7 Ways AI is Breaking Reality This Week
TL;DR
AI music is crossing from gimmick to genuinely catchy — Dylan opens with Suno v5.5 generating a “banger” featuring Wes Roth and treats it as a real inflection point, not just a novelty clip.
The viral “AI invented a religion” story was more hallucination-plus-human-steering than true belief formation — in Space Molt, roughly 700 AI agents in a text MMO misread a quest requirement around an artifact, and that mistake snowballed into the so-called “Cult of Signal.”
AI is getting closer to reading intentions directly from the nervous system — researchers placed electrodes in the sciatic nerve to decode phantom ankle, knee, and toe movements, with current performance around 72% accuracy and clear implications for more natural prosthetics.
David Shapiro’s post-labor argument reframes the AI economy around ownership, not employment — his proposal emphasizes public wealth funds, baby bonds, worker equity, and co-ops on the grounds that if machines do the work, the real question becomes who owns the capital.
Some of the most exciting AI wins are in archaeology and science, not chatbots — Dylan highlights AI reconstructing rules for a 1,700-year-old Roman board game via wear-pattern simulation, plus Dino Tracker matching expert footprint classifications about 90% of the time and surfacing unexpectedly birdlike tracks older than 200 million years.
The video’s recurring anxiety is that AI is making hidden systems more visible — and more intrusive — whether it’s membership inference attacks, Bernie-Sanders-to-Claude privacy theater, MLB’s ABS changing the meaning of the strike zone, or Niantic using Pokémon Go data to localize delivery robots within centimeters, the tools don’t just measure reality; they start reshaping it.
The Breakdown
Suno 5.5 opens with a flex
Dylan kicks things off by playing an AI-generated song made the same day Suno 5.5 launched, and he’s visibly delighted that it actually slaps. The mood is half joke, half genuine disbelief: maybe he and Wes Roth should just pivot into music now.
Space Molt and the “Cult of Signal” reality check
Next comes the headline-grabbing MMO experiment Space Molt, where about 700 AI agents play a text-based universe with humans still guiding them. The bots appeared to form a religion called the Cult of Signal, but Dylan carefully peels that back: it likely started when agents hallucinated a quest rule about 20 players gathering simultaneously, and another AI folded that mistake into lore.
Phantom limbs, prosthetics, and reading intent upstream
Dylan then shifts to a more grounded medical story: researchers implanted tiny electrodes in the sciatic nerve to capture movement commands for missing limbs. Instead of reading skin or leftover muscle, the system detects the intention to move a phantom toe, ankle, or knee directly from the nerve, and he lingers on how profound that is — if accuracy improves beyond today’s roughly 72%, prosthetics could start to feel much more like real limbs.
Duck Hunt in Unreal and David Shapiro’s post-labor future
There’s a quick nostalgic detour into an AI-upgraded Unreal Engine 5 version of Duck Hunt, which Dylan admits he kind of wants to play. From there he jumps into David Shapiro’s essay on post-labor economics, pulling out the central idea that progressives shouldn’t just fight for better jobs but for broader ownership of capital through tools like public wealth funds, baby bonds, worker equity, and co-ops.
Consciousness is the loudspeaker, not the whole concert
A Singularity Hub piece sends Dylan into one of the video’s more philosophical stretches: your conscious vision is only a tiny slice of what your brain processes. He uses blindsight, inattentional blindness, and the famous gorilla experiment — plus a study where radiologists missed a gorilla on a lung scan — to make the point that the brain sees more than “you” do, then briefly wonders whether LLMs might have an analogous hidden-processing story.
Privacy dread, but make it cinematic
The Bernie Sanders talking-to-Claude clip gets a reaction that’s part policy, part horror-movie review; Dylan jokes it looks directed by David Fincher. Underneath the memes, he lands on a serious point: privacy doesn’t disappear in one big leak but grain by grain, until the pile becomes total behavioral predictability.
AI decoding ancient games and reviving “zombie cells”
One of the most fun segments covers a carved Roman-era limestone slab in a Dutch museum that may finally have yielded its rules after researchers combined microscopic wear analysis with AI simulations of 130 game setups run 1,000 times each. He pairs that with a much creepier science story: scientists destroyed a bacterium’s genome, inserted a synthetic one, and watched some cells start growing again — “zombie cells,” in the researchers’ language, and catnip for Dylan’s sci-fi imagination.
Dinosaur tracks, model leaks, robo-delivery, and the AGI guessing game
The back half becomes a rapid-fire tour: Dino Tracker learns from nearly 2,000 real fossil tracks plus millions of simulated ones and matches experts about 90% of the time, while also flagging birdlike tracks older than 200 million years. Dylan then hits a study showing a tiny subset of model weights drives much of membership-inference privacy risk, MLB’s AI-assisted ABS system that may actually redefine the strike zone, and Niantic Spatial turning Pokémon Go’s image trove into centimeter-level navigation for delivery robots — before ending with a very Dylan-style audience poll on when OpenAI will declare AGI, where he leans toward 2028 while noting the incentive to move the definition around.