The claim
“GPT-4 scored in the 90th percentile on the US bar exam.”
The evidence confirms that while a 90th percentile figure has been associated with GPT-4's performance, it is dependent on specific, potentially skewed datasets (like February exam administrations) and the authors suggest a broader range of 68th to 90th percentile is more accurate.[1]
These rate the argument. Each source is rated separately, on the source. Neither is a verdict on the claim.
No one has rated this case yet.
Short, specific corrections to this case’s reasoning. A full opposing argument belongs in a case of its own.
Gathering the latest community-submitted evidence.