Vector (@_vector15) header
Vector (@_vector15) profile picture

Tech and Research | ML, Systems and backend | Ex-AI eng intern @ Startup | Ex-Quant research consult @ WQ brain

127.0.0.1
Joined February 2025
Photo by Vector (@_vector15)Photo by Vector (@_vector15)Photo by Vector (@_vector15)Photo by Vector (@_vector15)Photo by Vector (@_vector15)Photo by Vector (@_vector15)Photo by Vector (@_vector15)Photo by Vector (@_vector15)Photo by Vector (@_vector15)Photo by Vector (@_vector15)
Vector retweeted
my girlfriend and i love playing wordle, so naturally, i ended up building my own version. i don't particularly like the nyt site, and everyone on the internet gets the same word every day. most alternatives are full of ads and aren't quite what i wanted. it's clean, and you can play remotely with friends by taking turns guessing on the same board (even my mum plays it now). my favourite part, though, is the word bank. it accepts some very questionable guesses, and if you're stuck, you get hints after three guesses that tell you just enough without giving the word away. every word you solve comes with its meaning and etymology, and you can ask for harder words too, so it's a neat lil way to expand your vocab. probably my favourite thing i've built so far, simply because i actually prefer using it over what's already out there. give it a go (looks especially nice on mobile btw)
19
10
75
1,539
Vector retweeted
gta 6 é daora e tal mas acho que vou jogar um pokemon fire red
110
2,908
242
14,142
177,847
Vector retweeted
Today I learned ⊃∪∩∈ is a story about set theory.
163
1,653
126
19,953
584,840
Vector retweeted
i've been desensitized to exams i don't even feel the stress last minute for the post procrastination lock in to kick in, well whatever nvm.
7
3
55
982
Vector retweeted
i've been reading quite a lot of research papers over the past couple of months (even got to help out on one), and i liked the idea of picking one problem and going deeper than necessary instead of building something i didn't need. so i tried that with prefix caching in llms. i built a profiler to see what happens between a coding agent and the inference engine, byte by byte, got some numbers, then spent way more time figuring out what they meant. this is me summing all that up in a neat lil writeup (kinda long tbh).
18
11
3
105
2,494
when the test set (midterm) is a completely different data distribution than the train set (homework problems) and i can’t generalize
19
171
6
3,196
44,596
theory: in <domain you care about> there are only about 500 people in the world who really matter (ie: do the work, have impact, genuine, non grifter, non blowhards, will deepen and broaden your own skills and perspectives). it is your job to find them and make friends with them.
51
309
42
4,670
145,269
My advice is to stop optimizing yourself around becoming a better software engineer by conventional standards. We need to raise the standard so high that computers themselves have to stop computing the way they do today . just be psycopath Don't spend your entire career chasing the next programming language, framework, technology stack, or engineering trend. Those are tools, and the cost of using them is changing rapidly. Instead, become the person who invents things that don't exist yet. Find problems nobody has properly understood, explore ideas nobody is working on, and develop expertise in extremely niche areas that could become entirely new technological moats. Go deeper into mathematics, computer science, physics, economics, biology, and other disciplines. understand the underlying principles, discover connections between fields, challenge existing assumptions, and develop entirely new ways of thinking. AI doesn't make deep technical knowledge irrelevant. It makes understanding correctness, complexity, architecture, performance, and the fundamental limits of computation even more important. Try something so niche that it eventually becomes its own category. Explore different kinds of experiments, even if nobody pays attention to them initially. Don't chase attention or validation Being extremely niche is a new moat
What a software engineer should do now
8
74
1
1,178
48,161
Vector retweeted
التأجيل والتسويف أنعم أشكال الخوف، كذبوك فقالوا كسل
38
439
28
6,361
169,377
Vector retweeted
同じ場所で撮った写真20枚です!!!
190
4,463
184
36,144
685,085
Posting because these are words that resonate deep inside my bones, and it deserves a permanent spot on my feed.
2
7
253
203
11,426
114
92,855
984,187
Vector retweeted
de chiquito tienes que tener una compu de mierda que solo corra emuladores y juegos de hace 15 años, de lo contrario no desarrollas alma
510
12,486
657
91,284
1,151,791
and as a final trick i’ll believe in myself so deeply that my reality has not choice but to catch up.
1
9
3
56
3,407
started building a coding agent of my own without ai this will teach you a lot about how ai systems work under the hood involving context management, memory architectures, tool registries and other ai fundamentals, super excited to see how it turns out :)
27
6
1
169
5,077
"what a privilege it is to be exhausted by an ambition of your own choosing"
what a privilege it is to be exhausted by an ambition of your own choosing
98
31,743
341
227,385
2,214,615
Vector retweeted
writeup on MCPs, building your own MCP server and integrating third party MCP servers is out now. [ link in comments ]
5
6
27
702
Vector retweeted
Bounce back man, get your spark back. You deserve to be happy.
119
2,294
112
21,633
460,224
anyone hiring for backend/ml? brownie points if you share this, i’ve got exp, but no oss recently. have to do that again :(
8
10
38
1,550