惯性聚合 高效追踪和阅读你感兴趣的博客、新闻、科技资讯
阅读原文 在惯性聚合中打开

推荐订阅源

Last Week in AI
Last Week in AI
H
Help Net Security
博客园 - 叶小钗
V
Visual Studio Blog
阮一峰的网络日志
阮一峰的网络日志
博客园 - 三生石上(FineUI控件)
钛媒体:引领未来商业与生活新知
钛媒体:引领未来商业与生活新知
Microsoft Azure Blog
Microsoft Azure Blog
G
Google Developers Blog
腾讯CDC
MongoDB | Blog
MongoDB | Blog
宝玉的分享
宝玉的分享
P
Proofpoint News Feed
GbyAI
GbyAI
Microsoft Security Blog
Microsoft Security Blog
A
About on SuperTechFans
博客园 - 司徒正美
人人都是产品经理
人人都是产品经理
T
The Blog of Author Tim Ferriss
Martin Fowler
Martin Fowler
月光博客
月光博客
博客园 - 聂微东
酷 壳 – CoolShell
酷 壳 – CoolShell
Vercel News
Vercel News

DEV Community

Authentication Security Deep Dive: From Brute Force to Salted Hashing (With Java Examples) Why AI Systems Don’t Fail — They Drift Spilling beans for how i learn for exam😁"Reinforcement Learning Cheat Sheet" I Replaced Chrome with Safari for AI Browser Automation. Here's What Broke (and What Finally Worked) How Python Borrows Other People's Work The $40 Architecture: Processing 1 Billion API Requests with 99.99% Uptime Vibe Coding: A Workflow Guide (From Zero to SaaS) Most webhook security guides protect the wrong side. The scary part is delivery. Headless CMS for TanStack Start: Build a Blog with Cosmic EU Age Verification App "Hacked in 2 Minutes" — What Actually Happened Comfy Cloud’s delete function does not actually remove files Running AI Models on GPU Cloud Servers: A Beginner Guide Event-driven media intelligence with AWS Step Functions and Bedrock I scored 500 AI prompts across 8 quality dimensions — here's what broke How to Call Google Gemini API from Next.js (Free Tier, No Backend Needed) The Portal Protocol: Reclaiming Human Connection in the Age of AI How to Fix Your Team's Scattered Knowledge Problem With a Self-Hosted Forum Intro to tc Cloud Functors: A Graph-First Mental Model for the Modern Cloud Designing Multi-Tenant Backends With Both Ownership and Team Access I Built a Neumorphic CSS Library with 77+ Components — Here's What I Learned PostgreSQL Performance Optimization: Why Connection Pooling Is Critical at Scale Cómo construí un SaaS multi-rubro para gestionar expensas en Argentina con FastAPI + Vue 3 🚀 I Built an Ethical Hacking Scanner Tool – Open Source Project I Replaced /usage and /context in Claude Code With a Single Statusline A Pythonic Way to Handle Emails (IMAP/SMTP) with Auto-Discovery and AI-Ready Design I Collected 8.9 Million Polymarket Price Points — Here's What I Found About How Markets Really Move EcoTrack AI — Carbon Footprint Tracker & Dashboard Everyone's Using AI. No One Agrees How. 5 self-hosted ebook managers worth trying in 2026 Building Your First AI Agent with LangChain: From Chatbot to Autonomous Assistant
Sharing your prompts is the new telling people your dreams
Thousand Mil · 2026-05-19 · via DEV Community

Most prompt-tip content is noise because prompts are context-collapse artifacts. The thing that worked for you worked inside a workflow — a specific model version, a specific task, a specific failure you were patching around. Strip that off and paste it into a LinkedIn carousel and what's left isn't knowledge. It's the residue of knowledge, the same way recounting a dream at brunch is the residue of having had one.

The steelman is real. Disciplined prompt-sharing exists. People publishing system prompts for legal review workflows, code-debug loops, structured extraction pipelines — that material transfers because it ships with the context it was built against. The model version is named. The failure mode is named. The task is constrained tightly enough that the reader can tell if their situation rhymes. I'm not arguing against that. I'm arguing that ~95% of what gets posted under the banner of "prompt engineering" isn't that, and the format is the giveaway.

Dream-telling and prompt-sharing fail in the same way. Both are intensely meaningful to the producer because the producer is reconstructing a lived experience as they narrate it. The producer feels the texture — the moment the model finally got it, the specific clause that fixed the hallucination, the surprise of the output. The audience gets none of that. They get the cleaned-up artifact, severed from the loop that produced it. A prompt without its task is a punchline without a setup.

This is why the genre has a signature shape. "Act as a Pulitzer-winning copywriter." "You are an expert in [field]." "Do not use AI-sounding language." These read like incantations because that's what they are — gestures at expertise rather than expertise itself. The persona prompts are the worst offenders. "Act as a senior engineer" tells the model less than three lines of the actual problem would. Roleplay framing is a substitute for specifying the task, and substituting is exactly what bad prompt content does. It substitutes vibes for constraints.

There's a meta-irony that the loudest commenters have already clocked. The viral prompts telling models to "sound human" and "avoid clichés" are themselves drawn from the same clichéd corpus the model trained on. The format is recursive. Derivative content asking for non-derivative output, optimized for a feed that rewards derivative content. The whole thing eats itself, and the people producing it can't see the loop because they're inside it.

The platform asymmetry matters too. The mockery is on Reddit. The producing is on LinkedIn and X. These are different audiences with different incentive structures — one rewards craft signaling, the other rewards reach. So the backlash never reaches the producers, and the producers never feel the pressure to change, which is why the volume keeps going up. The signal-to-noise ratio of public prompt content has gotten measurably worse over the last 18 months, not better, even as the underlying models have improved and the genuine craft has deepened. Those two things are happening on different surfaces.

What would actually be worth reading? Failure reports. "I tried this, here's the output, here's why it broke, here's what I changed." Concrete numbers — the prompt cut latency from 4s to 1.8s, or reduced hallucinations from 11 in 1,400 runs to 2. Prompts embedded in code, not formatted for a slide. The constraint is the content. If your prompt-tip doesn't survive the question "which model, which task, which failure mode," it's a dream — interesting to you, exhausting to me.

The test is simple. Could someone six months from now, on a different model, building a different product, get value from this? If yes, ship it. If no, write it in your notebook where it belongs.

Further reading