The AI Power Struggle

Welcome to another edition of the Neural Net…

Where AI meets reality, the future updates itself, and we’re all just trying to keep up.

In this issue:

• The AI Power Struggle

• Harrison Ford Doesn’t Need AI To Destroy His Legacy

• OpenAI Researcher Confirms: ChatGPT Can’t Steal Your Charm

• AI Spots Brain Lesions That Doctors Missed

• Open-Book Doesn’t Mean Open-ChatGPT

• Anthropic Secures Additional Billions

Powering AI: The Energy Race Behind the Tech Revolution

Who is Fueling the AI Revolution?

AI is getting smarter - and consuming massive amounts of energy in the process. Every chatbot response, every self-driving car, every deep-learning breakthrough is guzzling more power than ever before.

💡 Did you know? A single ChatGPT response uses 10 times more energy than a Google search, and some AI systems already consume 33 times more power than traditional software.

All this demand is driving up emissions - Microsoft’s carbon footprint has jumped nearly 30% since 2020, and Google’s emissions in 2023 were almost 50% higher than in 2019, thanks to energy-hungry data centers.

The big question is: what will fuel AI’s future?

Natural Gas

Toby Rice, CEO of EQT, one of America’s largest natural gas producers, is making bold claims that your ChatGPT habit might be powered by good old natural gas for the foreseeable future. In a Friday CNBC interview, he tooted his own horn and declared natural gas will "take the lion's share" of the energy needed to feed the AI obsession. The large amounts of emissions natural gas produces is largely considered to be its greatest downside, but as Mr. Rice pointed out, at least it’s better than coal.

Nuclear 

Beyond being a favorite word of George W Bush, nuclear power is carbon-free and ultra-reliable, but it’s expensive and slow to build. Despite public skepticism, tech giants are investing big in nuclear energy. Microsoft is partnering with Constellation Energy to revive the infamous Three Mile Island nuclear plant in Pennsylvania. The exclusive contract locks in 20 years of power for its AI data centers in the region and aligns with Microsoft’s goal to be carbon-negative by 2030. In the spirit of W, don’t misunderestimate nucular power.

Renewables

Renewables (wind, solar) are lauded for being clean and sustainable but have clear reliability problems. AI data centers need constant, uninterrupted power to function, but solar and wind are unpredictable. The sun sets, the wind dies down, and AI can’t afford downtime. For this reason, many companies have opted for other means of power generation to power their data centers.

In reality, no single source will be enough to sustain the demand by AI on the electricity grid. For now, natural gas and nuclear are leading the power surge, with renewables still needing work.

AI models like ChatGPT are always on, consuming power 24/7, so the energy demand will only continue to rise. With an already strained power grid, Toby Rice sees the AI power crunch as the potential straw that breaks the camels back. Smarter AI use - like deploying smaller, more efficient models for specific tasks - could keep innovation moving without breaking the grid.

Heard in the Server Room
Quick hits roundup of AI news

• Harrison Ford Not Worried About AI, Able To Sell Out On His Own

  Harrison Ford is unfazed by AI's potential posthumous appropriation of his likeness, quipping, "You don't need artificial intelligence to steal my soul" while acknowledging he's already "selling that piece by piece" through his ongoing work (if you don’t believe him, watch the Indiana Jones and Star Wars reboots). The action star remains focused on the next generation of talent rather than digital replicants, confidently declaring "There won't be any need for me. There's somebody behind me. Doing what I did."

• OpenAI Researcher Says Soft Skills Aren’t Going Anywhere

  AI may be coming for jobs, but according to OpenAI researcher Karina Nguyen, there’s one thing it still can’t do - be human. In a recent interview, Nguyen emphasized that soft skills like creativity, emotional intelligence, and leadership will remain valuable, even as AI takes over technical tasks. She pointed out that AI struggles with areas requiring human intuition, decision-making, and complex interpersonal interactions, making roles centered on these skills more future-proof. Because we’re all masters at complex interpersonal interactions and decision-making right?

• Epilepsy AI Tool Detects Brain Lesions Doctors Miss

  Scientists have created an AI tool that can spot epilepsy-causing brain abnormalities that typically fly under radiologists' radar, potentially revolutionizing treatment seizure sufferers. While the algorithm detects about two-thirds of these sneaky lesions, described as finding "one character on five pages of solid black text”, experts caution we'll need more clinical evidence before this digital detective officially clocks in at hospitals. The researchers aren't keeping this breakthrough behind a paywall either, making their code available for hospitals worldwide to test-drive while awaiting the regulatory green light.

ChatGPT Can Do Your Homework… But It Can Also Get You Expelled

Haishan Yang, a former Ph.D. student at the University of Minnesota, found himself expelled and his student visa revoked after being accused of using AI on an exam. Yang, who already has a Ph.D. in economics, took a remote, open-book exam that allowed students to use class materials, but strictly banned AI tools like ChatGPT. After reviewing his exam, professors flagged his answers as too ChatGPT-like, and the university expelled him.

Now, he’s suing, claiming the school altered evidence and relied on unreliable AI detection tools.

Adding to the controversy, the university cited a previous incident from a year earlier, where Yang was accused of using AI on a homework assignment. A professor flagged his work after noticing a paragraph ended with a "note to self" that read: “re write it, make it more casual, like a foreign student write but no ai.” It’s not a great look, like leaving a “totally not cheating” note on your test and hoping no one notices.

His case is shining a spotlight on the murky waters of AI detection in education, where accusations can be made, but proving guilt is another story. Without a clear, provable standard, punishing students based on AI “similarities” is challenging and will always invite controversy.