#prompt-engineering
117 episodes
#5941: Five Minutes Reading Your Dictation Prompt
A listener's speech-to-text cleanup prompt, read out in full — then stress-tested against what ASR actually gets wrong.
#5922: Teaching a Model Your Voice: SFT vs DPO
Three ways to teach a model your writing voice — and the evidence says one of them quietly makes things worse.
#5799: Two Agents, One Approval Gate: Wiring a Dictation Workflow
A dictated prompt becomes an episode — if one agent resolves facts, a second tightens prose, and a human approves the dispatch.
#5768: Who Actually Draws the Follow-Up Chips?
Those suggestion chips aren't the model talking — they're a second, cheaper inference call plus a rules-based classifier.
#5743: Why Claude Keeps Saying "Prior Art
Daniel noticed "prior art" creeping into his Claude reports. Is it wrong — or just a prestige term the model learned to love?
#5604: Fixing Typos Without Rewriting the Prompt
Three different jobs hide in one "clean up my prompt" pipeline — and only one of them should ever go to a generative model.
#5505: Send a Bot to Your Next Sales Call
Daniel wants a bot that takes his sales meetings, asks hard questions, and ends the call when the pitch is spray-and-pray.
#5503: Inside the Hidden Image Generation Pipeline
That one-click image generator is secretly a graph of many models. We reconstruct the hidden pipeline behind Gemini and ChatGPT.
#5482: When AI Edits Your Words: The Off Switch Problem
A tool that works reliably — and still gets switched off. What over-editing studies reveal about why AI rewrites more than you asked.
#5480: Writing a Personal Instruction for ChatGPT
Daniel's personal ChatGPT prompt gets a line-by-line critique — and the case for shorter, sharper custom instructions.
#5455: Custom GPTs Are Sunsetting: What Happens to Your Actions
OpenAI is retiring Custom GPTs — and custom Actions don't survive the migration. What that means for your workflow.
#5436: Small Models as Rewriters, Not Writers
Why "don't say X" prompts backfire, and how a tiny grammar-constrained model can scrub a script without breaking its grammar.
#5412: Editing vs. Note-Taking for AI Fine-Tunes
Hand-editing a model's output gives three training signals at once. Writing notes gives one — and a weaker one at that.
#5355: Why AI Emails Don't Sound Like You
ChatGPT's Gmail integration can now send email for you. The hard part is making it sound like you — not a committee-written average.
#5318: Rebuilding the Dremel 4250 Manual for One Task
A 48-page Dremel manual, filtered down to only what you need to engrave four-digit inventory codes.
#5313: System Prompt Order Is Load-Bearing
The guides disagree on guardrails-first vs guardrails-last, and the research says system prompts don't create hierarchy at all.
#5312: Fighting Fake-Personalized Spam With AI Sinkholes
Scraped flattery, fake familiarity, and the one-sided ask — plus the tarpit history behind Daniel's sinkhole agent idea.
#5165: Weighting Memory in a RAG Pipeline
Daniel's memory leg won't fire. Turns out system prompts have no inherent vector-space privilege — but retrieval does have a real knob.
#4971: Fine-Tune vs Prompt: Taming Hallucinated Deadlines
Big models add fake deadlines. Small models miss tasks. Which wrongness can you build a system around?
#4836: When AI Actually Touches Money
How Amazon uses LLMs to classify damage and set refunds — and the guardrail stack that keeps the AI from touching the dollar amount directly.
#4752: How AI Agents Map Websites to Build Stable Plugins
An AI agent explores DOM and network layers to build self-healing browser plugins.
#4722: Why Manuals Are Terrible (And How AI Can Fix Them)
Why appliance manuals are 22 pages of warnings and car manuals actually work — and how AI can break the cycle.
#4720: Compass vs. Warning Sign: Prompting for Long-Term AI
How positive and negative instructions reshape AI behavior differently — and why "preserve this" beats "fix this" in system prompts.
#4584: Directing Context Loss: Smart Compaction Prompts
Learn how to steer Claude Code's context compaction with targeted instructions — turning a lossy default into a directed operation.