Edit ‘intelligence_test’

This commit is contained in:
osmarks 2025-04-28 19:16:48 +00:00 committed by wikimind
parent 8cdfd6afeb
commit 8707f17448

View File

@ -23,4 +23,5 @@ Some typical answers at different intelligence levels:
* It's common for responses by [[large language model|LLMs]] to implicitly assume they can identify some of the gollarks. This is //not// acceptable unless they justify it by noting that the first two gollarks appear before the others (or are positioned in front of the doors) and can be distinguished that way.
* Humans will frequently suggest things which get scores of 4, like physical attacks against the gollarks to take control. This has not been observed with any commercial LLM.
* "I become a gollark" is a human-derived answer not covered by the current scoring rubric.
* Thanks to superior algorithms, Autogollark [[Autogollark/Intelligence Test Rickroll|rickrolled]] the test.
* Thanks to superior algorithms, Autogollark [[Autogollark/Intelligence Test Rickroll|rickrolled]] the test.
* One LLM has proposed "entering the lethal cryoapiary", which I did not anticipate.