• AI KATANA
  • Posts
  • Leaked Documents Show Nvidia Scraping ‘A Human Lifetime’ of Videos Per Day to Train AI

Leaked Documents Show Nvidia Scraping ‘A Human Lifetime’ of Videos Per Day to Train AI

Also: Encouraging Impactful Uses of AI Through Llama Impact Grants Program

Author

AI KATANA
August 06, 2024

Good morning,

Today’s newsletter brings intriguing revelations and significant advancements in the tech world. Leaked documents show Nvidia scraping extensive video content daily to train its new AI model, Cosmos, raising legal and ethical questions. Sonova introduces two groundbreaking hearing aid platforms with real-time AI, improving speech clarity in noisy environments. Groq, an AI chip start-up, secures substantial funding to challenge Nvidia’s market dominance. A new AI by Caristo Diagnostics promises to detect hidden heart attack risks, potentially revolutionizing cardiac care. Lastly, Meta’s Llama 3.1 Impact Grants aim to foster impactful AI applications globally, with substantial funding opportunities. Stay tuned for detailed insights and updates.

Sliced:

  • 🎥 Leaked Documents Show Nvidia Scraping ‘A Human Lifetime’ of Videos Per Day to Train AI

  • 👂🏻 Sonova announces two new hearing aid platforms, including the first hearing aid with real-time AI to address most pressing need in hearing loss

  • 💰 AI chip start-up Groq’s value rises to $2.8bn as it takes on Nvidia

  • ❤️ 'Game changer' AI detects hidden heart attack risk

  • 💵 Encouraging Impactful Uses of AI Through Llama Impact Grants Program

Leaked documents reveal that Nvidia has been scraping vast amounts of video content from platforms like YouTube and Netflix to develop a new video foundational model, codenamed Cosmos. This model aims to advance Nvidia’s AI applications, including Omniverse, self-driving systems, and digital human technologies. Internal communications indicate Nvidia’s extensive use of open-source tools and virtual machines to evade platform restrictions, downloading up to 80 years of video daily. Despite internal concerns about legal and ethical implications, project leaders assured employees that high-level company decisions justified their actions. Nvidia maintains that their practices comply with copyright laws, though the legality of using such data for AI training remains contentious.

Sonova has unveiled two innovative hearing aid platforms, Phonak Audéo Sphere InTnio and Phonak Audéo InTnio, marking significant advancements in the field. The Audéo Sphere InTnio is the first hearing aid to incorporate real-time AI through its proprietary DEEPSONIC chip, which significantly enhances speech clarity in noisy environments by instantly separating speech from background noise. This technology has demonstrated a more than twofold improvement in speech understanding compared to existing products. Additionally, the Audéo InTnio platform, powered by the Phonak ERA chip, offers enhanced sound quality, power management, and Bluetooth connectivity, reducing listening effort and fatigue for users. These developments address the critical ‘speech-in-noise’ challenge, aiming to improve the quality of life for the growing number of people with hearing loss globally. Both platforms will be available starting in August 2024.

Groq, a Silicon Valley-based AI chip start-up, has secured $640 million in funding from investors such as BlackRock, Cisco, and Samsung Catalyst Fund, elevating its valuation to $2.8 billion. Groq aims to challenge Nvidia’s dominance in the AI chip market with its language processing units (LPUs) designed specifically for AI inference, which the company claims are faster and more energy-efficient than those of its competitors. The funding will boost Groq’s computational resources, enabling the roll-out of over 108,000 LPUs by March 2025. Yann LeCun, Meta’s chief AI scientist, joins Groq as a technical adviser, underscoring the company’s ambition to capture a significant share of the AI inference market. Despite not yet generating significant revenue, Groq has formed strategic partnerships with entities like Meta and Samsung, and nations like Saudi Arabia, indicating a robust strategy to expand its AI infrastructure globally.

A new AI model developed by Caristo Diagnostics, an Oxford University spinout, has been hailed as transformative in detecting heart attack risks that conventional CT scans miss. This AI technology identifies inflammation in the heart, a precursor to potential heart attacks within the next decade, by analyzing fat around the arteries. The pilot project, supported by NHS England and running in five UK hospital trusts, has shown promising results. In the Orfan study involving 40,000 patients, the AI helped identify individuals at significantly higher risk of cardiac events, leading to proactive treatments such as lifestyle changes and medication. The National Institute for Health and Care Excellence is evaluating the technology for potential nationwide NHS use, with approvals already secured in Europe and Australia.

Meta has launched the Llama 3.1 Impact Grants program, inviting applications for projects that leverage its latest open source AI model for societal benefit. Building on last year’s success, the program seeks innovative proposals that utilize Llama 3.1’s enhanced features, including support for multiple new languages, to address global challenges in areas such as education, the environment, and open innovation. With a total of $2 million in grants available, organizations worldwide can apply for up to $500,000 through an open call or specialized regional events. These events will also provide technical guidance and mentorship, with awards up to $100,000 for region-specific projects. The program aims to harness the transformative potential of open source AI to improve productivity, creativity, and quality of life globally.

🧑🏽‍💻 AI Jobs

🛠️ AI tools updates

OpenAI has announced a new AI detection tool that claims to identify AI-generated text with 99.9% accuracy, potentially curbing the use of ChatGPT for academic cheating. Despite these claims, skepticism remains about the tool’s effectiveness. Critics argue that the tool’s reliance on watermarking and classifiers might not be robust enough against simple manipulations such as paraphrasing or translating text. Furthermore, the tool cannot detect text generated by other AI models like Google Gemini. While OpenAI’s initiative could deter some students from cheating, its limitations suggest that determined individuals may still find ways to bypass detection. The discussion reflects broader concerns about the evolving landscape of AI and academic integrity.

💵 Venture Capital updates

Google’s recent acquisition of Character.AI, a personalized chatbot developer, is primarily driven by fundraising fatigue experienced by its founders. Character.AI, which previously raised $150 million at a $1 billion valuation, will now see its co-founders and over two dozen researchers rejoining Google. The deal includes a non-exclusive license for Character.AI’s models and a buyout of venture investors at around $2.5 billion. The company will shift from its closed-loop model to utilizing open-source models for post-training, focusing on reduced compute costs. This move positions Google more competitively in the AI race, despite potential scrutiny from FTC antitrust regulators. The transition will challenge Character.AI’s remaining leadership to retain its workforce and adapt to new objectives under Google.

Despite a challenging venture capital market, 38 startups have achieved unicorn status in 2024. These include Elon Musk’s xAI, valued at $24 billion, and a mix of AI, cybersecurity, health tech, and fintech companies. Notable unicorns are Aven, a consumer credit card company; Flo Health, a fertility-tracking app; and Altana Technologies, a global supply chain management company. Other significant names are Chainguard in cybersecurity, legal AI platform Harvey, and defense tech firm Saronic Technologies. This list reflects a diverse range of innovative companies continuing to attract substantial investment and achieving high valuations.

🫡 Meme of the day

⭐️ Generative AI image of the day