AI-Truth
starsforks
How to install AI-Truth?
kunkun2116/ai-truth
$git clone https://github.com/Kunkun2116/AI-TruthSee when AI is making things up.Make AI more honest — and show its receipts.
README.md
| 1 | # AI Truth |
| 2 | |
| 3 | **See when AI is making things up.Make AI more honest — and show its receipts.** |
| 4 | |
| 5 | Evidence labels for Claude, ChatGPT, and Gemini. |
| 6 | Show what was **searched in this conversation**, what comes from **memory**, and what is **inference** — instead of presenting everything in the same confident tone. |
| 7 | |
| 8 | <p align="center"> |
| 9 | <img src="https://raw.githubusercontent.com/Kunkun2116/AI-Truth/main/docs/ai_truth_homepage_mockup.png" alt="AI Truth — evidence labels for AI responses" width="720"> |
| 10 | </p> |
| 11 | |
| 12 | ### Demo |
| 13 | |
| 14 | https://github.com/user-attachments/assets/f06293c7-12ce-4cde-b0c4-b5b708c0c122 |
| 15 | |
| 16 | --- |
| 17 | |
| 18 | ## ✨ Why this exists |
| 19 | |
| 20 | LLMs often mix **facts**, **memory recall**, **analysis**, and **guesses** into one seamless answer. |
| 21 | The problem is not only hallucination — it is that everything can *sound equally certain*. |
| 22 | |
| 23 | AI Truth makes that difference visible. |
| 24 | It combines a structured credibility prompt with a Chrome extension that surfaces evidence labels directly in the chat UI. |
| 25 | |
| 26 | Read the [Design Journey](./docs/design-journey.md) — the reasoning, the mistakes, and the iterations behind the framework. |
| 27 | |
| 28 | ## 📦 What this repo includes |
| 29 | |
| 30 | - **A credibility prompt framework** for Claude, ChatGPT, and Gemini |
| 31 | - **Two prompt versions** for different use cases: |
| 32 | - **Compact** for always-on personalization / custom instructions |
| 33 | - **Full** for tighter control in new chats, research, and fact-checking |
| 34 | - **Inline evidence labels** such as `[S1]`, `[M2+R2]`, `[S3+R2+F]`, `[U+C]` |
| 35 | - **Risk prefixes** for high-risk domains like `⚠Legal`, `⚠Finance`, and `⚠Medical` |
| 36 | - **A Chrome extension** that can: |
| 37 | - visualize labels in AI responses |
| 38 | - copy the framework prompt in one click |
| 39 | - inject prompt versions into supported chat interfaces |
| 40 | |
| 41 | ## 🏷️ What the framework does |
| 42 | |
| 43 | ### 1) Pre-output review |
| 44 | |
| 45 | Before answering, the model runs a **hidden 6-point check** on key claims: |
| 46 | |
| 47 | 1. factual accuracy |
| 48 | 2. unsupported conclusions |
| 49 | 3. time-sensitivity |
| 50 | 4. reasoning gaps |
| 51 | 5. missing premises |
| 52 | 6. completeness |
| 53 | |
| 54 | Claims that fail should be revised, downgraded, or marked uncertain before the final answer is shown. |
| 55 | |
| 56 | ### 2) Evidence labeling |
| 57 | |
| 58 | The final answer labels claims by evidence family, so the user can see what kind of support each statement actually has. |
| 59 | |
| 60 | | Family | Tags | Meaning | |
| 61 | |--------|------|---------| |
| 62 | | **S** (Searched) | S1, S2, S3 | Actually searched in this conversation: multi-source verified → single strong source → weak / secondary source | |
| 63 | | **M** (Memory) | M1, M2, M3 | Stable consensus → possibly outdated → time-sensitive, should search first | |
| 64 | | **U** (User) | U | User-provided, not externally verified | |
| 65 | | **R** (Reasoning) | R1, R2, R3 | Mechanically verifiable → framework-dependent → open synthesis | |
| 66 | | **C** (Creative) | C | Generative ideation or design | |
| 67 | | **F** (Fragile) | F | Insufficient, conflicting, or missing support | |
| 68 | |
| 69 | Examples: `[S1]` `[M2+R2]` `[S3+R2+F]` `[U+C]` |
| 70 | |
| 71 | High-risk content should also use domain prefixes such as: |
| 72 | `⚠Legal` `⚠Finance` `⚠Tax` `⚠Medical` `⚠Safety` `⚠Compliance` `⚠Engineering` |
| 73 | |
| 74 | ## 🧠 Which prompt should you use? |
| 75 | |
| 76 | Two versions are included: |
| 77 | |
| 78 | | | **Compact** | **Full** | |
| 79 | |---|---|---| |
| 80 | | **Length** | ~400 tokens | ~2500 tokens | |
| 81 | | **Best for** | Personalization / custom instructions / always-on use | New chats / research / fact-checking / tighter labeling | |
| 82 | | **Behavior** | Lighter, faster, good default coverage | More explicit rules, stronger boundary control | |
| 83 | | **Tradeoff** | More drift in long chats, more edge cases left to model judgment | Better consistency, but too long for most personalization settings | |
| 84 | |
| 85 | **Recommendation:** Start with **Compact** for everyday use. Use **Full** when you want tighter boundaries, stronger labeling, and less drift. |
| 86 | |
| 87 | Prompt files: |
| 88 | |
| 89 | - **Compact** → [`model_instructions/prompts-v7-compact.md`](model_instructions/prompts-v7-compact.md) |
| 90 | - **Full** → [`model_instructions/prompts-v7-full.md`](model_instructions/prompts-v7-full.md) |
| 91 | |
| 92 | Suggested setup: |
| 93 | |
| 94 | - **Claude** → Settings → Customize Claude |
| 95 | - |
People who liked this also viewed
6 picksType
- model-alignmentModel Alignment is a python library from the PAIR team that enable users to create model prompts through user feedback instead of manual prompt writing and editing. The technique makes use of constitutional principles to align prompts to users' desired values.PromptsOct 2025
- prompt-decoratorsA standardized framework for enhancing how LLMs process and respond to prompts through composable decorators, featuring an official open standard specification and Python reference implementation. Claude Code plugin and MCP server integration.PromptsJul 2026
- tree-promptTree prompting: easy-to-use scikit-learn interface for improved prompting.PromptsOct 2023
- claudifyAI-powered Spotify playlist generator using Claude API. Create personalized playlists from text prompts or transform your liked songs into curated collections. Seamlessly blends AI creativity with music discovery.PromptsJul 2024
- system-prompts-and-models-of-ai-toolsFULL Augment Code, Claude Code, Cluely, CodeBuddy, Comet, Cursor, Devin AI, Junie, Kiro, Leap.new, Lovable, Manus, NotionAI, Orchids.app, Perplexity, Poke, Qoder, Replit, Same.dev, Trae, Traycer AI, VSCode Agent, Warp.dev, Windsurf, Xcode, Z.ai Code, Dia & v0. (And other Open Sourced) System Prompts, Internal Tools & AI ModelsPromptsJul 2026
- CL4R1T4SLEAKED SYSTEM PROMPTS FOR CHATGPT, CLAUDE, GEMINI, GROK, PERPLEXITY, CURSOR, LOVABLE, REPLIT, AND MORE! - AI SYSTEMS TRANSPARENCY FOR ALL! 👐PromptsJul 2026
