Neurosymbolic Reasoning for the ARC Benchmark

  • Subject:Neurosymbolic Reasoning, Large Language Models, Vision
  • Type:Bachelor or Master Thesis
  • Supervisor:

    Lukas Kinder

  • Add on:

    The ARC benchmark requires understanding and reasoning about pixel-based images, where current AI models struggle to match human performance. To raise awareness of this challenge, the ARC organizers are offering $600,000 to any model that achieves 85% accuracy. You can learn more about the competition at https://arcprize.org/. In this thesis, you will explore how to leverage the reasoning capabilities of Large Language Models (LLMs) to perform on the ARC benchmark. This project can be done as either a Bachelor or Master Thesis. If interested, please email me at lukas.kinder@kit.edu.