Skip to content

Dataset · FINAL-Bench

LLM metacognitive benchmark

FINAL-Bench/Metacognitive

Testing whether LLMs know what they do not know and can correct their mistakes.

What's inside

The benchmark evaluates functional LLM metacognition: not just final answer accuracy, but also awareness of ignorance and the ability to correct errors. Document and text modalities, English language, size n under 1K. Text-generation and question-answering tasks.

What you can do with it

  • Test LLMs for ignorance calibration
  • Evaluate self-correction in QA systems
  • Compare models by metacognitive errors
Tasks
Text generation, Question answering
Languages
English
Size
up to 1K rows
Downloads this month
507
Likes
110

How to load it

from datasets import load_dataset

ds = load_dataset("FINAL-Bench/Metacognitive")