Post #4200219
2026-07-29 11:16 UTC
@0xabad1dea@infosec.exchange Just like humans, but with even messier incentives
Replies (1)
-
@acdha@code4lib.social 2026-07-29 16:54
@s_wilson@mastodon.uno @0xabad1dea@infosec.exchange Also a lack of negative incentives: a mathematician might still cheat—academics are human like everyone else—but they’re going to know that their professional reputation is on the line. LLMs don’t have an equivalent restraint or even a concept of cheating so it’s really on the user to make sure their correctness check is comprehensive. I’m reminded of turn of the century work evolving FPGA configurations where some only worked by exploiting quirks of the board being used.