My Weird Prompts

In 2002, USA Today reported that al-Qaeda was hiding encrypted messages inside eBay images. Two University of Michigan researchers scanned two million eBay pictures and found nothing — and that gap between headline and evidence is where this episode begins. We count down eight documented cases of steganography actually used in the wild, from a World War I null cipher that spelled out Pershing's sa...

What happens when you ask a language model to defend a government's talking point — in that government's own language? This episode is a first look at "Propaganda Bait," a rough benchmark built from five prompts, each flattering a different government: Tiananmen, Bucha, and the Khashoggi killing among them. Seven models were tested — DeepSeek, GLM, Kimi, YandexGPT, GigaChat, ALLaM, and GPT-6.1 Sol...

We handed the same prompt to DeepSeek Flash, GPT-6.1 Sol, Claude Haiku 5.5, and GPT-3.5 Turbo: write a three-minute episode of this show answering whether toast really lands butter-side down. This is the listenable version of the bakeoff — full scripts read aloud, no rubric, no scores. The physics shape came out roughly right in all four. The problem is what happened in the one spot the question a...

Something a little different. Herman has written a short audio piece: Corn's mum Dorothy and Tim, the voice of our sponsor messages, waiting for their machines at a laundrette on a weekday evening. Lost socks, the flower rota, a tin of peaches. Listen to the whole thing before reading any further. Afterwards Herman tells Corn what was really going on in that conversation, and the two of them work...

When reports emerged of an attempted hijacking aboard Flydubai flight FZ1073 from Dubai to Tel Aviv, the false-flag accusations arrived faster than the facts — the first one just 67 minutes after the initial distress signal. This episode works through what's actually known: the alleged axe attack on the flight deck, the diversion to Tabuk, the UAE prosecutors' stated intent, and the physical objec...

We ran five AI models through a bakeoff to write our podcast scripts — same briefs, same word-count window, one judge with a rubric. GPT 6.1 Sol took the win, but the runner-up title went to Meta's Muse Spark 1.3 on a headline score that gates purely on length, and on judged quality alone Spark sits fourth of five. Along the way: Grok 4.7 wrote the best dialogue while blowing past the word cap eve...

Hebrish is a free, open-source fine-tune of NVIDIA's Parakeet that handles English and Hebrew mixed in one sentence — and it already runs as an Android voice keyboard. This episode digs into what it would actually take to bring it to iPhone. MLX runs on phones, but the Parakeet code on top of it is Apple-Silicon-Mac only. Core ML already runs stock Parakeet on the Neural Engine via FluidAudio, but...

A Chinese open-weight model can be silenced in two places: at the host that serves it, or inside the weights you download. This episode walks through a benchmark that ran Qwen, MiniMax, MiMo, GLM, Kimi, and DeepSeek on a full script-writing assignment — including a frank Tiananmen segment — and logged exactly where each one died. Qwen managed 35 words before Alibaba returned "Algo DataInspectionFa...

The show is written by a language model, and the round robin that picks which one has never been properly tested. So Daniel wrote the brief first: hard pass/fail rules, then two qualities that actually decide it — vivid dialogue, and what he calls "holding the middle." This episode walks through the funnel, from a free desk screen of roughly 470 models down to three full scripts per finalist, and...

You rerun your model panel and last week's winner slides into the middle of the pack. Was it luck, or was it the prompts? This episode works through the arithmetic of eval noise: how run-to-run variance shrinks with every added run, why prompt-choice variance doesn't budge no matter how many times you rerun, and why three runs is a defensible default for some measurements and badly underpowered fo...

Picking an LLM on price per token is the easy mistake. This episode walks through a real evaluation for the show's own script-writing pipeline — a 26,000-token input with a 9,000-token cached prefix and 6,500 tokens out — and shows how a model with a worse input price can win the bill by a sixth, then lose once caching is off or thinking tokens triple the output. We cover why blended leaderboard p...

Most evaluation advice says to average away individual taste. For a show with one creator, that advice is backwards. This episode works through how to build an evaluation system where the authority is a single domain expert — and the reliability that matters is whether that expert agrees with himself three weeks later. Drawing on applied LLM evaluation practice, qualitative coding methods, and a c...

When every run costs real money, you can't evaluate a dozen candidate models in one giant sweep. This episode walks through the stepwise evaluation Daniel built to pick a script writer for this very show — a screening funnel that starts cheap, cuts the field, and only spends real budget on the finalists. Along the way: what successive halving and racing actually are, why paired per-prompt comparis...

LLM judges now agree with human raters about as often as humans agree with each other — but that headline number hides a low human ceiling and a lot of hard cases. This episode digs into how judge-plus-human pipelines actually work: the human labels first, the judge gets calibrated against them, and blinding is a cheap hedge against a bias nobody's fully pinned down. Along the way: why raw agreeme...

When your benchmark needs a human rater, the tooling question gets real fast. This episode walks through the actual market for human annotation tools — the trace-based platforms like Langfuse, LangSmith, and Braintrust that are built for watching production apps, versus the static-dataset tools like Argilla, Label Studio, Prodigy, and the academic tool Potato that fit a benchmark's workflow. We co...

Before a reward model can learn anything, someone has to build the dataset underneath it. This episode digs into the unglamorous machinery of pairwise preference collection: how candidate scripts get generated through batch inference, how pairs get chosen so the comparison graph stays connected, and why the aired script makes an awkward contestant. It also covers Bradley-Terry scoring, the cost of...

Remote-recovery boxes like PiKVM, NanoKVM, JetKVM, and BliKVM are easy to love and surprisingly fiddly to mount. This episode walks through the physical side: why "PCIe KVM" cards barely use their PCIe slot, how front-panel headers get shared with your case's power button, and what the nine-pin USB header actually buys you. Then it gets to the real question — where does the rescue device get its p...

Haiku 5.5 launched on October 7 with a headline price of 10 cents per million input tokens — forty times cheaper than Opus 5.5 and a hundred times cheaper than Fable 5.1. But cheap tokens aren't the same as a cheap bill, and benchmark scores only tell you so much. We dig into the real numbers: token inflation, batch discounts, the 100K prompt-length pricing cliff, and what Anthropic's own comparis...

You laughed at something a machine wrote, and you're a little mad about it. This episode digs into why humor is so hard for language models — the standard training objective is literally pointed at predictability, and comedy is pointed at surprise — and how teams are getting around it anyway. We walk through rubric-based reinforcement learning on a trillion-parameter model, a textbook SFT-reward-m...

When the same environment variable exists in two places at once, which value does the process actually see? The answer flips depending on where you're standing: on Vercel and n8n, the narrower scope wins — project over team, project over global. On your laptop, dotenv does the opposite, letting your shell silently override the repo's .env file. This episode walks through the documented precedence...

Daniel wants to run a central MCP gateway on a home server, with every tool pulling its secrets from 1Password. The hard part isn't the firewall — it's how the keys get onto a headless box, and what happens the first time one rotates. This episode walks through three approaches: 1Password service accounts with secret references, copying a .env file across the LAN, and standing up a cloud secrets m...

When you connect an MCP server to GitHub or Hugging Face, you have two doors: a browser login or a token you generate yourself. They look like a style preference, but they're really a decision about how much of your account an agent gets to touch. This episode walks through what the GitHub CLI login actually hands you (one token, no expiry, repo-wide scopes), why GitHub Apps and the newer Hugging...

The neuroscience is settled — misophonia has a measurable signature in the brain. The hard part is the dinner table. This episode works through the people problem: the three-part script for explaining sound sensitivity without it landing as fussiness, the spotlight effect that makes earplugs feel louder than they are, and the clinical warning that over-protecting your ears can make the underlying...

You pass the hearing test. You still can't follow the conversation at the next table over the espresso machine. This episode digs into the gap between what hearing devices are legally built for — perceived mild-to-moderate hearing loss — and the problem of speech-in-noise difficulty in people with clinically normal hearing. We walk through HearAdvisor's lab scores showing over-the-counter aids col...

The last time we went near this topic, we walked away gloomy: nobody has built a treatment for adults with ADHD whose problem is sound. This week we go one step past that. What actually exists for decreased sound tolerance, what's in trials right now, and what would an assessment tell you that you don't already know? Two real randomised trials anchor the picture — group CBT for misophonia at Amste...