News for hackers. Stuff that matters.
Re:GPT-4 by nsxwolf on Tuesday March 14, 2023 @02:52PM (#35156643)
Passing the LSAT is a fairly good indicator that a human can be a lawyer. It's not yet a good indicator that a large language model can be a lawyer.
Re:GPT-4 by dash2 on Tuesday March 14, 2023 @03:18PM (#35156969)
This is an underrated comment. Compare: playing top level chess is a good indicator that someone will be good at maths problems, but not that a chess computer will be.
Re:GPT-4 by falcor84 on Tuesday March 14, 2023 @03:05PM (#35156825)
So what is missing? Could we add up those missing competencies to create a new test of "general knowledge-worker capabilities" and then train an AI to pass this test?