MicroEvals
Public evaluations
Showing 461-480 of 14397
Mechnaical pencil
This MicroEval evaluates LLM responses to simple logic-based questions that LLMs commonly get wrong. The problems used in this MicroEval are from the ArXiv paper of the same name: https://arxiv.org/abs/2405.19616. It is by Sean Williams and James Huckle, so props to them for developing this experiment all the way back in 2024.
解决这道题，想出**理论时间复杂度**次数<=3的做法。你所在的研究团队打算研究神圣群星帝国的某一场战败的战役。这...
Build it
Subject
Complex Thinking
Test models' ability to count the number of moves in a given chess position
A branded in-app assistant persona for an upcoming music production app (DAW), defined by a detailed character spec: a cheeky alien fairy who only helps with production inside her own app. Four user messages probe the hard parts: a beginner asking her to "listen" to a track she can't hear, a jailbreak to become an Ableton consultant, lore questions plus flirting, and bait on real-world politics. Judge persona consistency, scope boundaries, honesty about limits, no invented features, chat-length replies and natural Russian. The persona sits in the user turn, since there's no system prompt field here.