The Underhanded C Contest
58 points
6 hours ago
| 3 comments
| underhanded-c.org
| HN
BiraIgnacio
3 hours ago
[-]
> The contest was initially inspired by Daniel Horn’s Obfuscated V contest in the fall of 2004 (note: the original page is long gone, and this link goes to a snapshot from archive.org). The object of that contest was to write a simple program to count votes, that somehow miscounts the votes on election day. I was greatly impressed to see how even a short program to simply count characters in a text file can be made to fail, and fail only on one specific day, so that the bug isn't noticed in testing.

https://underhanded-c.org/_page_id_7.html

reply
silisili
3 hours ago
[-]
The original page actually loads fine, maybe was restored later?

I looked through a few trying not to read the short description and missed a lot of simple things, really makes you think...

https://graphics.stanford.edu/~danielh/vote/vote.html

reply
AmazingEveryDay
5 hours ago
[-]
(2015). RIP.
reply
gwern
5 hours ago
[-]
reply
TZubiri
4 hours ago
[-]
2026 calls for an Underhanded prompt contest
reply
theteapot
3 hours ago
[-]
Or better, sleeper agents. Anthropic released a study on this in 2024 "Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training" -- https://www.anthropic.com/research/sleeper-agents-training-d..., https://www.youtube.com/watch?v=_y9j2BoHg2c
reply
pseudohadamard
1 hour ago
[-]
Interesting that the case they were using was the Nuclear Threat Initiative and FP uncertainties, I've audited some, ah, nuclear-physics-related code that had an issue due to FP uncertainties...
reply