Original Shanghai Zhencheng Network Technology ZCD Zhencheng Digital Technology April 13, 2025 15:09 Shanghai
ZCD Zhencheng Digital Technology
Leading enterprise AI transformation,
Shaping the future of the industry together.
·AI News Daily·
Today’s News at a Glance
😟 Meta AI research hub FAIR rumored to be cooling down, Llama development rights change hands
💰 OpenAI former CTO’s startup aims to raise $2 billion, GPT veterans join
🍎 Apple Siri upgrade secrets exposed: internal friction, chip shortages lead to “difficult birth”
🚀 Ilya Sutskever’s new AI company SSI secures huge funding again, valued at $230 billion
🎮 Pokemon the new favorite? Engineer uses Gemini 2.5 Pro to live stream game testing AI
🤔 Altman’s TED talk: ChatGPT is not AGI, the future will coexist with superintelligence
01
😟 Meta AI research hub FAIR rumored to be cooling down
Recent reports indicate that as Meta shifts its focus to generative AI products, its foundational artificial intelligence research lab FAIR seems to be facing internal challenges. Organizational restructuring merged FAIR with product teams, leading to its marginalization in the subsequent development of the flagship model Llama, while also facing serious talent drain issues.
Core Highlights
- Organizational Restructuring: FAIR was merged into the Generative AI (GenAI) product division led by Ahmad Al-Dahle, described by former employees as a “blow” to the research lab.
- Development Rights Change: The lead development of Llama 3 and subsequent models has shifted to the GenAI team, with FAIR no longer in a core position. The latest release, Llama 4, was entirely led by GenAI.
- Talent Drain: More than half of the original Llama paper authors have left, top researchers are continuously leaving, and the lab is said to be “slowly dying.”
- Leadership Changes: FAIR head Joelle Pineau announced her resignation, with Chief AI Scientist Yann LeCun temporarily taking over. LeCun denies the demise theory, emphasizing that FAIR will refocus on its long-term goal – Advanced Machine Intelligence (AMI).
- Official Response: Meta AI VP Ahmad Al-Dahle responded to controversies regarding Llama 4’s uneven quality and “open-book exam,” stating quality differences stem from application deployment adjustments and firmly denying training on the test set.
Researcher Thoughts
- For Practitioners: This reflects the strategic shift of large tech companies in the AI field from pure basic research to rapid productization, potentially bringing opportunities and challenges in organizational structure adjustments and talent flow.
- For Ordinary People: Meta’s internal adjustments may affect whether future AI products encountered are more application-oriented with rapid iterations or research-oriented with continuous breakthroughs.
Recommended Reading
- Fortune: Meta AI research lab FAIR faces difficulties
02
💰 OpenAI former CTO’s startup aims to raise $2 billion
Thinking Machines Lab, an AI startup founded by former OpenAI Chief Technology Officer Mira Murati, is seeking a seed round of up to $2 billion, targeting a valuation exceeding $10 billion. The company has attracted several former core OpenAI members, including main authors of the GPT series papers.
Core Highlights
- Huge Financing: The planned seed round funding doubled from the previously reported $1 billion to over $2 billion, with a target valuation of at least $10 billion, showing the capital market’s pursuit of top teams.
- All-Star Lineup: The team has absorbed several OpenAI veterans, recently joined by GPT paper first author Alec Radford and former Chief Research Officer Bob McGrew, boasting strong technical capabilities.
- Company Goal: Committed to creating more practical, smarter AI systems closer to human needs, emphasizing the practical application value of AI.
- Market Optimism: Despite not having released a product yet, it has attracted significant investment attention due to the founders’ and team’s background and reputation.
Researcher Thoughts
- For Practitioners: This shows the entrepreneurial boom in the AI field and extreme optimism from capital, especially for teams with top talent and clear vision. This means new job opportunities and potential industry disruptors are emerging.
- For Ordinary People: There may be more AI products from different companies with diverse design philosophies to choose from in the future.
Recommended Reading
- Business Insider: Mira Murati’s AI Startup Funding
03
🍎 Apple Siri upgrade secrets exposed
The highly anticipated major upgrade for Siri encountered setbacks and failed to launch on time. Reports reveal multiple underlying reasons, including internal team hesitation on technical routes, key leadership changes, poor inter-departmental communication, and even a period of facing GPU computing resource shortages.
Core Highlights
- Technical Wavering: The Siri team initially planned a combination of large and small models (codenamed “Mighty Mouse” and “Mini Mouse”). Later, due to leadership decisions shifting towards a single cloud-based large model, it caused controversy, delays, and violated privacy policies.
- Leadership Change: Former AI head John Giannandrea and his deputy had their responsibilities on the Siri project revoked, replaced by software engineering chief Craig Federighi and Vision Pro executive Mike Rockwell.
- Internal Conflicts: Federighi’s AI team and Giannandrea’s Siri team were reported to have poor communication, even deteriorating relations, affecting collaboration.
- Resource Constraints: Early GPU budget applications were limited, forcing the AI team to rent computing power externally (e.g., from Google, Amazon) and even use alternative chips for development. Apple later clarified that budget needs were gradually met.
- Release Plan: Despite the difficulties, Apple still plans to launch the “Apple Intelligence Siri” feature integrated with AI capabilities this fall.
Researcher Thoughts
- For Practitioners/Developers: Reveals that even giants like Apple face challenges in strategic decision-making, organizational coordination, and resource allocation when advancing complex AI projects. This is a typical case of internal innovation in large enterprises.
- For Ordinary People: This explains the reason for Siri’s slow improvement over the years, while also foreshadowing a potentially significant experience enhancement this fall.
Recommended Reading
- New York Times: Apple AI project details
04
🚀 Ilya Sutskever’s new AI company SSI secures huge funding again
Safe Superintelligence (SSI), an artificial intelligence company co-founded by OpenAI co-founder and former chief scientist Ilya Sutskever, reportedly raised an additional $2 billion at a valuation as high as $32 billion (approximately RMB 234 billion).
Core Highlights
- Sky-High Valuation: The new funding brings the valuation of this recently established company, whose products have yet to be released, to an astonishing $32 billion, showing the market’s extremely high expectations for its founding team and grand goals.
- Capital Pursuit: This round of financing was reportedly led by Greenoaks, adding to the $1 billion successfully raised previously.
- Single Goal: SSI publicly declares its sole goal and product is “Safe Superintelligence,” focusing on the ultimate challenge of AI safety.
- Mysterious Product: Currently, the company’s official website information is minimal, only containing a mission statement. Specific products and R&D progress remain confidential.
Researcher Thoughts
- For Practitioners: SSI’s huge financing reflects capital’s long-term confidence and massive bet on the “Artificial General Intelligence” and “AI Safety” track. This means the exploration of AGI/ASI (Artificial General/Super Intelligence) is still accelerating, with safety issues placed at the core.
- For Ordinary People: This represents a technology that could profoundly change the future being secretly developed by a small group of elite teams.
Recommended Reading
- Financial Times: Ilya Sutskever’s SSI Funding
05
🎮 Using Gemini 2.5 Pro to live stream game testing AI
A software engineer created the “Gemini_Plays_Pokemon” channel on the Twitch platform, live streaming the use of Google’s latest Gemini 2.5 Pro model to play the classic game “Pokemon,” as a novel way to test AI capabilities.
Core Highlights
- Novel Testing: The engineer uses playing games—a complex, dynamic task requiring strategy—to test and demonstrate the performance of Gemini 2.5 Pro.
- Real-Time Demonstration: The live stream clearly shows how the AI analyzes game data (location, map, screenshots) and makes decisions (like capturing Pokemon).
- Not the First: Previously, Anthropic also announced that its Claude 3.7 Sonnet model achieved significant progress in “Pokemon” training (obtained three badges).
- Official Attention: Google’s AI product lead also noticed this live stream and shared the link, stating “it’s very fun to watch,” indirectly acknowledging the value of this testing method.
Researcher Thoughts
- For Practitioners: This provides AI researchers and developers with a vivid, intuitive, and close-to-real-world interaction method for AI evaluation.
- For Ordinary People: Watching AI play games is a fun way to intuitively feel the current intelligence level and limitations of AI, making AI testing more entertaining and engaging.
Recommended Reading
- Twitch: Gemini_Plays_Pokemon Live Stream Channel
06
🤔 ChatGPT is not AGI
OpenAI CEO Sam Altman shared his views on AI development at the TED2025 conference. He explicitly stated that the current ChatGPT is not Artificial General Intelligence (AGI) and predicted that human society will coexist with AI far exceeding human intelligence in the long term.
Core Highlights
- AGI Definition: Altman emphasized that there are internal disagreements on the definition of AGI, but clearly stated ChatGPT is not AGI due to its lack of autonomous learning, self-improvement, and the ability to independently complete complex tasks.
- Future Outlook: He foresees a future world where products and services are extremely intelligent and efficient, making it hard for people to imagine an era without intelligent AI assistance.
- Employment Impact: Acknowledged concerns about AI’s impact on employment but leans towards viewing AI as a tool to enhance capabilities, calling for new business models to protect creators’ rights.
- Open Source Plan: Admitted OpenAI started late in open source but confirmed they are developing a powerful open-source model close to the cutting edge.
- Development Focus: OpenAI will focus more on user experience rather than just model parameter competition in the future, and is optimistic about AI’s breakthrough potential in scientific discovery (like disease prevention, new materials) and software engineering.
Researcher Thoughts
Altman’s speech sets rational expectations for the current AI capabilities (not AGI) for both the industry and the public, while outlining the long-term vision for AI development and OpenAI’s strategic direction (user experience, open source, scientific applications). This helps practitioners grasp the industry pulse and gives ordinary people a clearer understanding of AI’s potential and challenges.
Recommended Reading
- X.com: Sam Altman TED2025 conversation clips
Today’s Summary
- AI Financing Boom: Companies founded by former OpenAI executives are highly sought after by capital. Ilya Sutskever’s SSI is reportedly valued at $32 billion, and Mira Murati’s Thinking Machines Lab seeks over $2 billion in seed funding.
- Big Tech Strategic Adjustments: Meta’s basic research department FAIR reportedly faces marginalization and talent drain; Apple, after overcoming internal challenges, plans to launch a new version of Siri integrated with AI capabilities in the fall.
- AGI Definition & Outlook: OpenAI CEO Altman clearly stated in a public speech that current ChatGPT is not AGI and predicted humans will coexist with AI far exceeding their own intelligence.
About Us
🚀 Leading enterprise digital transformation, shaping the future of the industry together. We specialize in creating customized digital systems integrated with AI, achieving intelligent upgrades and deep integration of business processes. Relying on a core team with top tech backgrounds from MIT, Microsoft, etc., we help you build powerful AI-driven infrastructure, enhance efficiency, drive innovation, and achieve industry leadership.