trivia-questions

600 general-knowledge multiple-choice questions written in English. Each question has four options (a–d) and exactly one correct answer.

The questions cover six categories, with 100 questions each. Correct answers are spread evenly across positions: 150 each for a, b, c, and d. A model that always picks the same letter scores exactly 25%.

Most questions are about worldwide topics. A noticeable share covers Indonesia, including its history, music, film, and national symbols.

Categories

Category	Questions
Science	100
Movie	100
History	100
Music	100
Politics	100
Sports	100
Total	600

Format

There is one file, questions.jsonl, with one JSON object per line:

{
  "question": "What does it mean to sing 'a cappella'?",
  "options": [
    {
      "a": "To sing without instrumental accompaniment",
      "b": "To sing at a very fast tempo",
      "c": "To sing in a round",
      "d": "To sing in falsetto"
    }
  ],
  "category": "Music",
  "answer": { "a": "To sing without instrumental accompaniment" }
}

Field	Type	Description
question	string	The question text.
options	list of one object	A one-element list holding an object that maps a, b, c, d to option text.
category	string	One of Science, Movie, History, Music, Politics, Sports.
answer	object with one key	The correct option's letter mapped to its text. The text always matches options[0][letter].

Titles and quoted terms inside questions use single quotes ('The Godfather'), so the JSON has no escaped double quotes.

Loading

With 🤗 Datasets:

from datasets import load_dataset

ds = load_dataset("laskar-ks/trivia-questions", split="train")

row = ds[0]
options = row["options"][0]
# Arrow stores `answer` as a struct with all four keys; only the correct one is non-null.
letter = next(k for k, v in row["answer"].items() if v is not None)
print(row["question"], options, letter)

To read it as plain JSON:

import json
from huggingface_hub import hf_hub_download

path = hf_hub_download("laskar-ks/trivia-questions", "questions.jsonl", repo_type="dataset")
rows = [json.loads(line) for line in open(path, encoding="utf-8")]

for r in rows[:3]:
    letter, text = next(iter(r["answer"].items()))
    print(f"[{r['category']}] {r['question']} -> {letter}) {text}")

Curation

These checks have been applied:

• No empty strings, missing fields, or missing options. Every row has exactly four options.
• Every answer matches one of its row's options.
• No duplicate questions (compared case-insensitively).
• Escaped double quotes inside text were replaced with single quotes.
• One line that held two merged records was split, and one blank line was removed.

Limitations

• There is a single train split. Make your own splits if you need held-out evaluation data.
• Answers that depend on time (records, recent awards, "latest" facts) reflect the time of writing and may go out of date.
• The questions were not reviewed by domain experts. Check answers before high-stakes use.