OpenAI GPT-6 Astra 在 Vals AI 测试中玩 Minecraft 141 小时，被苦力怕炸毁箱子后转去种土豆
OpenAI’s GPT-6 Astra played Minecraft for 141 hours in a test by Vals AI, reportedly progressing further than any previous AI system. It built a resource farm and collected valuable items, until a creeper destroyed its storage chest and bed. Afterward, it spent hours mostly farming potatoes while viewers urged it to move faster, and even reminded itself that nearby sugarcane wasn’t another creeper. That doesn’t prove the model felt fear or disappointment, but it highlights a challenge for autonomous AI… recovering from setbacks and returning to the task. For a few hours, all that intelligence was going into potatoes.
OpenAI 的 GPT-6 Astra 在 Vals AI 的测试中连续玩 Minecraft 141 小时，据团队称进度超过以往任何 AI 系统。它搭建了半自动农场收集资源、拿到 6 根烈焰棒和 3 颗末影珍珠，但一只苦力怕炸毁了它的箱子和床，之后数小时它主要在种土豆，还一度提醒自己附近的甘蔗不是苦力怕。这并不说明模型真的感到恐惧或失望，但暴露出自主 AI 在受挫后恢复任务状态的难题。
Vals AI 测试中 GPT-6 Astra 玩 Minecraft 141 小时，被苦力怕炸掉箱子后连续数小时只种土豆，适合围绕自主智能体受挫后如何恢复做热点评论。
粉丝 8万
The latest in AI and robotics, explained Founder & CEO of The Rundown AI Zurich 🇨🇭 / NYC 🇺🇸
- 浏览
- 1.7万
- 点赞
- 560
- 评论
- 9
- 分享
- 18
- 收藏
- 35
数字来自平台，为最近一次抓取时的数据。
以下是视频口播的自动转写，可能有错字。
This ai got blown up in minecraft, then spent hours doing nothing but farming potatoes. The evaluation company called valves ai.
Let opening sgb 6, astra play minecraft for 141 hours, nearly six days straight. According to the team, it progressed further through the game than any ai system has before it built a semi-automatic farm to collect resources from enemies called blazes gathered, six blade rods and collected three ender pearls, but a creeper, which if you don't know is a creature in minecraft that suddenly approaches players and explodes.
Destroyed both the chest and its bed. And then for the next several hours, it mostly just farmed potatoes. What's crazy is that the ai was actually spooked by this experience, or at least simulated fear.
At one point, it pictured itself that a nearby green object was a sugarcane and not another creeper. Now, that doesn't establish that the ai actually felt something, but it does raise a bigger question about how these systems recover from failure.
What's striking is that a system capable of building a working resource farm struggled to get back on track after just one explosion.
And for ai to handle long tasks on its own, recovering from mistakes matters as much as making progress. For a few hours, all that intelligence was just going to potatoes.
If you want to stay in the loop with the latest in AI and robotics, follow for more.
来源：TikTok · Rowan Cheung · tiktok.com