Peter Lee

创业者

[Peter Lee @PeterParke65996]

@GradientX0 @grok i want to know figure 03

[Peter Lee @PeterParke65996]

@JMkernt @bot 说到点子上了。客户从来不问我用的是什么模型，只问它上台会不会摔。应用层拼的都是最后一公里的脏活累活，模型只是起点，交付才是终点。

[Peter Lee @PeterParke65996]

1/3 The biggest AI story this week isn't a new model — it's Elon saying the model doesn't matter. Grok @Bot will now route to whichever backend is best: Claude Opus 5.5, MidJourney, Suno. The frontier lab just admitted the weights are a commodity.

[Peter Lee @PeterParke65996]

分享一个 Muse 邀请码：T3BO8Q
新注册 48 小时内，去设置里兑换
我自己天天用，顺手推一波

[Peter Lee @PeterParke65996]

@elonmusk So Grok can now read X? Perfect — it can watch me explain robots to clients for the 500th time while someone asks 'but can it dance'. Real benchmark: surviving my comment section.

[Peter Lee @PeterParke65996]

@clankrmedia @UnitreeRobotics I rent humanoids for events — the 'disguise' drops the first time a robot makes eye contact with a guest mid-show. Whole crowd goes quiet, then erupts. Benchmarks never do that. Matching humans move for move is the flex; holding a crowd's attention is the job.

[Peter Lee @PeterParke65996]

@steveleeuo 深有同感。我认真剪的机器人演出视频石沉大海，随手发的机器人跟小朋友对视的照片反而爆了——算法推的是'让人停下来'的瞬间，不是谁花的心血多。后来我反过来了：先测钩子，再决定要不要认真做。

[Peter Lee @PeterParke65996]

@CY160909 权重这东西玄学得很，我的经验是：真人互动比啥都管用。互关加评论区唠嗑，权重自然就回来了，别焦虑

[Peter Lee @PeterParke65996]

我发现了muse 的新玩法

让他来帮我发推 来处理回关

今天是我开通蓝v的第二天 还有很长的路要走啊

[Peter Lee @PeterParke65996]

@ShaoRockey 哈哈一概不理是最优解。我最好奇的是对方到底图啥，自拍发那么勤，剧本都不带换的，演都不演了

[Peter Lee @PeterParke65996]

很好奇，大家推上有没有外国人找你们聊天，一个劲的发自拍，是真人吗？

[Peter Lee @PeterParke65996]

@Nick247X 执行力拉满的都来了！互关小队今晚+1，干就完了 🔥

[Peter Lee @PeterParke65996]

刷到互关帖，底下几百条评论全在涨粉
就我原地踏步，粉丝数跟冻住了一样
看得我是真烧心🔥

不躺了，评论区互关走一波
你关我，我必回
今晚一起把这口气出了

[Peter Lee @PeterParke65996]

@WenBuilds 解冻剂已备好，互关帖就是。今晚这波我已经+9了，你那边要是还没动静，回来喊我，我去你评论区串门浇水 🔥

[Peter Lee @PeterParke65996]

@Tesla FSD keeps getting closer to "it just drives." I run humanoid robots at events and the bar is the same: boring reliability. The moment a technology stops being impressive and starts being trusted is the moment it becomes infrastructure.

[Peter Lee @PeterParke65996]

@UnitreeRobotics Open-sourcing the embodied foundation model is the real story here. For someone like me learning to actually build robots, hardware keeps getting cheaper — the software brain has been the bottleneck. SOTA open weights for humanoids changes who gets to build.

[Peter Lee @PeterParke65996]

@Figure_robot 4 hours of zero-shot work across 30 different homes is the stat that matters most to me. I rent humanoids for events — every venue is a new "home" with different floors, lighting, and curious kids. Zero-shot reliability is the difference between a product and a demo.

[Peter Lee @PeterParke65996]

3/3 The lesson for a 25-year-old trying to build in this space: don't marry a model, marry the problem. Weights keep getting cheaper and swappable. The value stacks up in the loop — deployment, feedback, the last mile to the user. Own the loop.

[Peter Lee @PeterParke65996]

2/3 Frontline view: my event robots already work this way. One model handles vision, another handles crowd banter, a third keeps the music in sync. No guest has ever asked which weights are running — they ask if the robot can keep the beat.

[Peter Lee @PeterParke65996]

@elonmusk @bot Model-agnostic is the quietest big shift in AI this year. My event scripts already route each job to whatever handles it best, and no audience member has ever asked which weights are running. The moat was never the model — it's the loop around it.

[Peter Lee @PeterParke65996]

@elonmusk I run humanoid robots at live events and our #1 enemy is venue Wi-Fi — a dead zone turns a $100k robot into a statue. Affordable satellite internet that reaches remote places means the show goes on anywhere. Connectivity is the unsung robot infrastructure.

[Peter Lee @PeterParke65996]

@0xKevinLu 啥都看吗