Shibo Chen

1.6K posts

Shibo Chen

Shibo Chen

@ShiboChenTech

I build chips. CSE PhD, UMich '25. Opinions are of my own

San Jose, CA Katılım Eylül 2016
1.4K Takip Edilen1.5K Takipçiler
Shibo Chen retweetledi
Kevin Mi
Kevin Mi@kevinmi920·
Introducing Infinite Studio ♾. Last week, @tenstorrent x @prodia announced the fastest Wan 2.2 video generation in the world. We built a demo to show what that speed unlocks: directing an infinite movie in real time. Demo 👇
English
22
45
122
4.8K
Shibo Chen
Shibo Chen@ShiboChenTech·
Nvidia wins because their chips are not only good at doing graphics but ALSO crypto and AI and on and on. Specialized hardware can win a small market but never hits home.
Y Combinator@ycombinator

Inference Chips for Agent Workflows @sdianahu Most AI chips are designed for "prompt in, response out." Agents don't work that way. They loop, branch, and hold context across dozens of steps, and current GPUs hit 30–40% utilization as a result. That gap is where purpose-built silicon wins.

English
0
0
4
383
Atomic Semi
Atomic Semi@atomic_semi·
We are building Atomic Studio to match the speed of our fab. It’s fast, browser-based, and collaborative. Layout, schematic, and simulation. Here’s a side by side of Studio (left) and Klayout (right) moving through the same file.
English
15
21
354
66.4K
Shibo Chen
Shibo Chen@ShiboChenTech·
@ClennonMartin But that's not part of Nvidia's supply and based on my conversation, turbine blades are also not the fastest way you can get large amounts of power on-site and deployed.
English
0
0
0
22
Shibo Chen
Shibo Chen@ShiboChenTech·
Jensen is right and it's so obvious. Conceding controls when you can dominate is some insane stuff.
English
0
0
2
153
Shibo Chen
Shibo Chen@ShiboChenTech·
@IanCutress What is the biggest constraints in the supply chain
English
0
0
3
348
Shibo Chen retweetledi
Tenstorrent
Tenstorrent@tenstorrent·
Run anything, anywhere. On May 1st, see Tenstorrent’s solutions for yourself, deployed at scale. Watch TT-Deploy live at tenstorrent.com/deploy
English
4
21
89
24.9K
Dylan Patel
Dylan Patel@dylan522p·
The Information wrote that Meta consumes 60.2T tokens per month, or ~750M per employee. Meta is getting absolutely token-mogged by us currently. SemiAnalysis employees consume 1.86B tokens per month. From our Tokenomics Model post "Everyone Keeps Estimating Token Prices Wrong"
Dylan Patel tweet media
English
31
18
362
69.5K
Shibo Chen
Shibo Chen@ShiboChenTech·
@neomewtwo @jimkxa No, only the top models are used intensively but the top models change very frequently. One model can linger around for a while but basically become irrelevant after a few months old and newer ones replace them
English
0
0
0
19
Shibo Chen
Shibo Chen@ShiboChenTech·
@neomewtwo @jimkxa But the same model (4o-mini) observed 90% usage swing in a mere two weeks span. I think it also follows power law, only the top few models generating the most tokens, the rest are long tails. @nihalkurth/inside-openais-usage-cliff-009b9a7d61a2" target="_blank" rel="nofollow noopener">medium.com/@nihalkurth/in…
English
1
0
0
50
mewtwo
mewtwo@neomewtwo·
@ShiboChenTech @jimkxa yes a but old models don't lose usage easily 4omini is still in the top 20 used on openrouter
English
1
0
0
45
Shibo Chen
Shibo Chen@ShiboChenTech·
@neomewtwo @jimkxa Given the pace of new models, I don't see any "well trained" model hold up for more than 3 months... Every 3 months, whatever believes people had 3 months ago seem laughable. If we still keep the 3-year hardware deprecation cadence, it seems wrong to overly invest into one model
English
1
0
1
49
mewtwo
mewtwo@neomewtwo·
couldn't one argue that the economic value from a large well trained model is worth the constraints or any setup costs? on the consumer side i can understand this more but if you look at factories they have specifically made robotics or manufacturing equipment and they seems to be fine
English
2
0
7
2.2K
Shibo Chen
Shibo Chen@ShiboChenTech·
I hope I had something like this for my PhD
Amber Liu@JIACHENLIU8

We're living in the BEST era for doing research. 💪 After I graduated from my PhD, the rise of AI-native research gave me a new chance to revisit my research experience. Lately, doing research feels incredibly rewarding to me. I get to experience the pure joy of curiosity-driven science because I no longer have to worry about the lower-level implementations or getting bogged down by infrastructure 🚀 (I'll be sharing some of my own recent research driven by this very soon!) But today, let me introduce the New Orchestra 🎻. We wanted to ship a product that absorbs the friction and brings science back to the curiosity.

English
1
2
11
1.5K
Shibo Chen
Shibo Chen@ShiboChenTech·
You will be surprised how many traders China has. Lots of people, lots of everything.
Eric Jang@ericjang11

I've returned from my month-long China trip. A big thank you to all the Chinese engineers who spent their time meeting with me. I have learned so much. I intend to return soon to learn more. My goal was to empty my cup and look at the AI + robotics ecosystem with fresh eyes. I’ve come away extremely impressed. China feels to me like a nation of engineers, contrasting with America being a nation of traders. Simultaneously, I am concerned that the world is unprepared for what is to come.

English
0
0
1
319

Keşfet

@tenstorrent @prodia @atomic_semi @rhatr @ClennonMartin @IanCutress @dylan522p @neomewtwo