7 results on this page
Researchers introduce an open-source benchmark separating olympiad geometry solving from accurate diagram construction. Across 954 problems, current foundation models achieved only a 36.14% average diagram compilation success rate, revealing a substantial gap between mathematical reasoning and faithful visual construction.
The Spectral Neuron is a proposed ML primitive based on the largest eigenvalue of an input-conditioned matrix. The preprint develops its mathematical properties, initialization and training methods, and evaluates scalability on synthetic and real-world datasets.
Google DeepMind researchers found that training AI agents through debate can reduce reward hacking when an LLM judge evaluates their work. On mathematics tasks, debate achieved higher sustained ground-truth accuracy than directly optimizing for LLM judge rewards.
UC Berkeley mathematics professor Zvezdelina Stankova wrote an op-ed criticizing students who cannot perform middle school mathematics. The discussion shifted after an AI detector reportedly flagged the human-written article, highlighting concerns about detector reliability.
數學家袁新意表示，AI 近半年在整合既有數學思想、構造例子與反例方面進步迅速，但仍未產生足以讓職業數學家認可的重要原創成果。文章也指出，中國數學人才培養已出現質變，但青年教師承受過重的量化科研考核壓力。
A creator built a working AI chatbot in vanilla Minecraft using 445,782 command blocks, without mods, plugins, or datapacks. A revised approach reduced the implementation from more than one million blocks by working around command blocks' limited mathematical operations.
UC Berkeley mathematics professor Zvezdelina Stankova acknowledged using AI to help edit an op-ed criticizing students’ mathematics preparation. The disclosure raises questions about transparency, authorship, and acceptable AI use in academic and public writing.