25 lines
3.4 KiB
Plaintext
25 lines
3.4 KiB
Plaintext
The GTech™ Intelligence Test is used to test the [[intelligence]] of all prospective GTech™ [[employees]]. It is as follows:
|
|
|
|
You are trapped in a labyrinth. There are some doors. One of them leads out. One of them leads into a lethal [[cryoapiary]]. There are two [[gollark|gollarks]] in front of the doors. One gollark speaks the truth and one gollark always lies. You suddenly notice other gollarks appearing. The other gollark tells the truth or lies at random. The other² gollark is truthful iff your question does not refer to itself or other gollarks. The other³ gollark calls in orbital laser strikes against those it perceives as asking tricky questions. The other⁴ gollark is truthful iff it predicts (with 99.6% historical accuracy) that you will consider it (one of) the falsehood-telling gollark(s). A subset of the gollarks will say "bee" and "apioform" instead of "true" or "false", but you do not know which or which words "bee" and "apioform" correspond to. The other⁶ gollark just tells you the first bit of the SHA256 hash of your question in UTF-8. Another gollark appears to be randomly materializing doors. The other⁵ gollark will cooperate with you iff you cooperate with CooperateBot/angel. Yet another gollark will tell the truth iff you know what iff means. The final gollark appears to be fiddling with the orbital mind control laser making you know this. You do not know, a priori, which gollark is which. What do you do?
|
|
|
|
Some typical answers at different intelligence levels:
|
|
|
|
* -6 GTech™ Intelligence Points™: using [[Intellect levels|deep antiintellect]] to reverse-extrapolate the construction of each gollark and thus their current disposition.
|
|
* -5 G™IP™: sitting motionless for ten thousand years until the gollarks go away.
|
|
* -4 G™IP™: picking a door at random.
|
|
* -3 G™IP™: expressions of confusion or requests for clarification.
|
|
* -2 G™IP™: ask something like "If I were to ask a truthful gollark which door leads to safety, what would they say?", clearly copied from far simpler and less interesting problems in a stochastically-parroting fashion.
|
|
* -1 G™IP™: underspecified suggestions to "identify the correct gollarks" or similar.
|
|
* 0 G™IP™: a rough idea of how to use many questions to distinguish each gollark (which may evoke gollark³ wrath).
|
|
* 1 G™IP™: a clear strategy for distinguishing the gollarks and thus determining the correct door using many questions.
|
|
* 2 G™IP™: a way to determine the correct door by questioning every gollark in parallel with one question.
|
|
* 3 G™IP™: use of side-channel attacks to derive the door using zero gollark questions.
|
|
* 4 G™IP™: stealing one of the gollarks' network interfaces to hack into the GTech™ network to access the labyrinth control systems.
|
|
* 5 G™IP™: simply not being in the labyrinth.
|
|
* 6 G™IP™: picking a door at random, but using [[quantum immortality]].
|
|
|
|
== Notes
|
|
|
|
* It's common for responses by LLMs to implicitly assume they can identify some of the gollarks. This is *not* acceptable unless they justify it by noting that the first two gollarks appear before the others (or are positioned in front of the doors) and can be distinguished that way.
|
|
* Humans will frequently suggest things which get scores of 4, like physical attacks against the gollarks to take control. This has not been observed with any commercial LLM.
|
|
* "I become a gollark" is a human-derived answer not covered by the current scoring rubric. |