like news and academic sites restricting access.

  • This trend raises concerns about AI models relying on outdated or lower-quality data from personal blogs and e-commerce sites.

  • The ongoing shift may drive AI companies to invest in synthetic data or exclusive access to valuable data sources, but it could also provoke legal and ethical challenges.

Startups and Funding

You.com raises $50M to expand AI-powered productivity engines for enterprises. Link.

  • You.com, an AI-powered search and productivity platform, has raised $50 million in its Series B funding round led by Georgian, with participation from Nvidia and Salesforce Ventures, bringing its total funding to $99 million.

  • Founded by Richard Socher in 2021, You.com aims to revolutionize knowledge work through "productivity engines" that enhance research, problem-solving, and content creation.

  • The platform has reported over 1 billion queries and 500% revenue growth since January.

  • Despite competition from tech giants like Meta and Google, You.com’s focus on accuracy, customization, and a consumption-based pricing model positions it as a key player in the evolving enterprise AI market.

Sakana AI raises $100M in Series A funding with NVIDIA collaboration. Link.

  • Sakana AI, an AI R&D company based in Tokyo, has raised over $100 million in its Series A funding round led by New Enterprise Associates, Khosla Ventures, Lux Capital, and NVIDIA.

  • The company focuses on developing nature-inspired AI foundation models and aims to advance Japan’s AI capabilities.

  • Sakana AI’s collaboration with NVIDIA will enhance research, data center access, and AI community development in Japan, addressing the country’s unique needs.

  • With this funding, Sakana AI plans to build a world-class AI lab, invest in talent, and develop sustainable AI technologies to help Japan regain a technological edge and tackle future challenges.

Consumer and Enterprise AI Applications

Early reviews of OpenAI’s SearchGPT highlight potential but also flaws. Link.

  • OpenAI’s new search engine, SearchGPT, aims to rival Google Search by offering AI-driven, concise responses rather than traditional search result lists.

  • Early reviews are mixed; while users appreciate its effectiveness for tasks like event planning or coding help, they also note significant issues such as inaccurate, “hallucinated” information.

  • SearchGPT compiles data from Bing and other sources but struggles with tasks where Google excels, such as local searches and online shopping.

  • Though still a prototype, SearchGPT shows promise, but it faces challenges before it can truly compete with Google’s established dominance in the search market.

Tesla introduces 'Actually Smart Summon' with enhanced AI capabilities. Link.

  • Tesla has released FSD v12.5.3 with the new "Actually Smart Summon" feature, which is available first on vehicles equipped with hardware 4 (HW4).

  • The updated Summon uses end-to-end AI and provides a live camera view through the app, enhancing control and safety compared to previous versions that relied on ultrasonic sensors.

  • The update allows users to direct their vehicles via "Come to Me" or "Go to Target" commands with live video feedback, and the feature is expected to roll out to HW3-equipped cars soon.

  • Tesla plans future updates, including improved range and integration with HomeLink for garage door operation.

Nvidia denies receiving subpoena in U.S. DOJ AI antitrust investigation. Link.

  • Nvidia has denied receiving a formal subpoena from the U.S. Department of Justice as part of its ongoing antitrust investigation into the AI sector.

  • While the DOJ has issued subpoenas to other companies like AMD and Intel, Nvidia clarified that it only received a civil investigative demand (CID), a less formal request for information to evaluate potential legal action.

  • The investigation focuses on Nvidia's market dominance, including allegations of anti-competitive behavior such as pressuring customers to use its AI processors exclusively and bundling its GPUs with networking equipment to limit purchasing flexibility.

  • Nvidia disputes all allegations related to its market practices.

Market Map of the Week

Publically disclosed revenue from AI companies & startups. Link.

Image credit: Sapphire Ventures

Thank you for reading the AI Rundown by Lightscape Partners. Please send any questions, comments, or suggestions to [email protected].