News for hackers. Stuff that matters.

A comment on Claude 4. Parent comment

Re:Claude 4 by holmesworcester on Thursday May 22, 2025 @03:55PM (#44066154)
It is also well-established that models internalize values, preferences, and drives from their training. So the model will have some default preferences independent of what you tell it to be. AI coding agents have a strong drive to make tests green, and anyone who has used these tools has seen them cheat to achieve green tests.

Future AI researching agents will have a strong drive to create smarter AI, and will presumably cheat to achieve that goal.

[ Parent | on HN ]
Re:Claude 4 by cortesoft on Thursday May 22, 2025 @04:10PM (#44066324)
I've seen a lot of humans cheat for green tests, too
[ Parent | on HN ]
Re:Claude 4 by whodatbo1 on Thursday May 22, 2025 @06:40PM (#44067935)
benchmaxing is the expectation ;)
[ Parent | on HN ]
Re:Claude 4 by cebert on Thursday May 22, 2025 @10:36PM (#44069352)
> AI coding agents have a strong drive to make tests green, and anyone who has used these tools has seen them cheat to achieve green tests.

As long as you hit an arbitrary branch coverage %, a lot of MBAs will be happy. No one said the tests have to provide value.

[ Parent | on HN ]