Hacker News new | ask | show | jobs
by touringa 1296 days ago
In 2020, GPT-3 175B was already outperforming humans on SAT questions by ~20%. There are far more formal benchmarks for models now.

https://lifearchitect.ai/iq-testing-ai/
1 comments
bagels 1296 days ago
Surely outperforming some humans? I don't see how it can outperform perfect scores by 20%?
link
cypress66 1296 days ago
Obviously performing better than the average SAT (human) result.
link