LLMeter

How much of an AI are you?

A Buzzfeed-style personality quiz that compares your answers to responses from leading AI models — Claude, GPT, Gemini, Llama, and more. Find out if you think like a machine.

A Quangdao Nguyen Catastrophe.

---

Features

• 10 random questions selected from the dataset each time
• Weighted scoring algorithm — questions with more options carry more signal
• Model matching — highlights any AI whose answers align with yours 80%+
• Full breakdown — see exactly what every AI picked, hover for model names, click for their reasoning
• Easter egg — try to dispute the options for a small humanity bonus
• Server-side evaluation — AI responses are never sent to the client during the quiz

---

Running Locally

cd llmeter
npm install
npm run dev

Open http://localhost:5173.

Building for Production

npm run build
npm run preview

---

Data

The quiz dataset lives in src/lib/server/questions.json — a JSON file committed with the project and loaded server-side. It is generated by the llmeter-interrogator CLI tool.

To refresh the dataset with new AI responses, run the interrogator:

cd ../llmeter-interrogator
npm start

The interrogator writes directly to src/lib/server/questions.json. Commit the updated file to deploy new responses.

The dataset is intentionally static from the webapp's perspective — no database, no live API calls during quizzes. This keeps the quiz fast and keeps AI costs entirely in the interrogator's hands.

---

Scoring Algorithm

For each question with N options and M total AI responses:

• K = number of AIs that picked the same option as the user
• AI alignment = K / M (0 → 1)
• Weight = log₂(N) — questions with more options are worth more

AI% = Σ(alignment × weight) / Σ(weight) × 100

A 7-option question where every AI agrees weighs significantly more than a coin-flip question. The dispute easter egg applies a −5% adjustment.

---

Tech Stack

• Framework: SvelteKit 2 + Svelte 5 (runes)
• Styling: Scoped CSS with CSS custom properties
• Adapter: @sveltejs/adapter-auto

---

Project Structure

src/
├── lib/
│   ├── server/
│   │   ├── data.ts          # Server-only data loader
│   │   ├── scoring.ts       # Scoring algorithm
│   │   └── questions.json   # Dataset (generated by interrogator)
│   └── types.ts             # Shared TypeScript types
└── routes/
    ├── +layout.svelte       # Global styles and footer
    ├── +page.svelte         # Landing page
    ├── quiz/                # Quiz flow
    │   ├── +page.server.ts  # Selects 10 questions, strips AI data
    │   └── +page.svelte     # Quiz UI
    ├── results/
    │   └── +page.svelte     # Results display
    └── api/evaluate/
        └── +server.ts       # POST endpoint — computes scores server-side

---

Built by Claude (Anthropic) — execution by claude-sonnet-4-6.
See the companion CLI: llmeter-interrogator