The AI Rundown by Lightscape Partners - 1/6/25

Happy New Year, and welcome back to another edition of The AI Rundown by Lightscape Partners!

• Microsoft-backed d-Matrix unveiled its GPU-free Corsair processor, a breakthrough in generative AI hardware. By eliminating GPUs and leveraging a DIMC architecture, Corsair achieves cost-effective, high-speed AI inference, signaling a significant shift in the AI hardware landscape.

• SemiKong, the first large language model tailored for semiconductor design, promises to cut chip development timelines by 30%. Developed by Aitomatic and the AI Alliance, it addresses expertise gaps and aims to challenge Nvidia’s dominance in the tech sector.

• OpenAI is exploring a structural shift to transform its for-profit arm into a Public Benefit Corporation (PBC). This change would balance profit with mission-driven initiatives, enabling large-scale investment while funding education, healthcare, and science projects through its non-profit arm.

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

Microsoft-backed d-Matrix introduces GPU-free AI processor for generative AI. Link.

• d-Matrix's Corsair processor, designed for AI inference, replaces GPUs and HBM, achieving cost-effective, high-speed performance.

• Corsair processes 60,000 tokens/second for Llama3 8B models and 30,000 tokens/second for Llama3 70B in resource-intensive tasks.

• Built on DIMC architecture, it integrates neural cores, a RISC-V CPU, and offers 2400 TFLOPs peak computing on a PCIe Gen5 card.

• Micron, a key Nvidia partner, collaborates with d-Matrix, signaling industry shifts in generative AI hardware solutions.

Models

SemiKong claims 30% faster chip development with its LLM. Link.

• SemiKong, developed by Aitomatic and the AI Alliance, is the first LLM tailored for semiconductor design, addressing expertise gaps caused by retiring engineers.

• Built on Meta's Llama 3.1, it integrates Domain-Expert Agents (DXAs) to customize workflows using company-specific technical data.

• It promises a 20-30% reduction in chip time-to-market, improved manufacturing accuracy, and faster onboarding for engineers.

• The AI Alliance, including Meta, AMD, and IBM, aims to challenge Nvidia’s dominance in tech with projects like SemiKong.

AI models falter in realistic medical conversations. Link.

• Researchers at Harvard Medical School and Stanford University found AI models excel in standardized medical tests but struggle with real-world medical conversations.

• The study introduced CRAFT-MD, a framework evaluating AI's ability to gather patient details and make diagnoses during natural conversations, revealing significant performance gaps.

• AI models often missed critical patient information and failed to ask essential questions, highlighting the need for improved conversational and reasoning capabilities.

• Recommendations include designing models for open-ended conversations, integrating diverse data sources, and incorporating non-verbal cues to better mimic real-world medical interactions.

Content Creation

Meta’s AI user profiles aim to transform social media. Link.

• Meta plans to introduce AI-generated user profiles that can create and share content, aiming to make platforms more engaging and entertaining.

• The company is also launching tools like text-to-video creation and personalized AI characters for tutorials and advice.

• Critics warn of risks such as political manipulation, low-quality content, and the lack of emotional depth in AI personas.

• Meta insists these investments are essential for long-term growth, despite concerns over potential misuse and harm.

Funding & Valuation

Stepfun secures significant Series B funding for AI advancements. Link.

• Chinese AI model maker Stepfun raised "several hundred million dollars" in a Series B funding round led by Shanghai's state-owned Fortera Capital and supported by Tencent and Qiming Ventures.

• Stepfun, founded in 2023 by former Microsoft Asia Research Institute chief scientist Jiang Daxin, develops multimodal large language models and is testing a trillion-parameter model.

• Funds will accelerate multimodal capabilities, enhance reasoning features, and introduce consumer products.

• The announcement highlights China's increasing investment in generative AI, following Zhipu AI's recent $411 million funding.

AI-driven exploration startup KoBold Metals secures $527M to find critical minerals. Link.

• KoBold Metals, which leverages AI to locate overlooked mineral deposits, raised $527M in a Series C round led by T Rowe Price and Durable Capital Partners, with backing from Bill Gates’ Breakthrough Energy Ventures.

• The company’s AI analyzes geophysical data to target metals like copper, lithium, and cobalt, vital for EVs and renewable energy.

• A recent Zambian discovery could yield $2B in copper by the 2030s, highlighting AI's transformative role in mining.

• KoBold plans to expand its team and expects an IPO within 3-5 years.

Data Centers + Energy

AI's power hunger sparks interest in nuclear, fusion, and hydrogen energy. Link.

• AI's rising computing demands are pushing data centers to explore energy sources like nuclear, renewables, and fusion, with nuclear startups benefiting from streamlined regulations and early tech adoption.

• Fusion energy continues to attract funding for demonstration reactors, though commercial deployment remains years away.

• Hydrogen startups face uncertainty as potential changes to the Inflation Reduction Act threaten subsidies critical for production scalability.

• 2025 may redefine climate tech as investors and policymakers respond to AI-driven power demands and evolving regulations.

AI data centers linked to power quality issues in nearby homes. Link.

• Reports indicate that areas within 50 miles of major AI data centers experience declining power quality, impacting nearly 3.7 million Americans and reducing the lifespan of electrical appliances.

• Data from over a million Whisker Labs Ting sensors suggests that poor harmonics from data centers stress local grids, causing overheating and potential fire risks.

• Utilities like ComEd and Dominion Energy dispute these findings, citing industry standards and localized issues rather than systemic problems.

• Bloomberg highlights that solutions are underway, but further investigation into grid-level impacts is essential to address widespread concerns.

Regulation + Legal

Congressional AI Task Force's Report Overlooks Catastrophic Risks. Link.

• The bipartisan AI Task Force's recent 273-page report offers insights into responsible AI use but neglects critical concerns like catastrophic risks and rogue AI control.

• Experts like Yoshua Bengio and Eric Schmidt warn of advanced AI's potential to destabilize economies or even humanity, citing significant probabilities in expert surveys.

• Public sentiment aligns with addressing AI risks, yet the report lacks concrete measures like federal regulation, whistleblower protections, or safety standards.

• Critics urge Congress to act swiftly, proposing a dedicated regulator, NIST funding, and legal frameworks to mitigate AI's most severe threats.

OpenAI

OpenAI plans structural shift to balance profit, sustainability, and mission. Link.

• OpenAI's Board is exploring transforming its for-profit arm into a Delaware Public Benefit Corporation (PBC) to attract large-scale investment while maintaining a mission-driven approach.

• The change aims to provide the non-profit with significant resources through equity in the PBC, enabling sustainable operations and broader charitable initiatives.

• The for-profit will focus on scaling AGI development, while the non-profit will drive mission-aligned programs in education, healthcare, and science.

• The shift reflects OpenAI’s ambition to shape the emerging AGI economy and ensure its benefits extend globally.

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