This edition of The Download dives into the complex world of artificial intelligence and the burgeoning field of advanced nuclear power, offering insights into the latest developments and persistent questions.

The Most Misunderstood Graph in AI: Decoding Exponential Growth

The AI community eagerly awaits updates from METR (Model Evaluation & Threat Research), a nonprofit whose now-iconic graph has significantly shaped the discourse surrounding large language models (LLMs). This graph suggests an exponential rate of development in certain AI capabilities, a trend that recent model releases have demonstrably surpassed. For instance, Anthropic’s Claude Opus 4.5, released in late November, demonstrated the ability to independently complete a task that would typically take a human approximately five hours, far exceeding even the predicted exponential trajectory. However, the reality behind these impressive advancements is far more nuanced than these dramatic figures might suggest. A deeper exploration of this graph and its implications is crucial for understanding the true pace and nature of AI development. This article is part of MIT Technology Review Explains, a series dedicated to untangling complex technological landscapes.

Next-Generation Nuclear Power: Answering Your Burning Questions

Nuclear power remains a central topic in the global energy conversation. Following a recent online Roundtable discussion that brought together experts on next-generation nuclear power, hyperscale AI data centers, and the electricity grid, numerous audience questions emerged. While some of these were addressed during the discussion, many warranted further exploration. This piece aims to answer some of the most pressing questions about advanced nuclear power, delving into the intricacies and potential of this vital energy source. This article is contributed by Casey Crownhart and is part of The Spark, MIT Technology Review’s weekly climate newsletter.

The Must-Reads: A Curated Selection of Tech News

This section provides a curated selection of compelling stories from across the internet, highlighting key developments and emerging trends in technology.

1. Anthropic’s New Coding Tools Disrupt Markets: The emergence of Anthropic’s advanced coding tools is sending ripples across diverse industries, from publishing and law to advertising and software development. Legacy software companies, in particular, face potential disruption. The question of a "software-mageddon" is being debated, with varying perspectives on its likelihood and impact.

2. Apple’s Lockdown Mode Thwarts FBI Access: A reporter’s iPhone was successfully protected from FBI access due to Apple’s Lockdown Mode, a security feature designed to prevent sophisticated cyberattacks. While this feature proved effective against direct phone access, agents were still able to retrieve data from the reporter’s laptop.

3. Data Center Outage Impacted All TikTok Categories: A recent data center outage affected all categories of content on TikTok, not just political material as some users had claimed, according to a report from NPR.

4. Big Tech Invests Billions in India’s AI Sector: India is experiencing a significant influx of investment from major technology companies eager to capitalize on the country’s growing AI capabilities. A new 20-year tax break is expected to further accelerate this trend. This investment comes alongside concerns about the working conditions of female content moderators who are exposed to abusive material to train AI systems. Additionally, Indian officials are considering restrictions on social media use for minors, highlighting the country’s complex approach to AI and digital regulation. MIT Technology Review has previously explored India’s pursuit of AI independence.

   

5. YouTubers Harass Women Using Body Cams: A disturbing trend has emerged where YouTubers are reportedly abusing freedom of information laws by using body camera footage to harass and humiliate women. This raises questions about the intended purpose of police body cameras and how technology, including AI, is being utilized in their application. MIT Technology Review has previously examined the potential of AI to improve police bodycam technology.

6. Jeffrey Epstein’s Inbox Recreated by Pranksters: A group of individuals has reportedly recreated a functional version of Jeffrey Epstein’s Gmail inbox, complete with notable starred threads. This development reignites discussions about Epstein’s extensive ties to the tech industry and the potential for revelations to cause rifts within established factions.

7. New Model Explores Near-Death Experiences: A novel model has been proposed that may offer an explanation for near-death experiences, though not all researchers are in agreement with its conclusions. This raises profound questions about the nature of consciousness and the biological processes associated with death. MIT Technology Review has previously explored the question of "What is death?"

8. New App Mimics TikTok for "Vibe-Coded" Apps: A new application has emerged that functions as a TikTok equivalent for "vibe-coded" mini-apps, catering to a new generation of digital expression. This development highlights the evolving landscape of online interaction and the emergence of new digital subcultures. MIT Technology Review has previously delved into the concept of "vibe coding."

9. Rogue TV Boxes Gain Popularity Amidst Rising Streaming Costs: Frustrated by escalating subscription fees for streaming services, consumers are increasingly turning to unofficial or "rogue" TV boxes to access content, indicating a growing dissatisfaction with the current media consumption model.

10. Climate Change Threatens Winter Olympics: The future of the Winter Olympics is increasingly imperiled by climate change, with artificial snow becoming a critical, albeit short-term, solution. Meanwhile, in the realm of sports, Team USA is leveraging AI to gain a competitive edge.

Quote of the Day

"We’ve heard from many who want nothing to do with AI." This statement from Ajit Varma, head of Mozilla’s Firefox browser, explains the company’s decision to reverse its previous plan to transform Firefox into an "AI browser." Mozilla is now introducing an AI-blocking menu to upcoming Firefox builds.

One More Thing: Personal Data in AI Training Sets

New research has revealed that a significant AI training dataset, DataComp CommonPool, likely contains millions of examples of personally identifiable information, including images of passports, credit cards, birth certificates, and other sensitive documents. While researchers audited only a small fraction of the dataset, their findings suggest that hundreds of millions of images containing personally identifiable information, such as faces and identity documents, may be present. This underscores the pervasive nature of data scraping and the potential for personal information shared online to be incorporated into AI training sets. This article is part of MIT Technology Review’s ongoing coverage of AI and data privacy.

We Can Still Have Nice Things

This section offers a curated selection of comforting, fun, and distracting content to brighten your day.

• Tips for beating boredom on long marathon training runs.
• An exploration of Mark Cohen’s intimate street photography, offering a unique perspective on humanity.
• A dedicated gamer’s meticulous recreation of a Fallout vault within the Sims 4.
• A look at what music’s most stylish men are wearing, from bold leather to understated parkas.