Small And Open Source Still Has A Horse In This Race

Jun 4, 2026 21m

AI Summary Available

Get the full experience! Sign up to access transcripts, personalized summaries, and more features.

Episode Description

Google released Gemma 4 12B, a multimodal model that runs locally on 16GB devices. TSMC's CEO warned chip supply won't meet demand for years. Ramp raised $750M at $44B, and Anthropic says 80%+ of its merged code is now Claude-authored. Google releases Gemma 4 12B, an 11.95B-parameter unified, encoder-free open multimodal model that can run locally on devices with 16GB of VRAM or unified memory (VentureBeat) Public First: 26% of Americans support increased data center construction, the lowest share among 15 large countries, such as Brazil, Japan, the UK, and Canada (FT) Sam Altman and Dario Amodei are among the signatories on a public letter urging improved tracking of synthetic DNA that could be used in AI-developed bioweapons (Wired) TSMC CEO C.C. Wei says the company won't be able to fulfill the demand led by US customers even as more capacity comes online in the US over the next few years (Bloomberg) Corporate spending management platform Ramp raised $750M at a $44B valuation led by Iconiq, Singapore's GIC, and the OTPP, taking its total funding to $3B (Bloomberg) Anthropic details its progress toward recursive self-improvement, and its implications, and says 80%+ of the code merged into its codebase is authored by Claude (Anthropic) Learn more about your ad choices. Visit megaphone.fm/adchoices

AI-Generated Summary

Google Unveils Gemma 412b: A Local, Multimodal AI Model

Google has released Gemma 412b, an advanced 11.95 billion parameter open-source multimodal AI model designed for local device use. Unique for its encoder-free architecture, Gemma 412b processes raw audio and visual data directly into the core language model, significantly reducing latency and memory overhead. It can operate efficiently on devices with just 16GB of VRAM, making it ideal for secure, offline enterprise use, especially on laptops during travel or in environments without internet access. The model features a large context window of 256K tokens, native reasoning capabilities, and native function calling, opening doors for complex applications including autonomous agents. Available on Hugging Face, Kaggle, and Google’s AI Edge gallery, this model represents a shift toward more accessible, privacy-preserving AI at the edge.

AI's Impact on Privacy and Security Concerns

The episode discusses concerns about the proliferation of AI technologies in sensitive domains, such as biosecurity and data infrastructure. Notably, a public letter signed by industry leaders, including Sam Altman, highlights the risks associated with synthetic DNA and AI-enhanced bio-weapons, urging tighter controls. Additionally, public opinion polls reveal that only 26% of Americans support increased construction of data centers, contrasting with higher support in countries like Nigeria and India. This opposition stems from fears over job losses, environmental impact, and security risks. The discussion underscores the tension between advancing AI capabilities and the societal and political challenges they pose.

Semiconductor Supply and the Growth of AI Infrastructure

Taiwan Semiconductor Manufacturing Company (TSMC), a critical supplier for AI and electronics chips, has stated that demand will outstrip supply for years, despite expanding its capacity and planning new U.S. plants. The company emphasizes its commitment to stable prices and significant revenue growth, amid global efforts to increase chip fabrication worldwide. Meanwhile, Ramp, a corporate spending management platform, raised $750 million at a valuation of $44 billion, reflecting the buoyant state of the AI-driven enterprise sector. Ramp is evolving to integrate AI tools for automating financial actions, preparing for a future where AI agents play a significant role in business operations.

The Advancing State of AI and Concerns Over Recursive Self-Improvement

Anthropic reports rapid progress in AI capabilities, with more than 80% of its code merges being authored by its Claude AI, nearing human parity in code quality. Their research suggests that AI systems perform tasks at an accelerating rate, leading to discussions about the possibility of recursive self-improvement—AI systems autonomously designing successors at increasing speeds. The firm emphasizes the need for global coordination to slow or pause development, highlighting the risks of unregulated race dynamics in AI advancement. This segment raises awareness about the ongoing rapid improvements and the potential future scenarios for AI development.