sumit
@sumitdotml
Followers
10K
Following
4K
Media
413
Statuses
3K
swe on a long journey to learn ml until I reach a profound understanding of everything
日本 東京
Joined September 2022
had to triple check to make sure I wasn’t tripping - are you freaking serious dude????
Chad IDE (@cladlabs) is the brainrot code editor. AI coding takes 1–5 min between prompts—too long to ignore, too short to start something new. Chad integrates your brainrot (X, IG, Stake, Tinder, etc) into your agentic coding workflow and helps to manage your
1
0
19
got nerdsniped again, I ended up spending my whole evening figuring out how to properly add read counts to my blog pages with cloudflare's workers kv although making a little progress on some study logs in the process :) hopefully I can churn out at least 3 this week
2
0
32
so I did a little something…
my intel macbook might be finished… the way it’s performing it makes me feel it’s running on hdd it’s so bad desperately need a new laptop, the only small issue here is that I have no money left & the macbooks I want are way too expensive
31
28
2K
having to work this hard for something & that something is not matrix manipulation… this feels like I’m committing a crime fomo is real
0
0
0
one of the more ambitious projects that’ll have a considerable impact & I’m the sole dev in charge, well at least my resume will look good!
2
0
8
got nerdsniped by copy heads when I was going through a paper on attention mechanics & all I could think of was the "do nothing, win" meme since it seems that's what these chaps do across residual streams
1
0
1
I sometimes think to myself how simpler it would be for me to get some nice high-paying dev roles if I wanted to, mostly because of my current job how unseemly that this moron just wants to do matmuls & read papers
2
0
17
one of the more ambitious projects that’ll have a considerable impact & I’m the sole dev in charge, well at least my resume will look good!
was barely finished recovering from a big project at work & immediately got assigned to another big one
0
0
11
came back to say this is probably the best thing I've been fortunate enough to read this year, so much gold in here if hf was selling this as a physical book I'd buy it immediately
Training LLMs end to end is hard. Very excited to share our new blog (book?) that cover the full pipeline: pre-training, post-training and infra. 200+ pages of what worked, what didn’t, and how to make it run reliably https://t.co/iN2JtWhn23
2
0
14
felt a bit nauseous inside the house today, so decided on a slight change of pace
6
1
75
ok the picture I get it (not really) but there’s 0 reason for this tweet to get these likes… and these bookmarks?????? why are you bookmarking this???? deeply deeply unserious platform
what on earth
2
0
15
the sheer number of accounts that have come to me telling me about this typst app after that tweet of mine where I was writing in markdown… is this some kind of psyop? and yes I like using plain markdown in my neovim don’t worry about me
2
0
7
what on earth
16
14
2K
everyone talks about how doing neovim means less distractions, more focus, more efficiency etc etc well I’m a simpleton, the primary reason for me is that it just looks so damn pretty & I just wanna use it all the time lmao
2
0
36
makes you physically ill reaching out to people you deeply respect who feel 'approachable' for sharing ideas or asking for opportunities, like what if they don't respond what if they think x or y etc etc there's overthinking & there's me ugh (still reaching out though)
2
0
21
a lovely anthropic paper from 2021 on attention mechanics, was very helpful to someone like me who finds it difficult to visualize concepts https://t.co/xjfzWw0JJl
0
4
43
this should actually make you look cooler imo you’re in the same room as them despite not having the traditional credentials “wow this person must be crazy good for them to be here without a PhD”
nothing worse for my impostor syndrome than my title being "Scientist" while still not having done a PhD
1
1
12
I look at the crazy shareholder value I generate at my work every week & imagine how better it’d be if these 8-10 hours per day could be used for mechinterp research instead of this corpo swe slop hooo boy
2
0
16
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
212
3K