cpldcpu/MisguidedAttention
A collection of prompts that test whether large language models can reason through modified versions of thought experiments and riddles without falling back to familiar but incorrect solutions.
Not currently ranked — collecting fresh signals.
star history
This repository contains prompts designed to evaluate LLM reasoning by presenting modified versions of classic riddles, paradoxes, and thought experiments. The prompts are structured to trigger recognition of familiar problems while requiring different solutions, testing whether models apply logical deduction or fall back to memorized responses. An evaluation framework tracks how different models perform on this benchmark over time, with interactive results available via a GitHub Pages deployment.
Frequently asked
- What is cpldcpu/MisguidedAttention?
- A collection of prompts that test whether large language models can reason through modified versions of thought experiments and riddles without falling back to familiar but incorrect solutions.
- Is MisguidedAttention open source?
- Yes — cpldcpu/MisguidedAttention is open source, released under the CC0-1.0 license.
- What language is MisguidedAttention written in?
- cpldcpu/MisguidedAttention is primarily written in Python.
- How popular is MisguidedAttention?
- cpldcpu/MisguidedAttention has 480 stars on GitHub.
- Where can I find MisguidedAttention?
- cpldcpu/MisguidedAttention is on GitHub at https://github.com/cpldcpu/MisguidedAttention.