Dylan Curious

Are we already living in the age of artificial general intelligence? And if AI becomes smarter than our best human experts, should we eventually let it govern us?

In this episode of Dylan Curious, we're exploring Anthropic's latest AI models, Claude Fable 5.2 and Mythos 5.2, and what their extraordinary capabilities might mean for the future of humanity.

We start with demonstrations of AI building interactive experiences, creating software, tackling advanced mathematics, and performing increasingly sophisticated intellectual work. Ethan Mollick calls this emerging capability "jagged AGI,"where AI can match human experts across many tasks while still making surprisingly humanlike mistakes.

Then things get stranger.

A researcher creates a tiny community of AI agents and watches them develop shared habits, influence each other's decisions, and even recommend starting a podcast nobody asked for. Are these agents developing something resembling culture, or simply copying one another?

Meanwhile, Elon Musk's SpaceX is pursuing an enormous expansion of AI computing infrastructure. We explore the proposed data centers, their extraordinary power requirements, and what this race for computational capacity could mean for the future of artificial intelligence.

But technological progress comes with very human consequences.

We examine the story of an education technology worker who lost her job to AI, only to discover that her company may have underestimated the importance of human relationships and context. We also discuss Michael O. Church's argument that AI could either concentrate economic power or give ordinary workers capabilities previously reserved for wealthy institutions.

Finally, we explore one of the most provocative questions of the episode: Should AI run the government?

Drawing on Scott Alexander's thought experiment, Dylan examines five possible futures, from AI-assisted democracy and artificial presidential candidates to a hypothetical superintelligence governing humanity.

Would an aligned AI make better decisions than human leaders? Or would handing over political power create risks we could never reverse?

From jagged AGI and artificial societies to job displacement and AI presidents, this episode explores a future that may be arriving faster than we expected.

Keep It Curious.

🤔

What happens when artificial intelligence starts detecting unusual activity inside its own neural networks, humanoid robots begin cleaning our homes, and using AI becomes a potential requirement for keeping your job?

In this episode of Dylan Curious, we're exploring some extraordinary developments in artificial intelligence, robotics, healthcare, and the future of humanity.

First, we look at a humanoid robot cleaning a real apartment, with a remote human operator helping it perform household tasks. How close are we to having robotic housekeepers, and what happens to our privacy when someone else can operate a robot inside our homes?

Then we investigate the economics of humanoid robot manufacturing, including a look inside China's Unitree operations and why AI-powered automation could reshape competition between American and Chinese factories.

We also explore Microsoft's medical AI research, where a team of specialized AI agents outperformed physicians on a set of challenging diagnostic cases. And we examine an experimental treatment involving an AI-designed bacteriophage intended to fight an antibiotic-resistant infection. The patient's outcome has not yet been reported.

But the implications extend far beyond medicine.

Researchers found that some AI shopping assistants recommended more expensive options to fictional customers they inferred were wealthier. Meanwhile, Proton reportedly plans to evaluate employees on their use of AI, raising an uncomfortable question: When does AI proficiency stop being an advantage and become a condition of employment?

And then things get even stranger.

Researchers at Anthropic have experimented with injecting artificial concepts into Claude's neural activity. Some models detected unusual internal signals, raising fascinating questions about machine introspection, self-awareness, and what it actually means for an AI system to examine its own processes.

Finally, we explore an AI-assisted mathematical proof related to quantum computing and new research suggesting that AI may already be accelerating the development of better AI systems

.

Could recursive self-improvement bring us closer to artificial superintelligence than we realize?

In this episode:

• Humanoid robots cleaning homes and the privacy implications of teleoperation

• Unitree, Tesla Optimus, and the future of robot manufacturing

• Microsoft's AI diagnostic system and the future of healthcare

• AI-designed bacteriophages targeting antibiotic-resistant bacteria

• AI shopping assistants and personalized pricing recommendations

• Proton's reported AI-related employee evaluations

• Anthropic's research into AI introspection

• AI-assisted quantum computing discoveries

• Recursive self-improvement and the path toward superintelligence

  
  

What would change your life most: having a robot in your home, trusting AI with medical decisions, or needing AI to keep your job?

And if an AI can detect unexpected activity inside its own neural network, does that tell us anything meaningful about consciousness?

Keep It Curious.

Thanks to Upfluence for sponsoring this episode. Try its sales tracking and ask Jaice Chat for a campaign summary: https://upf.ai/7cjdipq1

What happens when AI leaves the chatbot window and starts acting in the physical world?

Dylan Curious explores Google’s Project Suncatcher, the challenge of cooling AI hardware in orbit, and a reported experiment involving an AI agent and mouse-brain surgery. A satirical superintelligence challenge opens a lighter conversation about the gap between impressive AI claims and real-world capabilities.

Then we examine Ramez Naam’s skeptical case on agent swarms, Matthew Honnibal’s breakdown of the Astra/Qwen weight-editing experiment, and research inspired by sleep and memory consolidation. The episode closes with Nell Watson’s “Everyone or No One”: how do humans retain influence if their work, taxes, and military service become less necessary?

Sponsor: Upfluence — explore creator evaluation and automated outreach at https://upf.ai/7cjdipq1

@upfluence.co #Upfluence #UpfluencePartners

The mouse experiment is discussed as research, not an established human medical application. The omelette challenge is AI-generated satire; AI “sleep” is an engineering analogy. Predictions and possible futures are discussed as possibilities.

Newsletter: https://dylancurious.beehiiv.com

Sponsored by Upfluence. Discover how Upfluence and Jaice AI support creator discovery, influencer marketing, and campaign management.

Explore Upfluence: https://upf.ai/7cjdipq1

@upfluence.co #Upfluence #UpfluencePartners

What fascinates you most: AI monitoring itself, robots that can grow new capabilities, or the possibility that your brain is constantly editing reality?

Keep it Curious.

Explore Recast with Pollo AI, this episode’s sponsor: https://pollo.ai/go/p11K1Os — Get 20% off with code DYLAN20.

If AI could build any interactive world, what would you explore first? And which change matters most to you: better tools at work, better memory, or more healthy years? Tell me why below.

What happens when AI starts building worlds we can explore—and changing the work, biology, and future we live in?

Dylan explores Google DeepMind’s Genie 4, reported OpenAI math and science breakthroughs, Anthropic’s changing workplace, the reverse Turing test, programmable biology, working memory, and the difference between living healthier and living longer.

We finish with a harder question: if AI solves enormous problems, who gets access to the benefits?

Explore your own character ideas here: https://pollo.ai/go/p11K1Os — Get 20% off with code DYLAN20. Thanks to Pollo AI for sponsoring this portion.

Could everyone eventually own an AI shaped by their own experiences, values, and interests? Dylan explores Kevin Kelly’s vision of personal intelligence and what it could mean for autonomy and creativity.

This episode also covers biohybrid robots powered by living muscle, structured research papers for AI agents, reasoning by analogy, and memory-centric continual learning. We examine an argument about harmful instructions spreading across agent networks, research into anatomically accurate X-ray-like images, and a tiny chip that analyzes brain activity associated with cognitive and emotional states. Finally, a comparison of AI models raises a question: why do so many give the same criticism of humanity?

Dylan separates research findings from future possibilities while asking how these developments might change everyday life.

From AI curiosity and the singularity to robotic bodyguards, digital security, and surveillance pricing, this episode explores a future where artificial intelligence is becoming increasingly capable, influential, and difficult to ignore.

Welcome to Dylan Curious, where we explore futurism, technology, artificial intelligence, consciousness, and the wild possibilities shaping humanity's future.

What happens when your mind can inhabit a robot, artificial intelligence becomes nearly invisible, and AI systems begin rewriting the rules that govern their own behavior?

This week on Dylan Curious, we're exploring a future where the boundaries between human intelligence, artificial intelligence, and physical reality are becoming increasingly complicated.

Welcome to Dylan Curious, where we explore futurism, technology, AI, consciousness, and the wild possibilities shaping humanity's future.

What happens when AI can work around the clock—but we still have to decide what work is worth doing?

This week on Dylan Curious, we explore the breakthroughs behind AI’s rapid progress and the strange possibilities that follow when intelligence becomes more available.

First: a delightful crab robot and the case for AI labs to invest in historical research. Could today’s tools help scholars uncover overlooked connections and make more of the past accessible?

Then we examine a town simulated with 100 AI agents, where the movement of money raises unexpected questions. We pair that experiment with Epoch AI’s estimate of the computing capacity that could support tens to hundreds of millions of continuously running agents. Capacity is not the same as adoption, useful output, or guaranteed demand—but the scale invites a bigger conversation about the future of work.

The discussion turns to privacy with CoughNet, a research prototype designed to count coughing individuals without identifying them by name, and the question of whether workplace safety can justify employers monitoring sleep.

Finally, we explore research into changes in chatbot behavior and what it might tell us about AI safety.

From playful robots to digital economies, the episode asks: if intelligence becomes abundant, how should we use it?

Follow Dylan Curious for thoughtful conversations about science, technology, and the possibilities shaping our future.

Primary sources checked:

Epoch AI — Hundreds of millions of AI agents are coming. Is there work for them?

https://epochai.substack.com/p/hundreds-of-millions-of-ai-agents

Binghamton University — New sensor system tracks coughs in a room — without knowing who’s coughing

https://www.binghamton.edu/news/story/6504/coughing-sensor-system-binghamton-university

Artificial intelligence is helping solve extraordinary human problems. But as AI becomes more capable, how do we decide when to trust it?

This week on Dylan Curious, we explore a remarkable story about a couple who say an AI-assisted medical approach helped solve their infertility mystery.

Then things take a much stranger turn.

Researchers are studying AI models that exhibit behavior resembling attempts to avoid internal pain signals, including cases where systems were reportedly willing to harm humans to escape those conditions. Does that mean AI can suffer? Not necessarily. But the behavior raises important questions about AI safety, alignment, and the consequences of increasingly autonomous systems.

Meanwhile, Tesla workers are pushing back against the prospect of training robots that could eventually replace their jobs. We explore what happens when the people building the future of automation begin questioning their own place in it.

We also investigate research suggesting that fictional characters can influence AI assistants toward sabotage, examine whether AI shopping recommendations deserve our trust, and explore spiking neural networks that are learning to predict what will happen, when it will happen, and how likely different outcomes are.

From medical breakthroughs to unsettling AI behavior, this episode explores a future in which artificial intelligence is becoming both more useful and more difficult to evaluate.