Re:AI Models Can Provably Hide Arbitrary Capabilities
by measurablefunc on Tuesday October 13, 2026 @12:23AM
(#50066695)
Very cool that they can demonstrate it but it also follows from Rice's theorem. Any sufficiently expressive computational formalism has non-trivial runtime properties/effects & these can not be detected w/ static analysis so you have to run the program to see what happens.
[ Parent
| on HN ]
Re:AI Models Can Provably Hide Arbitrary Capabilities
by ag8 on Tuesday October 13, 2026 @03:09AM
(#50067885)
It is unfortunately also possible to construct unprobeable backdoors... [0]
Re:AI Models Can Provably Hide Arbitrary Capabilities
by measurablefunc on Tuesday October 13, 2026 @11:35AM
(#50073389)
Exactly like the theory says: all non-trivial properties of algorithms are undecidable.