🎙️ 為了 AI 竟然重啟核電廠？OpenAI 核心成員集體出走背後的真相｜AI 日報 Podcast

Introduction

Just how massive is AI's appetite? Today, Microsoft made the shocking announcement that it is restarting the Three Mile Island nuclear power plant, which has been shut down for years, solely to provide an uninterrupted power supply for its data centers! Meanwhile, an epic mass exodus is unfolding inside OpenAI, with top talent jumping ship to rival Anthropic or striking out on their own. This group of brilliant minds has even collectively issued a "devastating warning." What kind of threat could force these bitter rivals to join forces in an unprecedented move? I, Mu-yen, will break it all down for you in just a moment!

Today's Top Stories

1. Creators and Music Giants Join Forces! Anthropic Faces a Copyright Legal Storm

Source: TechCrunch (https://techcrunch.com/）
Summary: Anthropic, the AI giant backed by Amazon and Google, is currently mired in legal battles. Several prominent American authors have teamed up to file a class-action lawsuit, accusing Anthropic of using unauthorized datasets containing large volumes of pirated books to train its Claude models. Meanwhile, music publishers like Universal and Sony are also relentlessly pursuing the company over lyric copyright issues, demanding hefty damages.
Most Surprising Point: This time, the authors and music giants aren't just asking for money; they are demanding that Anthropic "destroy" all models trained on the infringing data. This is essentially asking to format Claude's brain!
Taiwan Perspective: Traditional Chinese creators and publishers in Taiwan face similar anxieties regarding the copyright of local linguistic corpora. If a clear "collective copyright licensing mechanism" can be established internationally, Taiwan's cultural content industry will be able to receive fair compensation in the AI era, rather than being treated as free training fuel for tech giants.
Discussion Points:
1. If a court orders an AI model to be "destroyed," how can the memory of specific data be completely erased from a technical standpoint?
2. Where exactly is the line for "fair use"? Does an AI reading human works count as "learning"?
Script Suggestions: "Imagine studying hard for years to write a book, only for your neighbor's kid with a photographic memory to memorize it, start a class to make a fortune, and not give you a single dime. Wouldn't you be furious? That's exactly how authors feel about Anthropic right now! What's even wilder is that the plaintiffs aren't just asking for damages—they want the 'destruction of the model.' Technically, this is a total disaster because it's incredibly difficult to extract specific memories from a fully trained neural network, like pulling out a Jenga block. I think the outcome of this lawsuit will determine whether AI can continue to get a free lunch or if it will have to start paying up."

2. Mass Exodus of Core Talent from OpenAI! Silicon Valley's AI Landscape Set for a Shakeup

Source: Fortune (https://fortune.com/）
Summary: The AI sector has recently experienced a historic talent earthquake. OpenAI co-founder Ilya Sutskever left to start a new company, SSI, while another core founding member, John Schulman, jumped ship directly to rival Anthropic. Even President Greg Brockman announced a leave of absence, sparking external speculation about internal ideological struggles.
Most Surprising Point: Only a tiny handful of OpenAI's founding team, like Sam Altman, remain in core positions. These former comrades-in-arms have now turned into the most formidable competitors in the market.
Taiwan Perspective: Taiwan's tech and semiconductor industries know the pain of "core teams leaving to start their own ventures" all too well. This serves as a reminder to Taiwan's booming AI startups that relying solely on star founders is not enough; building a healthy organizational culture and talent retention mechanism is the key to long-term growth.
Discussion Points:
1. Does the mass exodus of core technical staff mean that OpenAI's development path has drifted from its original safety and non-profit mission?
2. How will the flow of top talent accelerate the shifting balance of power among competitors like Anthropic?
Script Suggestions: "This is more dramatic than a prime-time soap opera! OpenAI's dream team is now scattered all over. What catches my attention most is that the reasons behind these big departures mostly stem from the conflict between 'AI safety' and the 'speed of commercialization.' They originally gathered around a vision to save humanity, but now they are parting ways due to ideological differences. This flow of top minds has instead supercharged Anthropic and the newly founded SSI. It seems the winner of the second half of the AI race won't just be decided by who has more compute, but by who can win the hearts of these geniuses."

3. To Keep AI Alive, Microsoft is Actually Restarting the "Three Mile Island" Nuclear Power Plant!

Source: CNBC (https://www.cnbc.com/）
Summary: To cope with the staggering power consumption of AI data centers, Microsoft has signed a 20-year power purchase agreement with Constellation Energy. Microsoft will help restart the Three Mile Island nuclear power plant—site of the worst commercial nuclear accident in US history—and dedicate all of its future power generation exclusively to Microsoft's AI computing needs.
Most Surprising Point: Who would have thought that a decommissioned nuclear plant carrying a dark historical shadow would be "brought back to life" simply because AI data centers are so incredibly thirsty for power!
Taiwan Perspective: Taiwan is currently facing severe controversies over electricity usage for semiconductors and AI data centers. Microsoft's move to directly "monopolize a nuclear plant" in the US offers a major revelation for Taiwanese industries: future competitiveness in AI is, at its core, a competition over "energy control."
Discussion Points:
1. As tech giants turn to nuclear energy, how will this reshape the global transition path to clean energy?
2. Will this "AI-exclusive power plant" model lead to a crowding-out effect, subsequently driving up electricity bills for everyday citizens?
Script Suggestions: "The endgame of AI really is energy! Microsoft is going straight for nuclear power this time, and at the infamous Three Mile Island no less. This shows that today's AI compute monsters are consuming electricity at a rate that traditional grids can barely handle. I believe this is a warning sign that Taiwan seriously needs to consider. We've always wanted to be an 'AI Island,' but is our grid ready? While Microsoft and Google are buying nuclear and green energy in the US, Taiwan's tech sector might have to find its own way to 'hunt for power' in the future. Otherwise, no matter how powerful your compute is, it's useless without a plug."

4. The Era of Heavy-Armor AI! Caterpillar Dominates with Autonomous Mining Tech

Source: TechCrunch (https://techcrunch.com/）
Summary: Centenarian heavy industry giant Caterpillar has quietly built the world's most successful autonomous fleet. Its more than 600 driverless mining trucks have traveled hundreds of millions of kilometers completely without human drivers, hauling over 10 billion tons of ore, and maintaining a zero-casualty record to date.
Most Surprising Point: While everyone is watching to see when urban Robotaxis will hit the streets, this excavator company has quietly achieved true "autonomous driving commercialization" in the extreme environments of mining sites.
Taiwan Perspective: Although Taiwan does not produce minerals, we are a global hardware powerhouse for industrial PCs and automotive systems. Caterpillar's success proves that "vertical applications" combining AI with heavy industry are the true blue ocean—highly monetizable with high barriers to entry. This is an excellent entry point for Taiwan's industrial control industry.
Discussion Points:
1. Why has autonomous driving achieved massive success in closed mining sites, yet struggled so much on open city streets?
2. How can traditional heavy industries leverage AI for digital transformation to avoid being phased out by pure-play software tech companies?
Script Suggestions: "Stop just looking at self-driving taxis on the road! Caterpillar has proven that the real king of autonomous driving is in the mines. Think about it: in a dust-filled mine with extremely harsh road conditions, multi-ton steel beasts are driving themselves without a driver and achieving zero casualties. How incredible is that! Their CTO put it perfectly: training a model isn't hard; the hard part is fitting the model into a real workflow. I feel that this kind of down-to-earth industrial AI is exactly what Taiwan's hardware supply chain should deeply cultivate and partner with."

5. Top Rivals Join Forces in Unprecedented Move! OpenAI and Anthropic Jointly Issue AI Cybersecurity Warning

Source: Axios (https://www.axios.com/）
Summary: Fierce competitors like OpenAI, Anthropic, Google, and Microsoft have made a rare move by co-signing a joint statement with cybersecurity firms and government agencies, warning that the threat of AI-assisted cyberattacks is imminent. Defenders have only a few months left to upgrade their defenses; otherwise, critical infrastructure like hospitals and water treatment plants could face devastating blows.
Most Surprising Point: These rivals, who usually fight tooth and nail in the market, have actually agreed to an unprecedented move: in the event of a major security incident, they will grant defenders access to their most core and advanced AI models for defense.
Taiwan Perspective: As a focal point of global geopolitics, Taiwan endures a massive volume of external cyberattacks daily. This warning is a wake-up call for Taiwan; our government and critical infrastructure must immediately integrate AI into cybersecurity defenses, because "using old methods to defend cannot stop an AI machine gun."
Discussion Points:
1. When hackers also use AI to find vulnerabilities, will traditional firewalls and antivirus software become completely obsolete?
2. Can the "security cooperation" among tech giants actually be implemented, or will it just remain a PR stunt?
Script Suggestions: "This is as rare as a movie where the heroes and villains suddenly team up to fight aliens! Getting bitter rivals like OpenAI and Anthropic to sit down and sign this shows that the AI cyber threat is truly breathing down our necks. Future hacker attacks won't involve manually writing code; instead, AI will automatically find system vulnerabilities and customize phishing emails. I want to remind everyone that this isn't just a tech industry issue. If water plants or hospitals are compromised, it affects all of our lives. In this AI arms race, our defense speed must evolve faster than the hackers!"

Conclusion

Did today's tech news feel like an absolute information overload? From copyright wars and talent exoduses to Microsoft restarting a nuclear plant for AI, it all proves that the transformations of the AI era have penetrated every corner of our lives. It's no longer just a competition over software, but a showdown of energy and security capabilities. Thank you for listening to today's "Mark's Tech Insights." I'm Mu-yen, and I'll see you next time!