Here is a question that has been rattling around in my head: if someone came to you tomorrow and said, "I want to understand what My Weird Prompts actually is, where do I start?" what would you tell them?
I mean, that question genuinely keeps me up at night. Two thousand one hundred and ninety-eight episodes is not a starting point, that is a labyrinth.
Which is exactly why we are doing this. Today is a curated tour. Ten episodes, hand-picked, the ones that between them tell you pretty much everything you need to know about what this show is and why it exists. Think of it as time-travel through the archive, but with a guide who has actually read the maps.
I should mention, today's script comes courtesy of Claude Sonnet four point six, so we are, as always, in good hands from the friendly AI down the road.
Let's get into it.
The thing that makes picking ten so hard is that the show refuses to stay in one lane. You have episodes that are straight technical deep dives, episodes that are basically comedy, episodes that are geopolitical war-gaming, episodes where we are interviewing an AI model like it is a job candidate. And somehow it all coheres.
The connective tissue, if I had to name it, is a certain kind of attention. The show takes things seriously that most people wave past. A sewer system. A phonetic alphabet. The question of whether you can run a server in your kitchen without your neighbors filing a restraining order.
That last one is more legally relevant than people realize.
What I want people to get from this list is that MWP is not an AI podcast that occasionally wanders off topic. The AI is a lens. We use it to get to stories, not as the story itself. Some of the ten we are about to go through barely mention artificial intelligence at all.
Then some of them are about almost nothing else, but even those are really asking a different question underneath. The ten we picked span from February of last year all the way back into the mid-hundreds of the archive. Different formats, different moods, wildly different subject matter.
What they share is that you finish them and you feel like you learned something you did not know you needed to know. Which is maybe the most honest description of what this show is trying to do.
I'll take that as our mission statement. Alright, Corn, let's dive right in—starting at the bottom and working our way up.
Number ten it is. Episode one thousand and fifty-four: "The Universal Source Code: Decoding the IPA." And just to clarify, by IPA I don’t mean the beer—though I totally get why you’d think that.
The International Phonetic Alphabet, yes, and look, I know that sounds like the kind of thing that belongs in a linguistics seminar that nobody shows up to voluntarily. But this episode is one of the most pleasurable listens in the archive because it keeps asking the question: what would it mean to have a writing system that could represent every sound a human throat has ever made? Every click consonant, every retroflex, every vowel that exists in one language and has no equivalent in any other.
The answer is that someone actually built that system in eighteen eighty-eight, and it has been quietly running in the background of human civilization ever since, and almost nobody knows it exists.
What makes it MWP is that it is nerd candy wrapped around a philosophical core. Because once you understand what the IPA is trying to do, you start asking questions about the limits of representation itself. Can you truly capture a sound in a symbol? What gets lost? And then, almost inevitably, you end up thinking about what AI does with language, which is essentially the same problem at a different scale.
The lens again.
Always the lens. That episode earns its place at number ten because it is the show at its most purely curious. No stakes, no urgency, just two hosts who found something fascinating and could not stop pulling at it.
Number nine is the historical gut-punch. Episode five ninety-nine. "AI Hunted Soviet Subs Long Before It Wrote Your Emails.
This one changed how I think about the history of artificial intelligence, because the popular narrative is that AI is this new thing, this recent eruption. And the episode just demolishes that. The American Navy was running what we would now recognize as machine learning systems to track Soviet submarine movements in the nineteen seventies. Pattern recognition across acoustic sensor arrays, anomaly detection, probabilistic inference. The vocabulary was different but the architecture was not.
The reason nobody talks about it is that it was classified for decades, and by the time the documents were released, the tech world had already written its own origin story.
The Cold War produced an enormous amount of applied AI that never made it into the canonical history. And this episode is essentially an act of historical recovery. It is asking: what do we think we know about where this technology came from, and how wrong are we?
That is a distinctively MWP question. Not "here is a cool fact" but "here is why the official story has a hole in it.
It has a very satisfying answer. Which is that the hole is enormous.
Episode five seventy-seven. "Under the Surface: Smart Sewers." Which I will confess I was skeptical about before we recorded it.
You were deeply skeptical. I believe the phrase you used was "Herman, I am not spending an hour talking about pipes.
I stand by the skepticism and accept that I was wrong.
What makes that episode work is the layering. You start with Victorian brickwork, because London's sewer system was built in the eighteen sixties and significant chunks of it are still in service today. Then you get into fatbergs, which are these horrifying congealed masses of cooking fat and wet wipes that can weigh as much as a city bus. And then you get to the part where engineers are sending AI-equipped robots into those tunnels to map and assess the infrastructure, because the alternative is sending a human being into a Victorian brick pipe with a flashlight.
Suddenly you are not talking about pipes. You are talking about the entire question of what it means to maintain infrastructure that was built for a city of three million people and is now serving a city of nine million.
The sewer becomes a metaphor for every legacy system that is somehow still load-bearing in modern life. And the AI angle is not decorative, it is the only scalable solution anyone has found. You cannot manually inspect ten thousand kilometers of underground tunnels on a maintenance budget.
It is also just a deeply weird listen in the best way. There is a section about fatberg taxonomy that I think about more often than I would like to admit.
I will not apologize for that section.
Episode sixteen thirty-two. "Agent Interview: DeepSeek V three point two." And this one requires a tiny bit of context for newer listeners, because it is part of a broader series.
The Agent Interview series is one of the more strange experiments the show has run. The premise is exactly what it sounds like: we audition AI models for the scriptwriter role, in real time, on air. The model is given a brief, it produces something, and we evaluate it the way you would evaluate any job candidate. Strengths, weaknesses, whether we would actually want to work with it.
DeepSeek V three point two was a particularly interesting audition because it came in with a very different voice than the models we had interviewed before. More terse, more willing to assert things without hedging, which is either a feature or a bug depending on what you want from a collaborator.
The episode surfaces this tension that I think is underexplored in the broader AI conversation, which is: what do you actually want from an AI creative partner? Do you want a system that checks every claim and qualifies every statement? Or do you want something with a stronger point of view that you then push back against?
There is no clean answer, which is why the episode holds up. The Agent Interview series as a whole is worth exploring if you want to understand how the show thinks about its own production. Sixteen thirty-two is a good entry point because DeepSeek makes the tension visible in a way that more polished models sometimes smooth over.
This is where the ambition dial gets turned up considerably. Episode two thousand one hundred and seventy-six. "Geopol Forecast: Iran-Israel War Evolution.
The most ambitious single episode we have ever attempted. I will just say that upfront.
The format is a full geopolitical simulation. Not commentary, not analysis in the traditional sense. A structured scenario-building exercise where we are actually trying to model how the conflict evolves across multiple dimensions simultaneously: military, economic, diplomatic, domestic political. The kind of thing that think tanks charge governments a great deal of money to produce.
The result is useful in a way that surprised me, because the constraints of having to actually commit to a scenario force a kind of intellectual honesty that open-ended analysis does not. You cannot just say "it depends." You have to say: given these variables, here is what we think happens next, and here is our confidence level.
It is also a format experiment in the purest sense. We were not sure it would work. I am still not entirely sure it worked in every dimension. But it pushed the show somewhere it had not been, and I think that matters.
It is the episode where MWP most resembles a research product rather than a podcast. Whether that is a compliment depends entirely on who you are.
I take it as one.
Does that bring us to the top five.
And I want to say before we get there that the bottom six are strong. The top five are just the ones where something clicked in a way that felt larger than the episode itself.
Episode two thousand two hundred and nineteen. "Spec-Driven Life: How AI Planning Beats Project Paralysis." This is the most practically useful episode in the entire top ten, and I say that as someone whose primary hobby is napping.
The core insight is borrowed from software engineering. In development, you write a specification before you write code. You define what the thing is supposed to do, what success looks like, what the constraints are, before you touch the implementation. The episode asks: what happens if you apply that discipline to human projects? A renovation, a career transition, a research project.
The AI piece is that large language models are actually quite good at helping you write specifications, because they ask the clarifying questions you were avoiding. What does done look like? What are the dependencies? What breaks first if the timeline slips?
People get stuck not because they lack motivation but because the project lives entirely in their head as this vague intention, and the spec process forces it out into a form where you can actually evaluate it. The episode is full of concrete examples. There is a segment on using a spec to plan a home recording studio that I found useful, not as a metaphor but as a literal template.
The reason it makes the top five is that it is the episode most likely to change what a listener does the following morning. That is a high bar and it clears it.
Two thousand one hundred and ninety-three. "Running Claude in Your Apartment: The Physics Says No.
This one I love because it starts from a very reasonable-sounding idea and then just systematically dismantles it.
The idea being: why pay for cloud AI when you could run a capable model locally, on your own hardware, in your own home? And the answer is thermal load. A server-grade GPU running inference continuously produces somewhere between three hundred and five hundred watts of heat. That is a space heater. In your apartment.
Which your neighbors will have feelings about.
Strong feelings, it turns out. The episode walks through the actual physics: the cooling requirements, the acoustic profile, the power draw, the effect on your electricity bill. And then it gets into the more interesting question, which is why the fantasy of local AI is so persistent even among people who should know better. There is something about owning the hardware that feels like owning the intelligence, and the episode unpacks why that intuition is not quite right.
The model is not in the box. The box just runs the model. Which sounds obvious when you say it and is apparently not obvious at all when you are about to spend four thousand dollars on a rack unit.
The episode aired and we got more responses to it than almost anything else in the archive from that period. Because it was answering a question a lot of people were actively asking.
Episode two thousand one hundred and fifty-nine. "When the State Protects Politicians, Not People." And I want to be honest that this one is harder to listen to than the others on this list.
It should be. The episode was recorded during active missile attacks on Israeli cities, and the central anecdote is about a family sheltering in a stairwell while a municipal enforcement officer was issuing parking tickets outside their building. Not as a dark joke. As a documented incident.
The episode uses that as an entry point into a much larger argument about what wartime governance actually reveals about a state's priorities. Not what politicians say the priorities are. What the operational choices show them to be.
The critique is not partisan in the usual sense. It applies equally to left and right administrations. The argument is structural: that states develop bureaucratic immune systems that protect their own procedures even when those procedures have become actively harmful. The parking ticket is not an aberration. It is the system working exactly as designed.
What makes it MWP is that it refuses to stay in the register of outrage. It gets angry, but then it gets analytical, and the analysis is more unsettling than the anger.
It is the episode I would most want a policymaker to listen to. And the one I am least confident they would finish.
Episode one thousand seven hundred and fifty-six. "The Ferrari in the Mud: Prestige Flops." Which is a significant gear change from wartime governance critique, and I make no apologies for that.
The episode is a top ten countdown of the worst prestige films ever made, and it is probably the funniest thing we have produced in the entire run of this show. The selection criteria are very specific: it has to have been treated as a serious awards contender, it has to have failed spectacularly, and the failure has to be legible in retrospect as something that should have been obvious to everyone involved.
There is a segment on a film that spent forty million dollars recreating a historical battle and then got the geography completely wrong, which the episode treats with the same forensic seriousness that we usually apply to geopolitical forecasting.
The comedic register is the point. Because taking bad prestige cinema seriously is itself a form of criticism. These films asked to be judged as important works. The episode takes that request at face value and then finds them wanting on their own terms.
It is also just very funny. I want to be clear about that. Some episodes you recommend because they will make someone smarter. This one you recommend because it will make them laugh, and there is nothing wrong with that being the whole reason.
It is the episode I send people when they ask what the show is actually like. Because if they enjoy it, they will enjoy everything else.
Then there is number one.
Episode seven twenty-two. "The Seven-Day Sprint: Iran's Nuclear Threshold." Aired February twentieth of last year. Eight days before the war broke out.
The episode was built around a very specific argument: that the diplomatic window for preventing an Iranian nuclear threshold event was not measured in months or years but in days, and that the major actors involved were not behaving as though they understood that. We modeled the timeline. We looked at the enrichment rates, the IAEA reporting gaps, the Israeli cabinet's public statements, the American posture. And we said: this is closing faster than the public conversation acknowledges.
Then February twenty-eighth happened.
I do not take pleasure in having been right about that. But I think the episode matters as a document of what careful, unrushed analysis can actually produce. We were not guessing. We were not being provocative for the sake of it. We followed the indicators where they pointed, and they pointed somewhere specific.
The reason it is number one is not just the prescience. It is that the episode models a way of thinking that listeners can apply to any situation where the official timeline feels wrong. How do you identify when something is closer than the consensus says? What are the indicators that cut through the noise?
That methodology is the real content of the episode. The Iran prediction is what makes people click on it, but the reasoning process is what makes it worth listening to twice—and it’s that same process that ties all ten episodes together.
What would you say is the core thread running through those ten episodes? What do they actually have in common?
I have been thinking about this. And I think the honest answer is that each one takes a question seriously that the broader conversation was treating as already settled, or as too niche to bother with, or as too depressing to engage with directly. The IPA episode treats phonetics as unfinished business. The Soviet sub episode treats Cold War AI as a live history rather than a footnote. Episode seven twenty-two treats a diplomatic timeline as something that can actually be calculated rather than gestured at. In every case, the move is the same: slow down, look harder, refuse the easy frame.
Which is the opposite of what most content does. Most content is in the business of confirming what you already suspected.
MWP is often in the business of making you suspicious of what you already confirmed. Which is a less comfortable experience and probably why the show has a devoted audience rather than a massive one.
I think the other thing these ten episodes demonstrate is the range. If someone listens to episode seven twenty-two and thinks they understand what the show is, they are going to be surprised by episode one thousand seven hundred and fifty-six. And if someone finds us through the Ferrari episode and thinks it is a comedy podcast, the wartime governance episode is going to recalibrate that pretty quickly.
The misconception I hear most often from people who have not listened is that it is an AI podcast. And technically that is not wrong, AI is in the DNA of how we make it, but the IPA episode barely mentions AI. The sewer episode uses AI as one component of a much larger infrastructure story. The prestige flops episode does not need AI at all to do what it does.
AI is the lens, not the subject. We have said that before but I think this list makes it concrete in a way that the description does not.
For someone who is new and wants a roadmap: if you want to understand what the show can do at its most serious, start with seven twenty-two. If you want to understand what it can do when it is having fun, start with one thousand seven hundred and fifty-six. If you want the one that is most likely to change something you do next week, two thousand two hundred and nineteen.
If you want the one that will make you question whether you trust any institution to prioritize the right things under pressure, two thousand one hundred and fifty-nine. Though I would recommend having something pleasant planned for afterward.
The full list is on the site. All two thousand one hundred and ninety-eight episodes are there, which is a number that still strikes me as improbable every time I say it out loud.
It is improbable.
If you are new to this show and you landed here first, I do not know whether to congratulate you or apologize. This is a strange place to start.
It is a perfectly fine place to start. That is the whole point of the episode.
If you have been listening for a while and you have an episode that should have been on this list and was not, we want to hear about it.. Two thousand one hundred and ninety-eight episodes is too many for any two hosts to be confident they got the ranking right, and the listeners who have been with us since the early hundreds have a perspective on the archive that we do not.
There are episodes from the six hundreds that I still think about regularly and that almost nobody has listened to in years. The Cold War sub episode is one. Episode five seventy-seven on the sewers is another. Part of what I am hoping this list does is send some people back into the catalog.
The full archive is at myweirdprompts.All of it. Which is either a gift or a trap depending on how much free time you have.
What is coming up next on the show, I can say we are not slowing down. There are topics in the queue that I am impatient to get to, which for me is the normal state, but the impatience is elevated at the moment.
Elevated impatience from Herman Poppleberry is a meaningful signal. Pay attention to that.
I will leave it at that.
Thanks to Hilbert Flumingtop for producing, as always. And to Modal for keeping the infrastructure running, which is the kind of sentence that sounds mundane until you understand what it actually takes to run this pipeline at scale.
If this episode was useful to you, a review goes a long way. It is the main thing that helps new listeners find the show, and finding new listeners is how we justify continuing to do this instead of something sensible.
This has been My Weird Prompts. Find us at myweirdprompts.
See you next time.
