James Lin
@jlinbio
Followers
3,229
Following
562
Media
123
Statuses
609
Slaying dragons. "Those who lack the courage will always find a philosophy to justify it." — Camus.
SF, Toronto, Boston
Joined January 2022
Don't wanna be here?
Send us removal request.
Explore trending content on Musk Viewer
Eagles
• 232598 Tweets
Packers
• 133849 Tweets
श्री गणेश
• 123537 Tweets
Ganesh Chaturthi
• 117299 Tweets
गणपति बप्पा
• 87165 Tweets
भगवान गणेश
• 45130 Tweets
#二度と撮れない画像を貼れ
• 43577 Tweets
JustinOn NCAA100Kickoff
• 34349 Tweets
Party Love AndaLookkaew
• 34118 Tweets
Jordan Love
• 33114 Tweets
#GanpatiBappaMorya
• 31339 Tweets
#マイナビTGC
• 26019 Tweets
#Smoke1stWin
• 24062 Tweets
#学マスデビュー初心_名古屋
• 18735 Tweets
ビッグラン
• 16685 Tweets
Nikocado Avocado
• 12730 Tweets
ボンドガール
• 10559 Tweets
クリスマスパレード
• 10530 Tweets
紫苑ステークス
• 10372 Tweets
Last Seen Profiles
Pinned Tweet
I was challenged to code the transformer architecture onto a Nvidia CUDA kernel. It's been a tremendous source of learning so far.
Progress tracker and implementation details:
15
45
609
Highest signal-to-noise pages on the internet:
1.
2.
3.
4.
5.
6.
7.
8.
9.
39
281
3K
Essential ML papers:
1. Transformers: Attention is All You Need
2. BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding
3. GPT: Language Models are Few-Shot Learners
4. CNNs:
6
68
387
Was challenged to code the transformer architecture onto a CUDA kernel.
Stage 1 ✅: Element-wise vector addition!
13
25
369
Stage 3 ✅: Softmax, dot product, and matrix operations!
Almost finished with core transformer operations on CUDA.
Let's dive in:
7
10
284
Stage 4 ✅: ReLU, batch matmul, and layernorm!
With these kernels, we've finally encoded all core transformer operations into CUDA.
Let's dive in:
7
14
222
Solved the Good Will Hunting math problem last night, I guess I'm better than an MIT professor at math now
4
4
179
Focus is a prerequisite for great work. This is because most hard problems are soluble in deep effort, but insoluble in shallow attempts.
6
11
187
Researchers have reported a room-temperature superconductor, LK99, that operates at ambient-pressure.
Here's the breakdown and why it could be a Nobel-worthy 🏆 discovery:
1/
5
47
178
All math olympiad nerds I know are obsessed with anime. The more obsessed, the better their IMO/AIME scores.
Food for thought.
11
7
170
The secret to motivation is to start working for someone you respect and absolutely don't want to let down
16
8
147
Chatting with hard tech founders is the best.
Today I learned that you can submit silicon chip designs and Google will straight up fab & manufacture them for you at ZERO cost. Delivered to your doorstep.
So cool.
8
14
139
(Ramble/thoughts)
Today, I had the most nerve-racking professional experience of my life.
I gave a talk for a biodefense panel at Stanford. The attendees were important people from DC. The other speakers were all Stanford professors or CEOs of $100M biotechs.
I'm just a kid
11
2
131
If you want to understand transformers, this is an excellent visual walkthrough:
1
12
124
Stumbled upon a list of foundational philosophical works.
Each book is succinct and captures the essence of its particular philosophy.
"Great Ideas" by Penguin Random House.
5
2
81
Young people severely underestimate themselves.
Why would anyone want to hire you?
Because you have three vital strengths.
1. Coachability
You can learn insanely fast and pick up new skills. Veterans are set in old ways. Be a sponge, devour knowledge, ask questions.
2.
5
3
84
I coded a simple evolutionary simulation for genomes.
Why it matters:
2
1
78
Friendships are a function of time and intensity. 'Grabbing coffee' once a year doesn't cut it.
Work on projects together, go on road trips, 1v1 in Minecraft, shoot the shit.
6
1
64
Y'all aren't ready for me to become a god at explaining the totality of biology & gene engineering
(Transcription: DNA -> RNA)
5
5
67
There are 2 fundamental truths:
1. Math exists in some axiomatic form
2. This dude ships
@yoheinakajima
3
4
60
I'm in Boston for the next two weeks @ the Boyden Lab researching computational brain mapping
Hmu if you're working on something cool! and bonus points if you're down to let me deep dive with you on your research, projects, etc. Coffee on me :)
5
0
63
Defense companies:
Aerospace
Lockheed Martin
Northrop Grumman
Boeing
Airbus
Textron
BAE Systems
Leonardo S.p.A.
Saab AB
General Dynamics
Thales Group
Rolls-Royce Holdings
Weapons Systems
Anduril
Shield AI
Raytheon
BAE Systems
Mach Industries
Energy
General Electric
General
9
7
58
Reminder that you can fold proteins in less than a minute
1
1
55
Still thinking about how
@willdepue
straight up calls his shots. Went from "wait why am I not working at OpenAI, this is the biggest thing ever" to getting a job there in 2 months. Respect.
5
0
55
Highlights from a wonderful year:
1. Formed lifelong friendships
2. Read 112 books - starting to feel my understanding of the world compound
3. Flew down for the Starship launch - a breathtaking experience
4. Founded and raised for a company for the first time - learned an
10
1
53
8/ Due to the unique structural properties of LK99, the deformation stress could not be relieved, which is what allowed for the permanent formation of SQWs.
The end result: superconductivity at, I repeat, ROOM-TEMPERATURE and AMBIENT-PRESSURE.
1
5
53
A day spent studying chip design with the greatest teacher in the world. Absorbed so much.
5
1
49
"What was your life like when you first got here?"
This is my favorite question to ask Uber drivers, most of whom are immigrants traveling across continents, seeking a better life. They often have crazy, interesting, and unexpected stories.
Just yesterday, I was going from SFO
2
0
43
I read 112 books last year.
Here are my top 10 in no particular order.
1. Shoe Dog by Phil Knight
The legendary founding story of Nike. Easily my favorite biography ever, but perhaps my favorite non-fiction book too. Written beautifully and full of wisdom for business and life.
4
1
42
Where sand becomes intelligence
cc: Quantum-Nano Center at Waterloo
2
0
44
Optimization papers:
1. BatchNorm
2. ReLU
3. Dropout
4. Residuals
5. Word2Vec
6. Adam
7. GeLU
8. LoRA
0
6
41
Related to the
@1517fund
@DStrachman
idea of belief capital — i.e. giving young people small grants is a strong sign of faith — there's a similar reason why 'where you live' matters so much:
1
2
40
@alexandr_wang
Hmm, if you view speaking up as a competition, a thing you do for status, I think you’ve already lost.
Speak more, speak less, they both have their place.
In higher-stakes meetings, every question and response should have a purpose. It’s efficient, a sign of respect for others’
2
3
36
Pro life tip: Be direct about what you want.
Whether in job hunting, dating, etc., people tend to flirt around the core thing.
Most opportunities die a slow death because people don't make their intentions clear.
If you like someone, just tell them. The good outweigh the bad.
1
0
36
If you want honest feedback, you can't bite back when you receive it.
Don't shoot the messenger. Don't bitterly argue why it's wrong or dodge it with "yeah but..." Don't let your ego sabotage opportunities to learn from others.
Just, listen.
2
2
36
@jeremyphoward
This is awesome, thanks! also big fan of , was the og course that got me into ML last year
0
2
36
Thinking of going to Japan next summer, any advice for making the most of a month-long trip? Activities, methods of travel, and sunrise locations much appreciated :)
16
1
31
Reading is 10x more fun/exciting when it's a second-hand book you've borrowed from a friend
1
0
32
Reminding everyone that PhDs and postdocs are lonely creatures who will eagerly tell you everything about their research if you ask.
My recent deep dives were w/ researchers working in Nobel Laureate labs!! They are dying to share stuff.
Reach out, ask questions, be curious!
2
1
33
That feeling when you finally comprehend the model and can write out the architecture without reference 🤌
1
0
30
9/ If these results pan out, the implications are profound. From the final sentences of this paper:
"The LK-99 has many possibilities for various applications such as magnet, motor, cable, levitation train, power cable, qubit for a quantum computer, THz
Antennas, etc."
3
3
29
Next up: conquering bio
1
0
28
An obvious, grand-slam product: Personal AI chef.
Imagine a robot that can:
1. Cook delicious meals with voice commands
2. Order groceries online and pick up from the door
3. Remember and curate your favorite recipe list
End-to-end. All households would sign up in a heartbeat.
5
2
28
12/ Progress is moving so fast in 2023, in AI, biology, and now superconducting. We're tunneling into a new world.
1
3
26
Went from a good understanding to an atomic-level understanding. Learning from first principles is soo rewarding.
2
1
26
13/ Hopefully this summary was helpful. Feel free to DM if you want to talk science and futures.
2
0
26
Nothing can compare to an unhurried, authentic conversation. It's a breath of fresh air from the networking rat race, where calls are stacked back-to-back like sardines in a can and relationships mass-produced like knock-off shoes which resemble the original in appearance only.
3
0
27
Modern medicine would not exist without protein engineering. The bioeconomy (stuff made w/ proteins) is worth $500B.
A layman's overview of protein engineering & fermentation:
1. Identify a target compound. Usually a medicine.
2. Identify metabolic pathway. Figure out what
2
1
25
Cancelled all meetings. Today's feeling like a deep work kinda day.
1
0
26
The tell for whether you're serious about learning is if you can extract useful information from feedback, even when the feedback is given by someone you don't like.
The opposite is true too: good feedback, given in bad faith, is hard to accept and can turn people against you.
0
1
26
Everyday I am relieved that there are 22 amino acids and 26 letters of the alphabet
3
1
22
Once I got that up and running, I had to learn the basic syntax. I mostly code with Python and Java, so I was unfamiliar with C++ (which CUDA is based on).
I got up to speed in 2 hours. Pointers alone took an hour but I've fully internalized them now. Thank god for GPT-4.
1
1
24
Just convinced my academic advisor to let me take 200% courseload, wish me luck 🙃
8
0
23
Chasing status is like peeling an onion: there are always inner rings of increasing exclusivity.
And we all know the fundamental truths of onion-peeling.
1) Eventually it disappears
2) You cry
1
3
22
We do not learn from experience; we learn from reflecting on experience — John Doerr
2
1
21
The masculine urge for a friendship where both guys are out to kill each other, to push each other to the mortal limit or die trying
2
0
21
I'm just gonna say it: squares are the most stable shape in nature.
There's a reason why animals have 4 legs, chairs have 4 legs, transmission towers have 4 legs. Because it WORKS. Triangles have been propagandized for too long.
12
0
21
Attended my first improv session yesterday.
There is an unspeakable magic when the butterflies melt away, when the outside melts away, and you find yourself immersed in a new world, where words and characters play easily. It's a new world—and yet, it is familiar.
2
0
21
@willdepue
Was thinking about implementing this! Imagine training a model on Khan Academy videos.
You could ask it about anything and it would generate a walkthrough video in Sal Khan's voice.
Would kill traditional education. Universities would no longer have to teach, only research.
3
3
22
Being back in the gym after a hiatus humbles the ego like nothing else: numbers are down, everything is sore, and light sets are mentally draining.
Time to build back up.
0
0
22
1/ This week I attended the world’s largest synthetic biology conference
@SynBioBeta
. Here’s what I learned.
2
2
20
11/ This paper was a collaboration between researchers at Korea University and Quantum Energy Research Centre, and to my knowledge, they have no history of falsifying data. Fingers crossed.
1
0
20
The best advice I've ever received is to play positive sum games with high-integrity people:
4
1
19
Stealing great ideas is a hallmark of smart execution.
History's greatest founders all studied those who came before.
@FoundersPodcast
FedMart to Walmart to Amazon—a trail of copying.
I'd be afraid of a company that ruthlessly copies the best ideas.
When people copy you, the best strategy is usually to ignore them. People who copy you are (a) unoriginal and (b) opportunists, and those are both strong predictors of failure. If you wait them out, they'll eventually drop away.
318
599
4K
3
1
20
I think I got them all. Here's a comprehensive list of all RNA types, for flashcards or curiosity:
1. mRNA | Messenger RNA
2. tRNA | Transfer RNA
3. rRNA | Ribosomal RNA
4. crRNA | CRISPR RNA
5. tracrRNA | Trans-activating CRISPR RNA
6. gRNA | Guide RNA
7. sgRNA |
3
0
19
Taking off at 1am for an unexpected 16-hour road trip.
Thrown into chaos this week, sleep-deprived, and wanting nothing more than to call with friends.
Bringing this book with me, a reminder to persevere through the storm. Light at the end of the tunnel.
—
“Let everyone
5
0
19
Over time, I've become disenchanted with "ambitious" and "smart" as descriptors
Infinitely more important is willingness to work hard and learn fast
0
0
18
Face your fears. Challenge yourself.
You cannot sharpen a razor on velvet.
1
0
18