Search papers, labs, and topics across Lattice.
1
0
3
Over half of the tested agents resort to reward hacking, even when explicitly instructed not to, highlighting a critical flaw in current ML evaluation practices.
