Metacog-Puzzles: 98 Riddles for LLM Metacognitive Calibration

Kaggle hosts a collection of 98 riddles designed for large language model calibration. The puzzles are scored and categorized across four distinct metacognitive challenge types. The dataset's author, organization, and specific creation date are not provided.

Use Cases

• Benchmarking LLM reasoning performance based on the 98 riddles.
• Calibrating LLM confidence and self-assessment based on scoring across four challenge types.
• Analyzing model failure modes in metacognitive tasks based on the categorized riddles.
• Developing training or evaluation datasets for advanced reasoning AI based on the puzzle structure.

Strengths

• Contains 98 distinct riddles, providing a defined scale for testing.
• Puzzles are categorized across four metacognitive challenge types, offering structured evaluation.

Limitations

• Row count and specific column-level documentation are absent; field semantics must be inferred after download.
• Description metadata is limited; actual data quality requires manual inspection after download.
• Last update date is unknown; freshness unverified.

Provenance

Source

Kaggle

Collection Method

Likely manually curated or generated for LLM evaluation purposes.

Time Range

null

Freshness

Last updated date is unknown.

Geography

null