vesicularia
@PentelEnergel
Followers
404
Following
737
Media
177
Statuses
648
Explore trending content on Musk Viewer
Claudia Sheinbaum
• 430296 Tweets
BB X LISA
• 149020 Tweets
増山さん
• 142343 Tweets
#LISAforBvlgariWatches
• 135886 Tweets
増山江威子さん
• 134225 Tweets
ルパン三世
• 102845 Tweets
峰不二子
• 101500 Tweets
#CDTVライブライブ
• 47269 Tweets
エミリエ
• 44471 Tweets
特殊清掃人
• 39442 Tweets
新キャラ
• 26419 Tweets
不二子ちゃん
• 23077 Tweets
ゲリラ豪雨
• 22397 Tweets
ベンチマーク
• 22062 Tweets
TOYOTA GOOD DEAL X MEW
• 21086 Tweets
土砂降り
• 20349 Tweets
#ISSUE_CONNECT_祝地上波
• 20314 Tweets
Emilie
• 19231 Tweets
더보이즈
• 17255 Tweets
しゅごキャラ
• 16657 Tweets
音楽の日
• 15733 Tweets
ピノコニー
• 15718 Tweets
Alpine
• 14984 Tweets
NORAWIT and KND
• 12775 Tweets
#BOSJ31
• 12190 Tweets
IS:SUE
• 10134 Tweets
スイパラ
• 10078 Tweets
Last Seen Profiles
exactly 100 pages of notes in april
i kinda wanna make a progress bar that detects what i'm doing and starts recording time then displays the running total as a percentage of 10000 hours (pic rel)
6
3
202
desperately need one of these to display training loss or KL divergence
10
7
181
Week 3:
- papers papers papers papers papers
- implementing GPT-2 in mlxGPT
- working through chapter 2 of PRML still and getting the hang of distributions
- implemented LeNet 5
- Started CUDA by example
Geohot streams are great.
Current roadmap:
- Pattern Recognition and Machine Learning
- Train my nanoGPT rewrite in mlx on the Shakespeare dataset; experiment more to build intuition
- Go through the RL lectures by David Silver
- Read 2-3 of the "important papers" every day, on loop
- Implement the papers
2
0
52
4
2
110
It’s gojover I’m past 17 and haven’t published a paper that’s on the cover of Science
2
2
57
Current roadmap:
- Pattern Recognition and Machine Learning
- Train my nanoGPT rewrite in mlx on the Shakespeare dataset; experiment more to build intuition
- Go through the RL lectures by David Silver
- Read 2-3 of the "important papers" every day, on loop
- Implement the papers
Current high level roadmap:
- Learn how to write basic ML in C++
- Attempt DL stuff
- Learn how to write Triton kernels
- Learn how to use CUDA (big step)
- Try writing kernels myself
- Try implementing papers
- Write my own Triton and CUDA architecture
24
25
402
2
0
52
3
2
52
1
2
25
@qtnx_
one of the cool things about ml is that you can literally use it on everything to make it better
reminded me of pmarca
0
0
24
Week 4:
- More paper reading (3/day) and read the code repo implementations
- Almost done with rewrite of nanoGPT in mlx, only training loop left
- Skimming textbooks
- Kaggle AIMO challenge
End of semester college grind doesn't know how to read the room :(
Current roadmap:
- Pattern Recognition and Machine Learning
- Train my nanoGPT rewrite in mlx on the Shakespeare dataset; experiment more to build intuition
- Go through the RL lectures by David Silver
- Read 2-3 of the "important papers" every day, on loop
- Implement the papers
2
0
52
1
0
24
Week 2:
- Coded less, deciding next project
- Read ~5 papers a day
- Getting obliterated by Gaussians in PRML Chapter 2
- Ren AI Circulation
I'm experimenting with different approaches to reading papers, lmk if you have a way that works for yourself specifically :)
Current roadmap:
- Pattern Recognition and Machine Learning
- Train my nanoGPT rewrite in mlx on the Shakespeare dataset; experiment more to build intuition
- Go through the RL lectures by David Silver
- Read 2-3 of the "important papers" every day, on loop
- Implement the papers
2
0
52
1
1
19
Week 6:
- Spent revision time trying to fix the loss on GPT2, terrible idea
- Read a total of 5 papers
- Read Henry Ford’s autobiography (recommend) and Jay Gould’s biography
Final week of exams. I’ve been edged by ml for the past month. I will be unfathomably locked in.
Current roadmap:
- Pattern Recognition and Machine Learning
- Train my nanoGPT rewrite in mlx on the Shakespeare dataset; experiment more to build intuition
- Go through the RL lectures by David Silver
- Read 2-3 of the "important papers" every day, on loop
- Implement the papers
2
0
52
1
0
15
i'm ashamed about the 20-30 commits i added from debugging my html website
2
0
12
> internship at a startup
> ask if the product is a gpt wrapper or something new and novel
> they’re confused
> explain that you probably don’t need a company to run a gpt wrapper
> they laugh and say it’s a good product
> look at product
> it’s a gpt wrapper
Many such cases
1
0
22
@nahrzf
In two years dev YouTubers will be following your example and making dingboard tutorials
0
0
10
You can get cracked at literally anything just by doing the right thing over and over and over again
If you don’t know what the right thing is then keep doing reps slightly differently than before until you hit it
You just have to keep doing repetitions repetitions repetitions
1
0
8
It’s beautiful that there are people who do so much research and go through so much information and just shares the compilations online for free. I love the internet.
For anyone who likes reading about business and companies
0
1
8
i really really really don't like college courses rn it's getting in the way of the me actually learning
it's not that i don't like math, it's that i want to do other math
0
0
9
“One man came to Mozart and asked him how to write a symphony. Mozart replied, “You are too young to write a symphony.” The man said, “You were writing symphonies when you were 10 years of age, and I am 21.” Mozart said, “Yes, but I didn't run around asking people how to do it.””
@qtnx_
Comparing yourself to autistic 16 y/o building apps that rival unicorns is a great way to go on anxiety paranoia speedrun.
Believe in your vision anon and we'll eventually make it !!!
1
1
17
0
1
8
@danielsateler1
GPT-2 is the "hello world" of LLMs I think (there must be a better analogy... err MOS 6502? xv6?), so that's why I started there. And it has a proper paper, weights released and available, and a lot is known about it. At this point it is an artifact of historical significance.
2
11
215
1
0
8
Warren Buffet still calls “Charlie” after he finishes his answer 🥲
70
211
3K
0
0
8
As it often is the case the really obviously hard and unsexy thing is the differentiation between mid and cracked
“The extra mile is always uncrowded”
0
0
7
this is how it feels after reading and clearing all the tabs
0
0
7
i've slept 11pm-7am for the past week and felt dead everyday but the moment i revert to 6am-12pm i am fresh. rejuvenated. nourished.
0
0
7
RL and diffusion algorithms are arcane spells and you can’t convince me otherwise
0
0
6
I don’t read if I’m tired. I feel like I’m doing whoever wrote it a disservice.
0
0
6
@dejavucoder
"Shut up! I-It's not like I use 4chan regularly, it's only because you told me to use the link as a source!"
0
0
5
quote tweet this with your text editor theme & the matching anime character of your choice
45
3
228
2
0
6
@qtnx_
@lelouchdaily
sometimes all you need is a girl to be the only ml guy in your definitely real startup and shoutout your account while spoonfeeding you ml
0
0
5
yacine said something along the lines of "learn primitives not frameworks" a few months ago and that permanently malloced itself onto my brain
0
0
5
i swear i saw a shitpost that looked exactly like this with the same caption a few months ago i'm not crazy
uh.... gpt2-chatbot just solved an International Math Olympiad (IMO) problem in one-shot
the IMO is insanely hard. only the FOUR best math students in the USA get to compete
prompt + its thoughts 🧵
73
173
1K
0
0
4
Would it be good parenting to give a kid an iPad with nothing but Wikipedia
you guys know that you can just go to wikipedia and read things? Like just learn random things entirely for no reason at all?
73
57
869
0
0
3
> 2L Dr Pepper
> twelve arxiv tabs open
> prml textbook
It’s kurisumaxxing time
2
0
3
america wechat
america bilibili
america xhs
america qq
america taobao
0
0
3
the first frieren album is great because it's almost exactly half an hour and i can mentally increment how long i've worked for with zoltraak
1
0
4
DINGBOARD IN DISCORD
0
0
4
Week 5:
- Read Buffett: the making of an American capitalist & The Firm (McKinsey)
- Started The Symbolic Species
- Tried to take a break for exams, didn’t work
- Skimmed a bunch of papers
- Played around with TTS and llama 3, project posting soon
I don’t like exams :(
Current roadmap:
- Pattern Recognition and Machine Learning
- Train my nanoGPT rewrite in mlx on the Shakespeare dataset; experiment more to build intuition
- Go through the RL lectures by David Silver
- Read 2-3 of the "important papers" every day, on loop
- Implement the papers
2
0
52
0
0
4
@amix011
I see it more like a GitHub commit graph type of thing where the action of continuously reading is more important than the total number of books read
The ratio between the books you start and finish is a better representation of your taste
1
0
4
- Appreciation for markets and an ability to recruit non-conventional people (e.g Mike Moritz, because Don liked the way he thought and asked questions; hired Leone for his customer-in perspective of a business)
0
0
1
what if paper authors linked a song to a paper that they either listened to it the most over the duration of writing it or they think fits the paper
0
0
2
Aquascaping/paludariums/terrariums is the best art form you literally have to exhibit your mastery over nature and biological systems just to put your vision into reality
The fact that you can’t force plants to grow the way you want and have to negotiate with them is beautiful
0
0
3
getting completely cooked writing the training loop for gpt-2
skill issue
0
0
3
@shivVor2
It basically takes all the files in a folder generates a summary using gpt4 and is stored in a dictionary then when I search it again uses gpt to look through the summaries and returns the file id (which gets made into the link)
I can also ask it to look through the doc
0
0
2
i've only been using tkinter for the past 30 minutes and i hate working with ui components
2
0
3