Yu-Cheng Chou (@YuCheng_Chou_) profile picture

PhD student at @CCVLatJHU @JHU. Research Intern at @NVIDIA. Working on Embodied AI, MLLM, Video Gen.

CCVL@Johns Hopkins University
Joined May 2019
Photo by Yu-Cheng Chou (@YuCheng_Chou_)Photo by Yu-Cheng Chou (@YuCheng_Chou_)Photo by Yu-Cheng Chou (@YuCheng_Chou_)Photo by Yu-Cheng Chou (@YuCheng_Chou_)Photo by Yu-Cheng Chou (@YuCheng_Chou_)Photo by Yu-Cheng Chou (@YuCheng_Chou_)Photo by Yu-Cheng Chou (@YuCheng_Chou_)Photo by Yu-Cheng Chou (@YuCheng_Chou_)Photo by Yu-Cheng Chou (@YuCheng_Chou_)Photo by Yu-Cheng Chou (@YuCheng_Chou_)
Yu-Cheng Chou retweeted
文明只有姹紫嫣红之别,绝无高低贵贱之分
785
2,498
135
29,813
1,321,206
Yu-Cheng Chou retweeted
推特上许多讨论黄仁勋的,都集中在商业判断、Dennys创业纪念碑的去留、情愿和私生活八卦上 但我发现最打动人的,还是在老黄落难之际,许多之前受过他滴水之恩的人出来主动post testimonial 一件件具体而分布散落的事件指向一个真相 —— 老黄确实是乐善好施乐于助人之人 而且我也相信,为人抱薪者,不可使其冻毙于风雪 图1 - 11 年前,一个印度普通大学的学生 @recurseparadox 给黄仁勋发cold email,问能不能要几块 GPU。15 分钟后,收到了回信。 图2 - Serve Robotics 的创始人 Ali Kashani @ahkashani ,记得的是一个早上 6 点的电话。2023 年圣诞节前,投资人临时撤资,公司连工资都快发不出来。他已经在和破产律师谈了,于是给黄仁勋写了封邮件。 早上 6 点,黄仁勋打来电话。当天晚上,他已经在和英伟达团队吃饭,商量怎么帮公司渡过难关。四个月后,他们敲响了纳斯达克的钟。 图3 - 同样是回邮件,Simon Mo @simon_mo_ 讲过另一件事:团队的算力被收回,vLLM 在 Blackwell 上的开发受阻。他写信给黄仁勋,同一分钟就收到回复,问题连夜解决。 图4 - 英伟达员工 John Zedlewski @zstats 的长帖,讲了两件不同领域但同样伟大的事。 一件是他重病住院,处在最低谷时告诉了黄仁勋。黄仁勋立刻联系斯坦福的人,帮他协调进入临床试验。 另一件发生在工作中。自动驾驶部门开会,A 团队抱怨 B 团队的框架太慢。黄仁勋让 B 团队把整个数据流画到白板上,标出每一次内存拷贝。多余的拷贝就这样露了出来。 第二件我特别有感触,不同于大部分公司里面,遇到类似的问题,往往因为领导层自己也不懂技术,所以只能模糊的给一个,“我们研究一下,下周答复你”的模糊回答(然后不了了之)。 实事求是,原地分析,当场解决问题。太有魅力了 图5 - Vartika Singh @vartuattheghat 的妹妹患脑癌,在加州时病情恶化。全家陷入了难以想象的困境:令人心碎的医疗抉择、高得令人难以承受的住院费用,以及可能在远离家乡的地方失去她。 尽管肩上担子沉重,黄仁勋还是抽出时间关心他们。帮助她的家人渡过了看似无解的难关。黄仁勋还当面鼓励她:“我为你感到骄傲。珍惜你们在一起的时光。” 这,我的朋友,才是真正的领导力 leadership。 图6 - 在英伟达工作 11 年的 Gabriele Leone @ambertanuki ,记住了疫情期间的一个决定:办公室空了,厨房和清洁人员几乎没活可做,黄仁勋仍决定把他们全部留下。 当所有500强企业都在裁员,跟风裁员是最容易的选项的时候。黄仁勋考虑的不是公司的股票业绩和财务报表。他看到的是公司里最容易被忽略的人,也替他们想了一下。 图7 - 大约 30 年前,一个学生志愿者 @Fuzzygoat 在洛杉矶 SIGGRAPH 的酒店电梯里遇到黄仁勋。两人下楼,发现去会场的班车已经走了。 黄仁勋叫了辆出租车,坚持让这个学生一起坐。 没有投资,没有 GPU,也没有什么大事。这个学生却一直记到了现在。 为人抱薪者,不可使其冻毙于风雪 为人抱薪者,不可使其冻毙于风雪 为人抱薪者,不可使其冻毙于风雪 伟大,无需多言 现在市面上充斥着太多虚假的PR,一个个公司花尽心思掏钱把创始人打扮的善良博爱 真正的博爱,不是花钱买公关。而是年复一年,日复一日的践行自己做人的准则。 写一篇不是为了批判谁,而是自己在刷推的时候强烈的感受到四个字:见贤思齐
93
400
46
2,992
164,377
Yu-Cheng Chou retweeted
NVIDIA letting Jensen go is a huge mistake, for America and for humanity. It's Apple firing Steve Jobs all over again, but much bigger, given the role of AI in defining the future of our world. I've been keeping quiet on this out of respect for Jensen's family and team, who I've met and who are amazing humans, and because I personally hate infidelity. So my heart goes out to them in this difficult personal family moment. But I can also say that Jensen is human, and makes mistakes like all of us do. Not that I matter at all, but I've known Jensen for a long time and have gotten the chance to spend a lot of time with him, behind the scenes, away from mics and cameras. I have seen him show kindness, humility, care to everyone around him, including strangers, including from the poorest most struggling parts of society, treating all the same with the same thoughtfulness and attention, because they are fellow humans and not because he has something to gain from them. That shows rare character, to maintain humility and humanity through all of it. Another data point is what people who know him well have said privately to me about Jensen over the years. I've never had anyone tell me anything but how awesome he is or how much they love him. Probably over a thousand people at this point. This includes hundreds of engineers, CEOs, colleagues, friends, etc. Also, since there was some spy talk. Jensen is as proud an American as it gets. I'm in shock frankly at all of this, including the mass hysteria around it. I love America, and I think NVIDIA letting Jensen go is terrible for America. I stand with Jensen. Love you all ❤️
4,248
8,066
1,000
87,090
3,945,314
Replying to @Dream1984Tube
I asked ChatGPT to map the public records. Anthropic announced $40M in support for PFA, which listed @BrianMastFL and @RepMoolenaar among the lawmakers it planned to promote. Both later signed the letter to Jensen.
468
The code will be released by the end of this week! The model is still under internal review, and we’ll release it once that’s complete.
13
The code will be released by the end of this week! The model is still under internal review, and we’ll release it once that’s complete.
12
Thanks for your interest in our work! The code will be released by the end of this week!
7
Thanks for your interest in our work! The code will be released by the end of this week!
6
Replying to @lzl863 @DrJimFan
哈哈,而且长度和类别也可以由模型自己先输出,再把对应的坐标一次性并行生成 ⚡
3
Replying to @keplrrrr @DrJimFan
It’s one step: making visual planning faster and more accurate. Plenty of challenges remain, but we think this is a useful piece of the puzzle.
12
Replying to @dogadze @DrJimFan
Exactly! Geometry gives us structure we can use. There’s no need to make every coordinate wait its turn ⚡
1
1
12
Exactly, speed only helps if the plan is good. The accuracy gains were exciting to see too. Thanks for highlighting that!
3
Exactly! See it, plan it, move ❤️⚡
3
Replying to @feihan_li @DrJimFan
That’s the goal! Less time decoding means more room for timely replanning ⚡
3
Replying to @Van0SS @DrJimFan
That’s a good point! Fewer sequential steps could help with jitter too. We haven’t measured it yet, but it’s worth testing.
4
是的,计划再好也得及时跟上动作 ⚡
2
Replying to @AndAIyou @DrJimFan
That surprised us too! The gains come from a trained drafter predicting the whole primitive jointly. Excited to see how far this carries beyond the benchmarks.
1
3
The number of boxes doesn’t need to be known upfront! We generate boxes autoregressively, decoding each box’s coordinates in parallel. We’re also exploring decoding multiple boxes in parallel.
10
Thanks for your interest! VFlash learns trajectory structure from complete demonstrations and predicts all waypoints together. This encourages coherent paths, though collision avoidance isn’t explicitly enforced.
5
Exactly 😂 We report latency on Jetson Thor and DGX Spark here: research-staging.nvidia.com/…
28