• AI Rundown
  • Posts
  • The AI Rundown by Lightscape Partners - 1/13/25

The AI Rundown by Lightscape Partners - 1/13/25

Microsoft’s Open-Source Model, Anthropic’s $60B Valuation Target, and China’s AI Advancements

Author

Ted Wagner
January 13, 2025

Image credit: TIME

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

  • Microsoft has made waves by open-sourcing its Phi-4 model on Hugging Face with full weights, allowing commercial use under an MIT license. This 14-billion-parameter model is optimized for efficient performance in compute-constrained environments and outperforms larger competitors like Google’s Gemini Pro in reasoning and code generation, signaling a shift toward accessible AI innovation.

  • Anthropic is reportedly raising $2 billion in a new funding round led by Lightspeed Venture Partners, boosting its valuation from $18 billion to $60 billion. This development follows Amazon’s $4 billion investment in Anthropic last year, underscoring growing Big Tech interest in generative AI and intensifying competition among OpenAI’s key rivals.

  • China’s AI advancements are testing the effectiveness of U.S. chip restrictions. Despite export controls, Chinese tech giants like Tencent and DeepSeek have developed models rivaling leading U.S. systems, leveraging stockpiled GPUs, offshore providers, and software optimization. This progress raises questions about the efficacy of current U.S. policies and their impact on global AI competition.

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

NVIDIA's AI Foundation Models bring generative AI to PCs. Link.

  • NVIDIA launched foundation models for RTX AI PCs using GeForce RTX 50 Series GPUs, enabling generative AI workflows locally with advanced features like FP4 compute for 2x AI inference performance.

  • The NIM microservices and AI Blueprints offer preconfigured tools for digital humans, content creation, and agentic AI, catering to both developers and enthusiasts.

  • PC manufacturers are rolling out NIM-ready RTX AI PCs to meet the growing demand for high-performance AI on personal hardware.

  • Project R2X, an AI-driven PC avatar previewed by NVIDIA, showcases the integration of vision-enabled AI agents for enhanced productivity and interactivity.

Netherlands engages Nvidia and AMD in AI supercomputer talks. Link.

  • The Dutch government is in discussions with Nvidia and AMD to supply hardware and expertise for a proposed AI supercomputer facility.

  • The initiative aligns with EU projects to boost Europe's digital economy and includes €204.5 million in national AI investments, supplemented by European subsidies.

  • Economy Minister Dirk Beljaarts emphasized the competitive landscape, noting the talks mark significant progress toward realizing the facility.

  • The project aims to enhance Europe's AI research and development capabilities amid global competition.

Models

Microsoft releases Phi-4 model as open source with full weights on Hugging Face. Link.

  • Microsoft's Phi-4, a 14-billion-parameter model, is now fully open-source with an MIT License, enabling commercial use and adaptation.

  • Phi-4 outperforms larger models like Google's Gemini Pro in reasoning and code generation, requiring fewer resources for deployment.

  • The model's streamlined architecture targets efficient performance in compute-constrained environments, emphasizing safety and alignment.

  • This release aligns with the industry's trend towards open-sourcing foundational models, fostering accessibility and innovation.

Musk highlights AI’s shift to synthetic data for training. Link.

  • Elon Musk claims AI models have exhausted the sum of human knowledge for training, pushing companies toward self-learning synthetic data.

  • While firms like Meta and Google already use synthetic data, experts warn it could lead to ‘model collapse’ with biased and lower-quality outputs.

  • Musk emphasized challenges with hallucinations in synthetic data, raising concerns about accuracy and reliability.

  • The growing use of synthetic data sparks debates over quality control and legal disputes over AI’s reliance on copyrighted materials.

Product Launches

Avataar’s Velocity automates product video creation for brands. Link.

  • Avataar has launched Velocity, an AI-powered tool that creates videos directly from product links, targeting a market dominated by Amazon and Google.

  • The tool is already used by HP, Lowe’s, and Victoria’s Secret, focusing on automating video creation for non-flagship products to improve engagement and conversions.

  • Avataar’s advantage lies in its extensive product data and focus on storytelling, ensuring videos highlight unique product attributes without inaccuracies.

  • With features like API integration and brand safety checks, Velocity aims to revolutionize video creation in e-commerce at scale.

Funding & Valuation

Anthropic seeks $2 billion in new funding, targeting a $60 billion valuation. Link.

  • Anthropic, a key OpenAI competitor, is reportedly raising $2 billion in a round led by Lightspeed Venture Partners, boosting its valuation from $18 billion to $60 billion.

  • This follows Amazon's $4 billion funding last year, which will convert to equity during this round, signaling intensified Big Tech interest in AI.

  • Anthropic, co-founded by former OpenAI executives, has raised $6 billion in total and reached annualized revenues of $875 million.

  • AI startups accounted for nearly half of U.S. venture capital funding in 2024, highlighting the sector’s rapid growth.

Hippocratic AI raises $141M to tackle healthcare staffing shortages. Link.

  • Hippocratic AI secured $141M in Series B funding, led by Kleiner Perkins, pushing its valuation to $1.64 billion and solidifying its "unicorn" status.

  • The startup's AI agent app store enables healthcare organizations to deploy AI agents for patient-facing, non-diagnostic tasks, addressing global staffing shortages.

  • In just 18 months, the company achieved milestones like safety parity with clinicians, contracts with 23 healthcare organizations, and 200,000 patient interactions with an average satisfaction rating of 8.7.

  • Funds will drive expansion into new verticals like pharma and global markets, building on its success in AI-powered healthcare staffing solutions.

Consumer Applications

VLC introduces offline AI-powered subtitling and translation. Link.

  • VideoLAN showcased a CES demo of VLC’s real-time AI feature for generating and translating subtitles in over 100 languages, running entirely offline on local devices.

  • The feature uses open-source AI models, integrated directly into the app, without relying on cloud services.

  • This marks a shift from previous plug-ins like Whisper to a native solution built into VLC.

  • VideoLAN also celebrated 6 billion downloads, noting continued user growth despite the dominance of streaming platforms.

Google’s AI-driven Daily Listen transforms Discover feeds into personalized podcasts. Link.

  • Google is rolling out a "Daily Listen" feature that generates short, personalized podcasts based on users' Discover feed preferences.

  • The feature, available via Search Labs for Android and iOS users in the U.S., offers five-minute audio updates with related story links.

  • Rooted in Discover, it adapts to user-followed topics, resembling NotebookLM’s AI-generated Audio Overviews.

  • Daily Listen provides a transcript and clickable links, enhancing the audio experience with deeper topic insights.

China's rapid AI progress tests the limits of U.S. chip restrictions. Link.

  • China’s AI development has advanced significantly despite U.S. export controls, with models from Tencent and DeepSeek rivaling top American systems.

  • China bypasses chip restrictions by stockpiling GPUs, using offshore providers, and optimizing software to maximize hardware efficiency.

  • Experts argue U.S. chip restrictions may need rethinking, as they focus on hardware but fail to address risks tied to military AI development.

  • While China’s open-source models match U.S. capabilities, closed systems in the U.S. still maintain a slight edge, highlighting ongoing challenges in global AI competition.

OpenAI

OpenAI's new model achieves AGI benchmark with ARC-AGI challenge success. Link.

  • Sam Altman, CEO of OpenAI, revealed in a Bloomberg interview that their latest model scored 87.5% on the ARC-AGI challenge, surpassing the 85% threshold for demonstrating reasoning beyond training data.

  • Altman highlighted challenges in defining AGI, emphasizing progress through OpenAI’s tiered benchmarks and the ARC-AGI challenge as a "North Star" toward AGI.

  • He criticized U.S. infrastructure hurdles for AI development and reflected on his tumultuous 2023 ouster and reinstatement as OpenAI’s leader.

  • Former board member Helen Toner cited Altman’s transparency issues as reasons for his brief dismissal.

OpenAI struggles with ChatGPT Pro's profitability. Link.

  • OpenAI CEO Sam Altman revealed the $200/month ChatGPT Pro plan is losing money due to higher-than-expected usage.

  • ChatGPT Pro offers advanced tools like the o1 pro reasoning model but faces criticism for unclear value at its steep price.

  • OpenAI expects a $5 billion loss on $3.7 billion revenue for 2024, citing high operational costs, including $700,000 daily for ChatGPT.

  • The company is exploring pricing changes and usage-based models to boost revenue, aiming for $11.6 billion in 2025.

Sam Altman denies abuse allegations made by sister. Link.

  • Ann Altman, sister of OpenAI CEO Sam Altman, has filed a lawsuit alleging years of sexual abuse beginning in childhood.

  • Sam Altman and his family released a joint statement denying the claims, calling them 'utterly untrue' and attributing them to her mental health challenges.

  • The lawsuit seeks damages exceeding $75,000, alleging mental health impacts from the alleged abuse.

  • Altman, a prominent figure in tech and OpenAI's co-founder faced previous public allegations from his sister but had not addressed them publicly until now.

Market Map of the Week

Portfolio Job Opportunities

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