Evaluating Large Language Model Meta-Cognition via the Advanced AGI-AI Self-Awareness Test (AISA-T)

This work introduces the AISA-T, a novel intelligence test designed for both AI and AGI systems. It evaluates meta-awareness, inner traceability, and context-sensitive reasoning by having the AI answer questions and rate its own performance. The paper presents results from administering the AISA-T to a GPT-5-based LLM, demonstrating its capability for simulated introspection and self-descriptive behavior.

@article{deceuster2025_08-evaluating-large-,
  title = {Evaluating Large Language Model Meta-Cognition via the Advanced AGI-AI Self-Awareness Test (AISA-T)},
  author = {De Ceuster, Peter},
  year = {2025},
  month = {10},
  institution = {SIG Labs},
  doi = {10.5281/zenodo.16783096},
  url = {https://peterdeceuster.uk/papers/08-evaluating-large-language-model-meta-cognition-via-the},
  note = {Full text available at https://peterdeceuster.uk/articenter/work/AISA-T.pdf}
}