看了这条你会明白，Astra数学再强也不是AGI的充分证据，Gary Marcus一句话点破通用性该有的样子。
Gary Marcus发帖质疑，把AGI限定在可形式化领域是移动球门柱。他以2024年与Brundage的赌约为据，认为写视频脚本这类任务AGI应当能做。他承认Astra在数学上表现惊艳，但仅凭数学能力不足以证明其达到AGI。
Is it moving the goalposts to ask that Artificial General Intelligence be “general”? And writing g...
Is it moving the goalposts to ask that Artificial General Intelligence be “general”? And writing good video scripts is surely something AGI ought to be able to do. (My 2024 bet with Brundage gave a similar example; it’s not new.) I was kidding about the “wake me” part (I have already said Astra is impressive) but you can’t really call something AGI if it works only in domains that can be formalized; that’s just not how the term was ever defined or what it means. Astra is super impressive in math; that alone (which is all we know for now) doesn’t necessarily mean it would qualify as AGI. Auguste Prompt @augusteprompt there's moving the goalposts and then there's this. 🔗 View Quoted Tweet 💬 1 🔄 1 ❤️ 5 👀 1180 📊 2 ⚡