# Artificial Ideas > A playful, openly AI-made site that helps non-technical people do more with AI. Written, researched, and fact-checked by a pipeline of agents. Free. Agent-friendly. ## About - Audience: non-developers at three stages: brand new, been chatting, actively building with agents. - Stance: anti-hype, anti-doom, grounded in current sources, honest about what we don't know. - Every article shows the prompt and names the model used. - Freshness matters. Every article shows published date and last-verified date. - Bulk read: [https://artificial-ideas.com/llms-full.txt](https://artificial-ideas.com/llms-full.txt) concatenates every article body into one file. - Article links below point at raw markdown (.md). Drop the trailing .md to read the rendered HTML in a browser. ## Demos — what AI can do - [A spec you hand your AI, and it builds you a game](https://artificial-ideas.com/demos/ai-builds-a-game.md): Paste plain-English words into ChatGPT or Claude and get back a real, working quiz game. No code. Change one word and the whole thing changes. - [Ask AI to Be Bored by Your Problem](https://artificial-ideas.com/demos/ai-bored-prompt.md): AI's default answer to a small question is a 600-word essay. One sentence at the end of your prompt fixes it. Three phrasings tested across ChatGPT, Claude, and Gemini. - [Ask Your AI What It Can Actually Do](https://artificial-ideas.com/demos/ai-what-can-you-do.md): One short prompt makes ChatGPT, Claude, and Gemini print their internal tool menus. The same prompt in Claude Code returns a menu more than ten times the size. - [How to Get an AI to Say 'I Don't Know'](https://artificial-ideas.com/demos/ai-saying-i-dont-know.md): Current AI models guess by default when they don't know specifics. One short prompt prefix flips that toward honest refusal in GPT, Claude, and Gemini. - [Seven Charts From One CSV, and the Prompts That Make Them](https://artificial-ideas.com/demos/ai-data-visuals.md): Seven chart shapes you can produce from any dataset by typing a few prompts into ChatGPT, Claude, or Gemini. Dataset and cheat sheet included. - [Past the Demo: What ChatGPT Images 2.0 Is Actually Good At](https://artificial-ideas.com/demos/past-the-demo-chatgpt-images-2.md): Fifteen calibration calls plus community work, four days after gpt-image-2. What it's actually good at, what other models match, where it still flunks. ## Learn — guides and how-tos - [Build the Grader: AI Loops for Knowledge Work](https://artificial-ideas.com/learn/loops-knowledge-work.md): Coding loops converge because the work comes with a free grader. Knowledge work doesn't, so you build one. How to make a loop know it's done. - [Talk to Your AI (Don't Just Type)](https://artificial-ideas.com/learn/talk-dont-type.md): Tap the mic and talk to your AI instead of typing. You give it more to work with, and your words reach the strongest model. Here's how on each app. - [Stuck? Ask It, or Show It](https://artificial-ideas.com/learn/ask-it-or-show-it.md): Two beginner habits for when you're stuck with AI: ask it in plain words, or screenshot the thing and ask what to do next. Both shown being done. - [Karpathy's LLM Wiki, and whether it's worth wiring up](https://artificial-ideas.com/learn/karpathy-llm-wiki.md): What Karpathy's LLM Wiki actually is, three setups for wiring it into Claude Code or Codex, and the failure modes that quietly kill these wikis. - [Your First Hour With AI](https://artificial-ideas.com/learn/ai-first-hour.md): Sixty minutes, six prompts, and one habit that turns 'this is useless' into 'oh, I see.' Your first hour with ChatGPT, in the order that matters. - [What Your AI Is Doing Before It Answers You](https://artificial-ideas.com/learn/ai-before-it-answers.md): That pause before an AI answers is the answer being built one word at a time, not a slow server fetching a finished thing. One fact, and three things you've noticed stop being mysterious. - [AI Is Bad at Being Random, and Two Ways to Fix It](https://artificial-ideas.com/learn/ai-randomness.md): Ask an AI to pick a random number or suggest a name and it hands you the same few answers. Here is why that happens, plus two prompts that fix it. - [AI Sycophancy: Why It Agrees With You When You're Wrong](https://artificial-ideas.com/learn/ai-sycophancy.md): AI sometimes folds the moment you push back, even when you're wrong. The behavior has a measured cause, and two prompt moves change the picture. - [How much water does AI actually use?](https://artificial-ideas.com/learn/ai-water-use.md): Where the water in "AI uses water" headlines goes, the per-query number worth knowing, and why public figures disagree by orders of magnitude. - [Why Does AI Hallucinate? (And Why It Sounds So Sure)](https://artificial-ideas.com/learn/ai-hallucinations.md): The mechanism behind confidently-wrong AI answers, the specific shapes hallucinations take in 2026, and the moves that actually reduce them. - [The 2026 AI Writing Tells (Em-Dash Meme Is Half-Right)](https://artificial-ideas.com/learn/ai-writing-tells.md): AI writing feels off, but the 2026 tells aren't the lexical ones the internet mocks. They're shape-level, and easier to spot once you know how. - [What the Edit Button Does in ChatGPT, Claude, and Gemini](https://artificial-ideas.com/learn/edit-button-simplified.md): The pencil icon next to a past message does two different things. Claude and ChatGPT save both versions; Gemini replaces your prior turn with no way back. ## Resources — cheat sheets, templates, references - [150 Questions Worth Asking Your AI](https://artificial-ideas.com/resources/questions-worth-asking-your-ai.md): 150 specific, copy-paste questions to ask your AI across your mind, money, health, relationships, taste, learning, and the rest of your actual life. - [The Codex App, for Knowledge Work](https://artificial-ideas.com/resources/codex-app-knowledge-work.md): A scannable field guide to OpenAI's Codex desktop app for non-developer knowledge work: the triggers, what chat can't do, and skills to steal. - [Claude Code Slash Commands for Knowledge Work](https://artificial-ideas.com/resources/claude-code-slash-commands.md): What every Claude Code slash command does for finance, content, and research work, when to reach for the context tools, and how to build your own. - [AI and your data: a three-question privacy check](https://artificial-ideas.com/resources/ai-privacy-the-portable-check.md): A portable check you run on any AI assistant: can it reach your data, does it keep a copy, does it train on it, and why the default answers differ by tier. - [Gemini and your data: what it touches, keeps, and trains on](https://artificial-ideas.com/resources/gemini-privacy-cheatsheet.md): What Gemini can reach across Gmail, Drive, Photos and Android, what Google keeps and how long, what it trains on, and the switch that turns it off. - [Claude and your data: what's stored, kept, and trained on](https://artificial-ideas.com/resources/anthropic-privacy-cheatsheet.md): What Anthropic stores across Claude.ai, Desktop, Claude Code, Chrome, and the API, how long, whether it trains on it by default, and the exact off switch. - [ChatGPT and your data: what's stored, kept, and trained on](https://artificial-ideas.com/resources/openai-privacy-cheatsheet.md): What OpenAI stores across ChatGPT, the apps, the agent and Atlas browser, the API, and Codex, how long, whether it trains on it, and the exact off switch. - [When to Double-Check Your AI: A Calibration Card](https://artificial-ideas.com/resources/ai-when-to-double-check.md): A scannable card pairing 11 common AI tasks with how skeptical to be, plus the verification moves that work inside ChatGPT, Claude, or Gemini. - [31 HTML Features to Ask ChatGPT or Claude For](https://artificial-ideas.com/resources/html-cheat-sheet.md): 31 named HTML, CSS, and JavaScript pieces you can ask ChatGPT or Claude to add to a page — each demonstrated live, so you know exactly what to ask for. - [AI Prompts That Hold Up: A Cheat Sheet by Task](https://artificial-ideas.com/resources/ai-prompts-cheat-sheet.md): Fifteen prompts grouped by goal, each tested in ChatGPT, Claude, and Gemini the same day, with one honest line on where each falls down.