May 05 AI Daily — Anthropic & OpenAI Enter Enterprise AI, DeepSeek Models

1. Anthropic and OpenAI Secure Massive Investments, Accelerating Enterprise AI Market Entry

AI leaders are moving beyond simply providing APIs and are building new business models that offer direct solutions to the enterprise market. Anthropic has established a joint venture with Wall Street financial firms, including Blackstone and Goldman Sachs, worth $1.5 billion to support the adoption of AI in enterprises. Similarly, OpenAI has raised $4 billion from TPG, SoftBank, and others to set up a similar 'deployment company,' directly addressing the 'last mile' problem of integrating AI into real business operations.

• Anthropic Joint Venture: A $1.5 billion initiative targeting mid-sized companies, providing 'Claude' model integration and consulting with forward-deployed engineers (FDEs).
• OpenAI Deployment Company: Valued at $10 billion and targeting a $4 billion raise, focusing on private equity portfolio companies.

Sources: aitimes.com  |  🔗 View Original

---

2. The Rise of Open Source LLMs: DeepSeek V4 and Mistral Medium 3.5 Released

The open-source AI ecosystem is making remarkable progress, quickly catching up to Big Tech's proprietary models. DeepSeek has unveiled 'DeepSeek V4', a Mixture of Experts (MoE) model capable of processing a massive 1-million-token context window. At the same time, Mistral AI launched its next-generation 'Mistral Medium 3.5' model alongside its coding agent platform 'Vibe', focusing on improving developer productivity. This signifies that open-source models have officially entered the competition in terms of performance and capabilities.

• DeepSeek V4: Adopts an MoE architecture with 1.6 trillion parameters, highlighting its 1-million-token processing capability and low operational costs.
• Mistral Medium 3.5: Automates complex development tasks such as writing code, refactoring, and testing through the cloud-based remote agent 'Vibe'.

Sources: Simon Willison's Weblog · aitimes.com  |  🔗 View Original

---

3. US White House Considers National Review Before AI Model Releases

The US White House is reportedly considering introducing formal government review and oversight procedures before new AI models are publicly released. This marks a significant policy shift away from previous deregulation trends, reportedly triggered by potential risks identified in Anthropic's 'Mythos' model. While the measure aims to strengthen the safety and accountability of AI technology, concerns have been raised that it could slow the pace of technological innovation.

Sources: The New York Times · aitimes.com  |  🔗 View Original

---

4. Intensifying AI Infrastructure War: Chip IPOs, Offshore Data Centers, and Memory Shortages

The competition in AI technology is expanding into a war to secure hardware infrastructure. AI chip startup Cerebras is attempting an IPO with a $26.6 billion valuation, while Meta has issued $25 billion in bonds to build AI infrastructure. Meanwhile, innovative efforts continue, such as Pantalassa—backed by Peter Thiel—building offshore data centers powered by wave energy. Simultaneously, the shortage of memory semiconductors due to surging AI computational demands is worsening.

• Cerebras: Emerging as a challenger to NVIDIA, pursuing an IPO of up to $3.5 billion at a $26.6 billion valuation.
• Meta: Raising its annual capital expenditure to a maximum of $145 billion, expanding AI infrastructure investments.
• Pantalassa: Secured $140 million to build offshore data centers utilizing wave energy.

Sources: aitimes.com · Benzinga · The Rundown AI  |  🔗 View Original

---

5. The Era of AI Agents Begins: OpenAI Unveils 'Symphony' and MS Launches Governance Platform

'Agent' technologies, where AI autonomously performs tasks, are advancing rapidly. OpenAI has released 'Symphony', an open-source framework that utilizes task management tools as command centers for AI agents. Meanwhile, Microsoft launched 'Agent 365', a governance platform designed to prevent the unchecked spread of AI agents within enterprises and strengthen security, demonstrating that the adoption and management of AI agents has become a core challenge for businesses.

Sources: The Decoder · aitimes.com  |  🔗 View Original

---

6. Meta Unveils 'Autodata', Allowing AI to Generate and Improve Its Own Training Data

Meta has released the 'Autodata' framework to address the quality issues of training data, which dictate AI model performance. This system is designed so that AI can autonomously generate high-quality training data through iterative analysis and improvement, acting like a data scientist. This is seen as a significant breakthrough in solving AI development bottlenecks and advancing the 'self-improvement' capabilities where AI enhances its own learning environment.

Sources: aitimes.com  |  🔗 View Original

---

7. iPhone Siri Can Be Replaced with a Free, Local AI Model Operating Offline

A method to connect a locally running free AI model to the iPhone's Action Button instead of Siri has been revealed. Using the 'Locally AI' app alongside open-source models like Google's Gemma, users can utilize various AI features—such as translation, summarization, and problem-solving—without an internet connection, preserving their privacy. This exemplifies the rapid advancement in the utility and personalization of AI on edge devices.

Sources: The Rundown AI  |  🔗 View Original