Unspecialized humans score 83.7%
In the SimpleBench test requiring high school knowledge, humans scored 83.7% while the top model Claude Fable achieved 81.9%.
- Field
- training language models
- What they did
- Researchers created the SimpleBench test, consisting of over 200 questions, where individuals without specialized training (high school students) score 83.7%, while top language models, including Claude Fable, score 81.9%.
- Why it matters
- This allows assessing how much modern models lag behind humans in tasks requiring spatial-temporal reasoning, social intuition, and protection from deception, unlike most other tests where models outperform humans.
#llm#simplebench#claude#reasoning



