神经符号AI新方法schema在ARC-AGI-3基准上取得进展

精选理由

这个新方法用程序世界模型解决了ARC-AGI-3难题，不用反复试错，比纯神经网络更聪明。

AI 摘要

Gary Marcus转发Wenjie Ma的工作，指出神经符号AI方法schema可能使ARC-AGI-3基准变得可解。schema将经验转化为程序世界模型，并基于完整历史验证，然后搜索解决方案，无需进一步试错。该方法已在交互演示中展示效果，相关分析和演示托管在schema-harness.github.io。

原文 · Gary Marcus

losing track of all the wins for neurosymbolic AI

losing track of all the wins for neurosymbolic AI Wenjie Ma @wenjie_ma Turns out ARC‑AGI‑3 may be saturable after all -- with the right harness. [schema] turns experience into a program world model, verifies it against the full history, then searches it for solutions without further real-world trial and error. See interesting interactive demos + analysis at schema-harness.github.io . 🔗 View Quoted Tweet 💬 2 🔄 1 ❤️ 9 👀 1560 📊 2 ⚡

查看原推