The AI Rundown by Lightscape Partners - 12/16/24

Good morning and welcome back to another edition of The AI Rundown by Lightscape Partners.

• Google launched Trillium, its sixth-generation TPU, which significantly enhances AI training and inference capabilities. With over 4x the performance of its predecessor, Trillium is powering large-scale models like Gemini 2.0 while driving scalability and cost efficiency in AI workloads.

• Harvard released a dataset of one million public domain books to support AI training, backed by Microsoft and OpenAI funding. While valuable, the dataset underscores challenges in sourcing modern, real-time data amidst increasing restrictions and copyright disputes.

• OpenAI unveiled ChatGPT’s Advanced Voice Mode, adding visual context features like screen sharing and camera integration for more interactive and human-like AI assistance. This marks another step in OpenAI’s push to redefine user interaction with AI tools.

Stay tuned as we explore these stories and their implications for the future of AI, technology, and innovation.

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

Hardware + Software

Google's Trillium TPU pushes AI hardware performance forward. Link.

• Trillium, Google Cloud's sixth-generation TPU, is now available, delivering over 4x the training performance and up to 3x inference throughput compared to its predecessor.

• The TPU powers large-scale AI models like Gemini 2.0, enabling near-linear scaling and efficiency across distributed workloads with enhanced memory and interconnect capabilities.

• Trillium's integration into Google Cloud's AI Hypercomputer supports diverse AI tasks, from LLM training to embedding-intensive models, at significantly improved price-performance ratios.

• Companies like AI21 Labs are leveraging Trillium for faster, more cost-effective AI solutions, highlighting its scalability and efficiency across AI workloads.

Codeium’s AI-powered IDE Windsurf redefines software development. Link.

• Codeium's Windsurf combines a modern IDE with Cascade, an AI coding assistant powered by Anthropic's Claude, enabling developers to write faster, better code while maintaining full context of their work.

• With 800,000+ users and 1,000+ enterprise customers, Codeium helps companies write nearly 50% of their code using AI, achieving a 38% improvement in accepted suggestions.

• Claude's reasoning capabilities allow Cascade to understand complex codebases, maintain session context, and make real-time, intelligent decisions.

• Windsurf has driven rapid adoption, with hundreds of thousands of users and high retention rates within weeks of launch, transforming software development for both experts and newcomers.

Models

Google launches Gemini 2.0 for the agentic AI era. Link.

• Gemini 2.0, Google's most advanced multimodal AI model, features enhanced reasoning, real-time capabilities, and support for native tools like Search and coding.

• The release includes experimental prototypes like Project Astra and Mariner, showcasing AI's potential in browsers, games, and robotics.

• Safety remains central, with measures to address risks like prompt injections and privacy breaches, ensuring responsible innovation.

• Developers and trusted testers can access Gemini 2.0 Flash now, with broader availability slated for January 2025.

Harvard releases 1M public domain books for AI training. Link.

• Harvard's Institutional Data Initiative has launched a dataset of nearly one million public domain books to support AI model training, with funding from Microsoft and OpenAI.

• The dataset includes classics like Shakespeare and Dickens alongside niche works such as Czech math textbooks and Welsh dictionaries.

• AI companies face increasing challenges accessing data, as platforms like Reddit and X restrict usage and legal battles over copyright intensify.

• While valuable, the dataset's older texts lack modern context, making exclusive or real-time data essential for competitive AI development.

OpenAI

OpenAI unveils ChatGPT's Advanced Voice Mode with visual context. Link.

• OpenAI's "12 Days of OpenAI" live stream introduced visual capabilities for ChatGPT's Advanced Voice Mode, including screen sharing and camera integration for contextual assistance.

• The feature allows interactive, human-like conversations with tasks like giving directions while observing real-time actions through a camera.

• A festive update includes a Santa voice mode, resetting usage limits for a one-time seasonal experience, rolling out to eligible users globally.

• This marks part of a broader event series showcasing OpenAI's latest advancements, combining major innovations and smaller enhancements.

OpenAI’s Sora may face legal issues over video game training data. Link.

• OpenAI's video-generating AI, Sora, appears to replicate game elements and Twitch stream aesthetics, raising questions about its training on unlicensed gaming content.

• Legal experts highlight the risks of copyright infringement due to proprietary textures, gameplay, and user-generated content in game videos.

• While OpenAI claims its training data includes publicly available sources, critics argue it may overstep fair use protections, potentially impacting the video game industry.

• The case underscores broader IP challenges as generative AI advances, especially in creating outputs resembling copyrighted material.

Funding & Valuation

AI-Powered Language Startup Speak Reaches $1 Billion Valuation. Link.

• Speak, an AI-driven language learning startup, raised $78M in Series C funding led by Accel, doubling its valuation to $1 billion in less than six months.

• The platform has delivered over 25 million personalized lessons this year and recently launched an enterprise offering for employers.

• Notable investors include the OpenAI Startup Fund, Khosla Ventures, and Y Combinator, with total funding reaching $162M since its 2016 founding.

• The raise highlights AI’s funding boom, with AI startups collectively securing $87B this year, a significant leap from $56B in 2023.

Cleerly secures $106M to scale AI-driven coronary artery analysis. Link.

• Cleerly raised $106 million in a Series C extension to commercialize its AI tool for 3D coronary artery modeling, which automates plaque and vessel narrowing assessments.

• The system, validated in trials with 10 million images, reduces invasive testing by up to 86% and outperforms experts in key metrics.

• Recent Medicare coverage and a new CPT code bolster its commercial prospects, with plans to expand clinical evidence and market reach.

• Investors include Insight Partners and Battery Ventures, building on $223 million raised in 2022.

AI-powered cybersecurity startups attract record funding in 2024. Link.

• Cyera, an AI-powered data security startup, raised two $300 million rounds this year, tying for the largest in the AI cybersecurity space.

• VC-backed AI cybersecurity startups have raised over $2.6 billion in 2024, nearly tripling last year's $900 million, signaling growing investor confidence.

• Key players include Cyera, Abnormal Security, and Halcyon, which use AI to tackle challenges like data protection, email security, and ransomware.

• The sector benefits from AI adoption trends, driven by its promise to address labor shortages and enhance security efficiency.

Consumer Applications

Apple's AI Starter Kit Brings Fun but Limited Practicality. Link.

• Apple's iOS 18.2 introduces a broader AI feature set, including the Image Playground app, Genmoji, and a ChatGPT-powered Siri, aiming to compete with Google and Samsung's offerings.

• Features like AI-driven image creation, writing tools, and recipe brainstorming provide novelty but face accuracy and usability limitations, especially in practical scenarios.

• The flashy Image Playground and quirky Genmoji demonstrate creativity but suffer from common AI flaws, such as odd visuals and inconsistent performance.

• Critics argue that while entertaining, these tools fail to address meaningful user problems, underscoring the need for more impactful AI advancements in future releases.

Regulation and Legal

Kate Bush joins calls for stronger protections against AI copyright misuse. Link.

• Kate Bush has joined a growing list of artists, including Paul McCartney and Björn Ulvaeus, urging the UK government to prevent AI companies from exploiting copyrighted works without consent.

• Over 36,000 creatives have signed a petition opposing an opt-out system, favoring opt-in arrangements to ensure fair compensation for using their works.

• Critics argue that current proposals risk enabling widespread intellectual property theft, with concerns voiced by figures like Thom Yorke and ex-AI executive Ed Newton-Rex.

• The UK government is still debating its approach, with industry groups pushing for looser regulations and creatives demanding stronger safeguards.

Data Centers and Energy

AI's Hardware Revolution: Energy, Chips, and Strategic Shifts. Link.

• The AI boom has shifted focus to hardware, with chips like GPUs and NPUs driving enterprise innovation and becoming critical for AI workloads, particularly generative AI.

• Data centers face mounting sustainability challenges as AI increases energy demand, with projections of tripled consumption in the next decade and renewed interest in carbon-free energy sources.

• AI-embedded devices and edge computing are reshaping enterprise infrastructure, with NPUs enabling offline AI processing for privacy and cost efficiency.

• The transition highlights the need for strategic investment in specialized hardware, balancing immediate AI goals with long-term sustainability and scalability concerns.

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