James Fang (@polymath_james) header
James Fang (@polymath_james) profile picture

22 | Agent Infra & Systems @architectlabs | UIUC CS '23, Master's '25 Views are of my own.

Joined January 2024
Photo by James Fang (@polymath_james)Photo by James Fang (@polymath_james)Photo by James Fang (@polymath_james)Photo by James Fang (@polymath_james)Photo by James Fang (@polymath_james)Photo by James Fang (@polymath_james)Photo by James Fang (@polymath_james)
James Fang retweeted
57% of voters think AI will hurt society. I think 57% of voters have never had a doctor say "stage 4 cancer" to their face. I have. - AI is what has allowed me to design a personalized cancer vaccine from my own tumor's DNA. - AI is what lets me understand everything about my health and show up to appointments prepared. - AI catches the stuff that slips through, like drug interactions, lab trends and the questions nobody thought to ask. And this is just basic shit my dumb ass is doing on my own. We've only just begun to see its true potential. Be scared of AI all you want. Just know the price of slowing it down gets paid by sick people. May not be you or your loved ones today, but it will be in due time. When you're in extreme agony from a terminal diagnosis with death lingering, I sure as shit bet you'll be wishing that AI had developed a treatment so you could spend just a bit more time with your friends and family.
NBC News Exit Poll asked them about artificial intelligence. 57% said AI will do more to hurt society than help it. 76% said they oppose building a data center in their own community. Ouch.
157
321
43
2,348
80,060
Check it out, Pranav’s cooking cool stuff :)
Today we’re introducing Aether @aetherruntime. The fastest cloud coding agents in the world that you can use with your existing subscription Give it a task and just 20 seconds later your build is running on its own cloud machine. You can pick up from your phone, watch every line, and step in whenever you want as if it was local. Close the laptop. It keeps going. It runs Claude Code or Codex on the Claude or ChatGPT plan you already pay for so there’s not a 2nd AI bill. You only pay for the machine. When it’s done, if code review is enabled, a reviewer agent runs the code with the changes, which catches classes of bugs that can't be found from static analysis. Before building Aether I built infra at Apple, Uber, and HeyGen. I built Aether because the cloud agents I tried before all had drawbacks, and I couldn't imagine building something without it now. It’s free to start. runaether.dev
1
2
175
The Spanish HSR accident in Toledo is tragic and I grief for the families for the victims. Sadly, accidents are not accidents. Accidents are a product of how a government treats accountability:
56
The point open source libraries is to abstract away design decisions so you don’t have to make them yourself again. Even if many functions seem easy to implement on inspection, it saves you the energy of rethinking a tool’s architecture.
65
Canada has plenty of advantages. It is a place where people want to live, especially before the pandemic. I lived in Canada from 2011-19. With the right policies, it can and will prosper again. Brain drain can be framed as a major disadvantage. But it also brings diffusion of knowledge, capital, and skills right across the border, from Canadians returning home to start businesses. If Canada gets economic policy right again, it will continue benefitting enormously from trade with USA.
canada has problems, but the US is an outlier, not the baseline. having US next door is a massive advantage and disadvantage. like roughly 1 in 4 STEM grads from waterloo, ut & ubc move to the US. plenty of reasons we already know of. it is not that we are “lacking ambition”, some of that ambition has a US address now. i also find the canada vs US scoreboard unhealthy. canada helps fund the education, the US recruits the talent. we both benefit from this connection, and are not two separate teams. we can definitely ask why people leave and what would make them stay. but i’m long on this relationship. making it work better for both countries will do more for canada than constantly asking why we aren’t america.
1
187
Picked up AoE2 this year, this thing was released like over 25 years ago My goodness this is an incredible game, easily the most difficult I’ve played though and takes so much time
Anyone who gamed between 1995-2010 What's the one game you still think about?
1
107
I’m actually very happy and jealous how much progress biotech and compbio has made in the last 4-5 years. This field has actually exploded. If I entered UIUC as an undergrad in 2025-26, I absolutely would have done the CS+BioE major instead of Engineering CS degree.
1
2
221
SF has so many tasty restaurants. Tech bros, you make more money than anybody else in the city, why are you eating at Souvla every single meal?
1
1
4
620
One terrible lie of modern society is that you’re only getting started with life when you are 18, 20, 21, 22, or some other arbitrary point in your 20s. If we’re optimistically expecting people to live until 90 years old, 22 is nearly quarter of the way through. We can’t be telling people that a quarter of life isn’t really living life. Life begins at conception. Life experiences begin the moment you are born. This pernicious lie discourages people from living out meaningful moments of life, discourages people from having agency, and is a major reason why many young people can’t grow up and mature, despite having more opportunities and resources than at any other point in human history. No, you’re not waiting another year for life to begin. You are living life right now.
7
3
1
128
14,799
One of the best books I’ve read in high school is called “Good Economics for Hard Times”. The biggest takeaway I had from the book is that it is easy to do something well when conditions are perfect. But actual skill arises when you managed to do something well even when conditions are imperfect. This applies to any craft, including my own, software engineering. Your environment will throw difficulties after you. Supply chains might stop working. The tools at your disposal might not be as good as you’ve hoped. You might not always have the resources you need. Your value is determined by how well you can work with other people and adapt to your environment.
3
120
CS students, one of the best things you can do for yourself today would be to go see this world. Literally. Go travelling, go to an art gallery, go see a concert. It is becoming more and more apparent building software isn’t just a science anymore. Software is inherently a creative profession. AI can nail the science part of it today. It can write sophisticated algorithms. But a human still has to design the software, even if this human uses AI to construct the design. Study how various art forms are made. For example, Monet didn’t just “come up” with a painting of water lilies. He went to his house in Giverny, he collected the right painting materials, he thought about where to paint, how to paint. It wasn’t effortless even if Monet wanted it to appear that way! It takes effort and meticulous planning to create something original to wow the world. If you want to thrive in the world of software, the more creative, original, imaginative you could be, the better your career would be.
3
207
Not a bad way to save money and use your money for fun things like travel instead, or to save up to buy a house elsewhere. The part I’ll be most scared about, unironically, would be how the building will fare during a destructive earthquake.
A Microsoft AI scientist chose to live in a tiny room in San Francisco’s Chinatown for less than $1,000 a month. @annakook gives us a rare look inside a single-room occupancy building, where residents have private rooms but share kitchens and bathrooms.
115
James Fang retweeted
🚨🚨🚨 vLLM is all you need to reproduce DeepSeek’s performance. With Wide-EP and EPLB, vLLM can match DeepInfer’s decode performance within 6% and generate $39/gpu/hr in revenue at DeepSeek’s off-peak pricing, at 200 P90 TPS on the AgentX workload. Specifically: • Under DeepSeek’s decode configuration (DeepSeek-V4.1-Flash, 131K context, nextn=3), vLLM on GB200 achieves throughput within 6% of DeepInfer’s published results across batch sizes 8–256. • On AgentX, vLLM pushes the Pareto frontier further on GB300, reaching up to 1M tokens/s/GPU and generating $38.96/gpu/hr in revenue at 196 P90 TPS, based on DeepSeek’s off-peak API pricing. • Also, vLLM extended low-latency kernel support for DeepSeek-V4.1-Flash, achieving up to 826 TPS per user on AgentX.
36
97
44
1,040
114,267
Maybe taxes should be on a gradient after all, this will actually force Americans to be proficient in calculus and there solve the American numeracy problem
74