python mines @mistralAI || rust oss when I have time || vim, nix, and haskell when i feel like it
- Tweets 6,600
- Following 1,534
- Followers 6,419
- Likes 25,453
Pinned Tweet
You asked, I deliver (2 weeks late).
A blogpost about what I did over the last 18 months to sign as a research engineer at a foundation model lab.
Link and summary below
1/n
Today I signed with @MistralAI as a Research Engineer.
Hard to put into words how excited I am.
46
234
3,151
640,502
> friend of a friend
> didn’t get a MIT exchange via his German Uni
> booked 1 way ticket to Boston
> knocked on doors til he got a random prof to sign off
> stayed there
> was literally a visiting researcher in undergrad
You can literally just do things
44
213
8,735
399,519
become a self-taught ml engineer
i published this blog yesterday, a year ago -- 8.9k reads to date.
i'm pretty excited that my alma mater invited me to give a talk on this tomorrow
(link to blog in comments)
18
166
2,219
186,027
Today I signed with @MistralAI as a Research Engineer.
Hard to put into words how excited I am.
109
13
1,360
511,544
I had a couple of people in my DM's asking for advice on how to become a ML Engineer.
My answer has been more or less the same each time.
So i put in some effort, did a proper write-up on my blog, and links to some great actionable resources
Link below & in bio
19
100
1,175
169,064
How to become an ML engineer in 5 to 7 steps.
I wrote that 11 months ago and it just reached 8k reads.
5
82
1,182
95,521
Credentialism and it’s consequences habe been a disaster for the human race.
The main thing i learned at Berkeley is that there is no hidden knowledge
You can, in principle, learn anything from books/the internet at home
29
59
831
66,125
Yes, he flew there and went to classes for two weeks and then went to the professors asking if they were interested in him as a visiting researcher and that he flew there on a one way ticket because that was his dream.
3
2
818
15,744
fellow ML grinders & anime pfp's
quote below reminded me of a great free ML resource on arXiv
"A high-bias, low-variance intro to ML"
Covers a lot of ground in only 100ish pages and is useful for anyone with some mathematical aptitude
great resource to get started, and to refer back to
link below
12
76
793
144,486
That was in 2017. I suppose he had a return flight he didn't took. The point is, he went there without any commitment from a prof, knowing it would be more likely to get one when you meet them in person.
Once you have the letter from the prof, it's fairly easy to get a J1.
1
2
611
22,618
I think you're underestimating the spanish. They build cars, make fashion, grow food, energy, valencia and barcelona are busy ports, telefonica, seat, santander, vodafone ... but tourism is a big part too.
4
1
508
21,038
How about not policing how others look and assuming they do everything in relation to you and your preferences...
3
6
481
108,728
Which is why I am so impressed. I know a couple of people (myself included) who networked their way into these universities. But cold blood flying there, going on campus and introducing yourself is an insane level of commitment which is what I think that sold him.
1
480
6,518
I still remember my last HS Finals where i first felt like „damn, if you study you actually know stuff and don’t have to triangulate your vague memory, common sense, and what you can infer from the exercise“
3
6
358
5,837
A lot of sociological research points to that the barrier for men to get intimate with someone is (way) lower than for women.
But for committing to relationships there are no statistical differences w.r.t standards and expectations from a partner between the sexes
4
4
304
29,299
The probability distribution of the location of the electron is a standing wave.
The atom itself is mostly empty space...
Fun fact, the probability of the electron to be inside the proton is larger than zero.
14
274
12,512
new website, who dis?
got an inaugural blog post on how to learn hard things bc. i'm doing that for a while and got things to say.
Also have a neat collection on learning resources i like.
will build it out over time.
let me know what you think.
link in bio 🌶️
8
5
270
24,207
Wow. kinda crazy close to 5k people clicked my ML Engineering blog post.
every now and then I chat with people about it in my DMs, too.
I probably should blog more...
4
7
259
14,539
What I learned during three months programming at @recursecenter in NYC. (link below).
5
13
237
19,984
I had many DMs asking for advice.
I've been working towards a career inflecting role since April last year. While I won't share Mistral specifics (duh!), I can talk in general terms about worked for me and what didn't.
Blogpost coming in the next few days!
Today I signed with @MistralAI as a Research Engineer.
Hard to put into words how excited I am.
9
2
245
25,938
Techno, nvim, reproducing results, split keeb', candle after having friends over for brunch.
What a sunday
12
2
228
8,370
> be me, study social science
> Love stats, hate yaps
> F off to physics
> Kinda prefer people to particles
> Learn ML for quant SocSci with online data
> Build SWE skills to do better ML
> I'm in love
> God put me on earth to SWE
I did it, anons, I found my passion
wagmi
10
211
16,963
The hoops continue.
It’s called management consulting.
Waking up and leaving the hoops was painful.
But worth it.
1
171
28,255
A 2 year course based masters is usually a prerequisite for european 3 year PhDs. So both have 5 years of grad school
6
1
161
17,551
Belated 2024 half time learning update
~ 500 after work hours on
- k8s
- Docker
- Python Deep Dive
- Deep Learning for NLP
- DSA Fundamentals + LC
- Svelte
- Designing Computer Programs
- nvim, tmux, typing faster, general terminal tools
- Git Deep Dive
5
1
166
9,964
What writing "Writing an Interpreter in Go" in Rust taught me about learning to code better and faster.
(link in comment)
5
5
169
11,297
3 months, six interviews, about 100 hrs of work, final 4% out of 2k Applications, and i just got gutted out of the process.
Fuck.
Let’s say, I‘ve had better days.
Cya tmrw
17
139
8,620
On one hand it’s apples and oranges because the laws of physics make founding and sustaining an extraterrestrial colony much harder that flight.
On the other, i have no doubt man will eventually live on other worlds. The first human born off earth is probably closer than the Pyramids on a time scale
3
1
115
27,635
Yeah, probably. He at least had the funds available to cover his flights and stay. Which is a privileged situation and not available to everyone.
But he also demonstrated agency with his actions to make the most out of it.
1
113
15,522
I attempted becoming a software engineer twice.
The first time failed miserably.
Today, I'm doing quite well.
I've wrote a blog about what worked and what you should definitely avoid.
I hope it's useful to some!
(link in comment)
7
2
97
3,975
I‘ve read that earlier this year and found it a digestible resource with a good trade off between mathematical rigor and application.
You can kinda skim through his anecdotes though.
Combine that with a bunch of LC (e.g. Neetcode 150 etc. ) and youre good.
4
1
92
6,153
here you go: arxiv.org/abs/1803.08823
2
3
88
4,889
I used to think that (not speak out loud, i'm not cocky).
Then i got utterly humbled by a few competitions.
Was a slap in the face that made me learn (again) that there is a difference between theoretical knowledge and practical skill.
1
88
9,247
My roomie once said
"everything you want to be good requires work"
He said it about a party we planned (was awesome, btw) but it stuck with me
You can literally make things good by putting in effort
2
3
78
7,676
The main value in uni is having someone curating your curriculum, spoon feeding you the content, and getting a nice piece of paper employers like.
Elite schools also may give you a very strong network into a countries elites.
The knowledge is not a differentiator
1
2
72
2,089
I’m becoming a 30 yo boomer today,
No Rust on day 23/64.
Tomorrow, it’s back to work. Today it’s celebrating life!
I’m looking at the next decade like I’m looking at that pawn.
Hungry for more!
nom nom nom
Day 22 w5d2 addendum:
Finished up the "Building Bitcoin in Rust" Book.
Chain runs and the whole thing took 9 days to build.
nice
24
83
5,263
But Greene, Ferris, Kyosaki, Carnegie definitely are.
Disagree on peterson.
Hawking definitely knows his stuff, but i am vary on pop sci physics books that avoid formulas.
8
66
4,295
As a self-taught SWE who spent quite some time at uni, so i got some thoughts
Is it necessary? No.
But
- It's a trust signal
- Buys time to learn useful stuff
- Gives you friends + powerful network
- Opens doors
For the full thing: link in bio🌶️ & first comment
5
3
73
5,384
Haskell is like Latin.
It doesn't help you much for your trip to spain but it makes a strong foundation for your career as a linguist.
6
5
71
5,839
It‘s simple.
Not easy.
But simple.
I started late '22 grokking ML, CS and SWE textbooks.
Throw in some moocs, do the exercises, build stuff, find a community, keep a steady pace you can sustain for years and remarkable things happen.
It did take 7 months of this to get an SWE job, another 7 months later i was invited as a speaker on SWE conferences.
Great things happen with consistent work
9
67
12,338
All i want from life rn is:
- lots of quality time with friends and gf
- lift heavy stuff and put it down again
- surf big and often
- write great software and get cracked
4
1
69
10,418
Especially when it becomes a regular thing.
Any consistent input into the body will shape it a certain way.
2
1
61
21,780
This is kinda weird bc. Amazon is one of the most data driven companies there is.
To understand data, you need qualitative inquiry. But you cannot escape your own observer bias with anecdotal evidence.
2
65
1,041
Important to note that he is nothing official but acts on his own (much to the malaise of local law enforcement who, as they say, have much better things to do).
2
61
3,952
There is a lot of BANG for very little time invested in using SWE practices in your data science and ML work.
Yes, also in research.
Free compute in Colab/Kaggle is great and Jupyter Notebooks are a great education tool.
Sadly, it nudges counterproductive behaviour.
1/2
>NOOOOO Ludwig you DON'T understand! We are SCIENTISTS! A REPL-like environment like Jupyter notebooks are actually OPTIMAL for the kind of quick-iteration explorative work we do!
6
1
58
5,050
Idk. Sure, the first application after college, scholarships etc. all look at gpa’s too. But us elite unis spend tons on their branding and thus the Cv stamp glows. And education is smth that stays on your CV long.
1
53
11,257
Started the self-taught CS grind late '22 and took this year much serious than last year.
6
51
1,904
I in HS i dropped Physics in 10th grade and took math pass/fail. Went to study sosci. There, i discovered I kinda like stats and experiments. So i started over with Physics. At the time i barely knew trig functions.
I now have a graduate degree. People have different timelines.
49
1,648
On more complex solo projects, i use feature branches with a dirty log, then squash merge and use the commit message as a dev diary.
Really useful to document why you did what (and not something else), benchmark results, whatever.
But also only when the project is somewhat large / mature.
2
52
3,411
DEVELOPERS DEVELOPERS DEVELOPERS DEVELOPERS DEVELOPERS DEVELOPERS DEVELOPERS DEVELOPERS DEVELOPERS DEVELOPERS DEVELOPERS
DEVELOPERS DEVELOPERS DEVELOPERS DEVELOPERS DEVELOPERS DEVELOPERS DEVELOPERS<voice cracks>DEVELOPERS DEVELOPERS DEVELOPERS
2
53
5,519
I started my academic career in sociology and in one of the first lectures my prof said: "The decisions you make from now to 30 are the most influential on your lifetime earnings, happiness, and health -- no pressure".
1
2
48
2,277
utterly humbled by leetcode today.
i‘m a paid professional
wtf
I should be able to solve this
must. get. better.
2
49
3,504
Finally finished.
It started of great, but i procrastinated through the last 150ish pages for the last 3 months.
As someone who sidetracked into SWE from data science and mostly build library and 1dev-1machine tools, this was pretty eye opening
2
51
2,768
Hard disagree on 6.
Choosing any resource to move forward is almost always better than searching for the perfect one and not starting.
Also 5 is bogus. You cheat yourself when you only learn high level and lack the fundamentals to understand it fully
7
3
50
10,939
Germanys wealth amassed during decades of export surplus allows us to indulge in expensive leisure activities such as political activism that hates change as much as stagnation.
(It‘s basically NIMBY‘ism paired with a genetically inherited trait of larger than life figures that don’t sportfunny mustaches).
2
47
6,500
2024:
- built OSS tool with >20k dwnld in Dec (with team).
- presented at PyData & AI Tinkerers
- made 6th/final round for ML research at a big lab (but didn't get it 🥲)
- >1000 after work hours in learning CS/ML fundamentals
- Read 6 textbooks
- reached intermediate level in Haskell, C, C++
- got pretty good in DevOps with docker and k8s
- Started learning parallel computing/CUDA
- 2nd place at a hackathon (utterly trashed at another, lol)
- increased typing speed by 30 wpm
- Benched 110kgs
- Grew this account a bit
- Quit my job 'cause personal and company goals diverged
- Digital nomad life in Spain for 3 months, and lots of quality time with family & friends in London, SF, Vienna, Frankfurt, Alps, at Home, on festivals
10/10, will continue
4
1
49
1,781
Residue theorem was one of the most mindblowing ones in my maths education. Like, i really did not see that coming.
1
43
2,226
Consistent hard working midwits soon outwork even semi motivated geniuses.
Put in the time, trust the process, you will be fine.
1
44
2,932
To be at MIT and learn that the people there are also just human instead of some stylized super geniuses, i guess. Idk
1
45
4,455
For a lot of self taught people it‘s probably about learning more about the tools you work with. CPUs, cores, cache lines, ram. There is a lot of demystifying happening when you go down that route.
1
46
2,493
As much as I would want to be type 2), I'm probably 1).
After all, I chose haskell and compiler theory over indie hacking something trying to make money.
Coming to terms with who you are is growing up, I guess.
1
45
7,891
I started making real progress in my self taught SWE journey once i let go of any notion of time.
No shortcuts
Learning takes time.
Accept that instead of rushing to an arbitrary 3 months deadline.
To me, the difference was night and day!
4
43
791
T - 1 week until becoming a dramatically better programmer for 3 months straight
Incredibly excited.
Still got a couple todo's until then.
1) Finish e2e paper draft & give to supervisor
2) Finish Rustlings (4 chapters to go)
3) Finish parallel computing course (~80% done)
6
44
3,439