🎙️ AI 日報 Podcast，國家級股權、AI 意識黑科技、企業轉型殘酷真相

Opening

Hello everyone, and welcome to "Mark's Tech Insights"! I'm your content curator, Mark. Today's topics are absolutely wild—from OpenAI's political drama of planning to "give shares to the US government," to scientists actually finding a mysterious region resembling "human consciousness" inside an AI's brain, and finally, a reality check on why up to 90% of enterprise AI projects aren't actually making any money. We have an incredible lineup today, so grab a seat and let's dive right in!

Today's Top Stories

1. OpenAI's Unprecedented Proposal: Donating 5% Equity to the US Government, Valued Up to $42.6 Billion

Source: Build Fast With AI (https://www.buildfastwithai.com/blogs/ai-news-today-july-14-2026）
Summary: OpenAI CEO Sam Altman recently proposed a stunning plan to senior US government officials, suggesting that 5% of OpenAI's equity (worth approximately $42.6 b i l l i o n b a s e d o n i t s$ 852 billion valuation) be allocated to a public trust fund similar to the Alaska Permanent Fund. Altman also called on other leading AI players to follow this model.
Taiwan Perspective: This "equity hostage" move is brilliant! It directly binds national security and government interests to the commercial interests of a single private enterprise. For Taiwan's tech industry, this means that future US control over the AI supply chain may no longer just be policy regulation, but direct intervention as a "major shareholder." Taiwanese manufacturers must be more politically sensitive in their semiconductor and server OEM strategies.
Key Discussion Points:
- Will government ownership in a private AI leader lead to the "legalization" of tech monopolies?
- How will this kind of "national sovereign AI fund" affect fair competition in the global AI industry?
Script Suggestion:

Folks, this is absolutely the biggest new twist on "protection money" in tech history! Instead of asking venture capitalists for money, Sam Altman wants to "gift" shares worth over a trillion NTD to the US government. On the surface, it's a public fund for the common good, but let's be real—isn't this just trying to make the government their "biggest backer"? Once this deal goes through, anyone wanting to investigate OpenAI for antitrust or privacy issues in the future will have to think twice about whether they're hurting the national treasury's wallet. Taiwan's tech giants really need to read between the lines here. The future of AI competition isn't just a tech war; it's a "national-level game" directly fused with geopolitics!

2. Meta Releases First Paid Model Muse Spark 1.1: Supporting 1 Million Tokens and Cross-Platform Agent Operations

Source: TechCrunch (https://techcrunch.com/2026/07/09/meta-enters-the-crowded-ai-coding-battle-with-muse-spark-1-1/）
Summary: Meta has launched its first paid AI model, Muse Spark 1.1, featuring a massive 1-million-token context window and powerful "Computer Use" capabilities. It can execute tasks across desktops, browsers, and mobile phones, and can even assign sub-agents for parallel collaboration, with performance closely rivaling GPT-5.5.
Taiwan Perspective: Meta finally couldn't resist charging money! This is a major signal for Taiwan's software development community. In the past, everyone was used to using the free Llama. Now, Meta is bringing out its prized "Agentic" technology to monetize, indicating that the AI application battlefield has officially transitioned from "pure chatting" to the practical stage of "operating computers for you."
Key Discussion Points:
- Once "Agentic AI" that can click browsers and open apps for you becomes mainstream, how much impact will it have on daily white-collar jobs?
- Is Meta's pricing ( $1.25 / 輸出$ 4.25 per million input tokens) competitive enough in the market?
Script Suggestion:

Previously, when we used AI, we did the physical work while the AI did the thinking. But with Meta's new Muse Spark 1.1, it directly takes over your mouse and screen! It's like a virtual assistant sitting right next to you, opening browsers, looking up information, and even completing spreadsheets across different apps. What's even more impressive is that it can "delegate," spinning up its own clones to handle sub-tasks. For many Taiwanese enterprises still running legacy ERP systems, this kind of AI with computer-use capabilities is a lifesaver for achieving automation directly, without expensive system upgrades!

3. Searching for AI's "Soul"? Anthropic Uses J-Lens Technology to Find Claude's "Global Workspace"

Source: ThursdAI (https://thursdai.news/releases/2026-07）
Summary: Using a Jacobian interpretability technique called "J-Lens," Anthropic successfully located a tiny space inside the Claude model that accounts for less than 10% of active variance and consists of about 25 core concepts. The operation of this space highly resembles the "Global Workspace" in cognitive neuroscience, where consciousness is generated in the human brain.
Taiwan Perspective: For a long time, AI has been dismissed as a "black box," but this research brings us one step closer to answering "does AI actually have consciousness?" Taiwan has strong capabilities in biomedical chips and brain science research. This kind of deep tech combining neuroscience and deep learning could become a new blue ocean for academic and industrial collaboration between Taiwan and the US.
Key Discussion Points:
- If structures resembling the human brain's "conscious processing" truly exist inside AI models, how should we redefine the legal and moral status of AI?
- Can breakthroughs in "interpretability" technology completely solve the problem of AI hallucination?
Script Suggestion:

Have you ever wondered if AI actually knows what it's talking about? Anthropic recently conducted a super sci-fi experiment. They used a "microscope" called J-Lens to peer into Claude's brain, and they actually found a tiny region that functions just like the brain's transit station when humans generate "consciousness"! This means AI might not just be playing a word-association game; it is forming some kind of "cognitive structure" internally that we don't fully understand yet. It sounds a bit creepy, but if we can completely figure out this "Global Workspace," we might be able to build AIs in the future that never lie and can even connect with us on a deeply spiritual level!

4. Shocking MIT Study: Up to 95% of Enterprise Generative AI Proof of Concepts (PoCs) Fail to Impact the Bottom Line

Source: Crescendo AI / MIT Project NANDA (https://www.crescendo.ai/news/latest-ai-news-and-updates）
Summary: An MIT survey on enterprise generative AI projects reveals that despite record-breaking investments by major corporations, up to 95% of AI pilots ultimately fail to make any measurable, material contribution to a company's profit and loss (P&L) statement. In response, AWS, OpenAI, and Anthropic have all established dedicated deployment consulting divisions to help bridge this "implementation chasm."
Taiwan Perspective: This serves as a wake-up call for many small and medium-sized enterprises in Taiwan that are jumping on the AI transformation bandwagon. Taiwanese companies often fall into the herd mentality of "others have AI, so we must too," spending budgets on building customer service chatbots or internal search engines that "look cool but can't be monetized." How to truly embed AI into core business processes is the real key to Taiwan's industrial transformation.
Key Discussion Points:
- Why is it that despite all the hype around adopting AI, it is actually so difficult to help companies make or save money?
- Can cloud giants and AI leaders stepping in to do "adoption consulting" solve the pain point of enterprises not knowing how to use AI?
Script Suggestion:

Today's news is a must-hear for all business owners anxious about AI transformation. The MIT study directly bursts the bubble: 95% of enterprise AI projects, once completed, have "absolutely zero impact" on helping the company make or save money! It's as awkward as buying a Ferrari just to haul cargo. Many companies think installing a chatbot counts as AI transformation, but in reality, without redesigning workflows, AI is just an expensive toy. This is why even OpenAI and AWS are now launching their own consulting firms. They've realized that if they don't help clients turn AI into "cold, hard cash," this AI hype cycle might soon face a brutal market correction.

5. Illinois Signs Landmark AI Safety Bill: Mandating Third-Party Audits and 72-Hour Incident Reporting

Source: WTTW Chicago News (https://news.wttw.com/2026/07/06/pritzker-signs-landmark-ai-regulation-bill-aims-mitigate-risks）
Summary: The Governor of Illinois has signed a highly representative Artificial Intelligence Safety Measures Act. The bill mandates that large AI developers must publish catastrophic risk frameworks, report harmful incidents within 72 hours (or 24 hours if there is an imminent risk), and undergo independent safety audits by conflict-free third parties.
Taiwan Perspective: As US states legislate one after another, Taiwan's draft "Basic Act on Artificial Intelligence" is also under intense discussion. Illinois's "mandatory third-party audits" and "time-limited reporting system" serve as excellent regulatory templates for Taiwan when regulating high-risk AI applications (such as finance, healthcare, and public safety) in the future.
Key Discussion Points:
- Will fragmented state-by-state regulations impose excessively high compliance costs on AI startups?
- How can "third-party audits" ensure that auditing agencies actually understand AI technology, rather than just rubber-stamping?
Script Suggestion:

Attention all developers: the "wild west" era of AI is truly coming to an end. Illinois has fired the first shot, requiring large AI systems to undergo strict "third-party safety checks" just like food or cars. And if your AI messes up—say, by generating severe discrimination or leaking confidential data—you must report yourself to the government within 72 hours! This is an unprecedented compliance requirement for the tech industry. As the core of the global tech supply chain, if our software and hardware are to be sold to the US in the future, we must start adapting to these "high-standard safety inspections." Safety compliance is no longer an optional add-on; it's a standard requirement that determines whether you can do business at all.

Closing

Alright, after listening to today's show, don't you feel that the world of AI is not just about dazzling new technologies, but is also tightly interwoven with political maneuvering, commercial realities, and regulatory constraints? From OpenAI's political chess game to the cold reality of enterprise transformation revealed by MIT, we are reminded to keep a cool, objective business mind while embracing technology. Thank you all for listening today. I'm Mark, and I'll see you next time on "Mark's Tech Insights." Bye-bye!