your friendly neighbourhood engineer, working on sample efficiency for robots
- Tweets 13,709
- Following 2,547
- Followers 11,764
- Likes 77,475
Pinned Tweet
born in right generation
huge ass neural nets
cuda errors
claude code
iced americano
2
8
182
52,087
I hate how research papers are written.
so much yapping in them.
just get to the point bro
368
20,636
165,459
3,794,888
there is this niche genre on YouTube, videos < 40k views. A extremely gorgeous human being just sitting and talking to the camera, high quality audio, life lessons, deep talk no fancy lightning or editing.
just a person being honest and talking in front of the camera. I enjoy it
49
537
10,058
311,265
how it feels to understand that every skill issue is temporary and can be solved with time and focus.
27
884
6,879
262,494
this video is all you need to understand what does "drive" mean and why motivation is fake.
34
306
6,392
306,675
-writes infra critical low level code in go
-doesn't use syntax highlighting
-uses a 25 year old code editor
-wears sunglasses while he codes
-sometimes works 9 hrs straight no bs
definition of cracked
51
307
5,069
387,372
> see someone god like at design
> check bio
> ex apple
always
apple is of design what stripe is of computer engineering.
23
86
4,551
183,525
people who do gym, solve maths, run marathons, write essays, build are way cooler in my books.
it takes effort to do things, especially the hard ones.
45
235
4,496
91,027
I think if you pick a really hard research problem, try to solve it in <8 weeks, blog everything, shoot videos, ship good oss code and repeat it.
you can get hired by a big lab in ~1 year.
52
185
4,329
272,248
6 days ago
- very basic knowledge of encryption/decryption
- 0 app dev experience
- vague knowledge about network security
today
- build a app for encryption and authenticating for one of the biggest user base out there.
- won a national hack
- planned to deploy to 1.3B users
69
80
3,159
170,631
lmao this tweet is so fucking funny bro just read research papers and tweeted he cracked diffusion language models HAHAHAAHAH
I’m a 19 y/o founder from Germany, and I cracked how Google’s Text Diffusion model works!
There are three main approaches on this topic:
Community note
The video in this post is from this blog:
hkunlp.github.io/blog/2025/drea
48
45
3,092
259,245
so many high status people are engineers because engineering blesses you with a attitude and the ability to pick up any skill or concept out of thin air if you are curious enough to learn it, you just need to put in the hours.
people who don't understand this usually complain.
33
187
2,409
90,537
karpathy sensei's new venture.
it will take you from basics of language modeling, machine learning to Multimodality, RLHF, deployment.
I can't get enough of this man.
github.com/karpathy/LLM101n
16
178
2,110
111,721
chinese bros back at it again
- train a Decoder-Only transformer < 3 hrs on a 3090
- fully studded with LoRA, DPO, SFT.
- well documented training
- vision, moe, and other goodies also avail.
19
204
1,996
123,578
investing in a good computer can basically change your life.
my baby runs 60+ chrome tabs, 30+ safari tabs, discord, 5 tabs of terminal, neovim with 14+ buffers, cursor with ONE javascript project open, and while Spotify blasting music.
and never makes one noise.
48
107
1,883
139,200
people don't want you to know this but you can make yourself get addicted to putting in high number of focus hours.
17
113
1,833
83,562
so funny when people tell me to do this.
I went balls deep into the linux rabbit hole when I was like 14. Wrote custom window manager for X in C.
attached images are some of my favourite creations
(some of which are custom themes curated to my taste)
52
71
1,868
197,988
as soon as you realise that reading and thinking of being productive is not real work.
you can save so much of your time.
basically get into cycle of doing actual work and getting the dopamine out if it.
repeat this. profit.
10
143
1,489
56,992
how fake was this demo
when you get access to gpt-5, try a message like "use beatbot to make a sick beat to celebrate gpt-5".
it's a nice preview of what we think this will be like as AI starts to generate its own UX and interfaces get more dynamic.
it's cool that you can interact with the synthesizer directly or ask chatgpt to make changes!
35
17
1,530
217,678
sudden urge to condense all my knowledge of computers, hardware, maths, machines, biotech, robotics
in a single markdown file.
28
34
1,005
46,706
fun fact, you can turn your life around in 6 months.
all it takes are the 3 d's
DRIVE
DISCIPLINE
DAWG (in you)
17
70
1,202
43,730
networking?
I just dm every follower of mine and ask them "what are you working" on.
I made so many friends this way.
this shit is easy
43
24
1,114
63,034
you should work on things you *really* want to work on; if you dont you end up killing your soul and your taste.
learned this the hard way
19
86
1,113
23,880
This is such an interesting way to assess someone's talent.
Give problem + deadline → observe approach. Reveals strengths, weaknesses, problem-solving, and time management. Simple, effective.
22
41
1,027
85,748
most people arent even trying that hard, you can just become world class in something and become relevant in the field if you try hard.
20
53
827
20,340
fuck man, to be a researcher at deepmind and to be able work on something like this.
We can now train AI inside the mind of another AI. 🤯
🌍 Our world model, Genie 3, imagines and generates new worlds on the fly.
🤖 Our embodied agent, Sima, is dropped in and learns to navigate them autonomously.
The entire loop—from the environment to the action—is generated by AI.
This is just the first step, soon we will have world simulators for training general embodied intelligence.
10
30
850
43,402
chatgpt
- buggy ui
- slower model
- occasionally down
- more yapping.
claude
- excellent ui and aesthetics
- better model imo
- better speed
- using the model feels like having a smart buddy that doesn't yap (comparatively).
- Artifacts actually makes sense unlike oai
55
27
779
58,025
someone started this
HYPE
twentytextbooks.bearblog.dev
i keep saying this
someone should make a website that just list 20 textbooks to get you ahead of 99% of people in that field.
11
75
730
72,105
dis is going to solve robotics btw
we just released RF-DETR segmentation preview
RF-DETR is 3x faster and more accurate than the largest YOLO11 when evaluated COCO segmentation benchmark
we plan to launch full family of models by the end of October
↓ fine-tuning notebook below
repo: github.com/roboflow/rf-detr
23
30
759
54,003
I hate medium dot com so much.
if you write there I have a personal level beef with you
26
14
594
34,362
going to pop a redbull tonight and speed run this with some jazz
deep-ml.com
14
29
557
159,180
meta just solved like 9 different problems for me with a single model wtaf
Introducing SAM 3D, the newest addition to the SAM collection, bringing common sense 3D understanding of everyday images. SAM 3D includes two models:
🛋️ SAM 3D Objects for object and scene reconstruction
🧑🤝🧑 SAM 3D Body for human pose and shape estimation
Both models achieve state-of-the-art performance transforming static 2D images into vivid, accurate reconstructions.
🔗 Learn more: go.meta.me/305985
31
98
2,748
290,475
there is prolly like a trillion dollar worth of value in GitHub repositories under this org
6
4
550
26,737
no one owns that domain lol, some random dude keeps on changing it to whoever owns the SOTA and it is the funniest thing ever
15
20
521
19,036
if you as an engineer are still using the claude webui, instead of api.
i have a news for you son.
36
5
463
31,887
i firmly believe that uni's are intensionally super slow with their curriculum , so that the students get used to slow and ugly corporate life.
aint no way there are people who need to spend 4 years learning basics of computers and electronics.
20
20
399
53,486
if you are a 20 or so kid in tech right now you have 2 ways.
way 1 go to the ai slop generation route and fry the brains of all coming generations.
way 2 go take the robotics pill and try and automate all possible human labour.
kinda red pill blue pill stuff
17
10
438
15,752
idek how people used to study before chatgpt.
I just chunk all the study material into it,
ask it tons of questions
make gpt ask questions to me.
repeat for every module
voila.
5 hrs max for churning 3000ish pages.
11
31
409
33,743
GRPO debate is truly flawed, deepseek and Kimi can complete with frontier models with GRPO so it doesn't really matter.
all the RL people know what actually matters is the priors (the base model we pre train on top of); if anyone says otherwise they are prolly cope maxxing
16
9
402
44,021
actual list for
"The optimal path to mathematical mastery"
sheafification.com/the-fast-…
by @anisomorphism
i keep saying this
someone should make a website that just list 20 textbooks to get you ahead of 99% of people in that field.
4
25
386
28,590
opens linkdin
finds a dude swapping his face onto @cory
closes linkdin
how is that website real lmao
23
4
398
34,959
kinda sucks most student research positions are only for PhD students like, kinda beats the point
11
8
389
28,600
oh boy it is already nov i dont have a internship offer lined up for the summer
i am cooked
26
2
346
28,260
> sees a blog, "implement rag from scratch"
> hmm interesting, opens the blog.
> import langchain
> realises I am on the medium website.
honestly my fault for ever trusting those mid-ium links
7
7
336
10,473
if you want to build something similar,
here are a few links go write some python
github.com/ml-explore/mlx-ex…
github.com/VikParuchuri/mark…
github.com/vegaluisjose/mlx-…
huggingface.co/deepseek-ai/d…
mermaid.live
exams coming soon time to pull out my exam study stack, about to put 40 pdfs into a llm context and make it write notes for me.
how was I even studying before.
8
38
328
26,466
half month (maximum) of 12 hr sprints is enough for the things they teach in 4 years.
kinda insane.
i firmly believe that uni's are intensionally super slow with their curriculum , so that the students get used to slow and ugly corporate life.
aint no way there are people who need to spend 4 years learning basics of computers and electronics.
30
8
318
27,641
the goal is to generate infinite runaway to be able to do things with friends and family that spark joy in life.
money was never the goal
8
17
295
7,990
btw you can just ignore what's on the job page and pitch the founder to hire you.
job pages are useless anyway
12
9
302
24,512
any to audio paper
any to audio paper
any to audio paper
neat tricks about multimodal data preprocessing
2.4B parameters
DiT model, consisting of 24 layers based on stable audio
3.2k GPU h800 hours
very cool work check it out
2
23
307
15,717
coming to the stores near you
@shourshot solved AR
25
11
291
35,579
- prefix caching
- flash attn
- torch compile
is all you need to almost beat vllm btw
16
6
290
20,236
imo being cracked means having personal taste in tools you use and mastering them
him using his preferred sets of tools and performing at his max potential is what makes him cracked
2
2
261
19,428
