ZDi
zdi1908.bsky.social
ZDi
@zdi1908.bsky.social
🇦🇷 I make machines (and myself) learn. Backpropagation, C++ enthusiast.
Currently doing ML speech synthesis (and general DL) research @ my bedroom

🌐: https://zdisket.github.io/page.html
🖋: https://zdtech.substack.com/
I am Napoleon
September 6, 2025 at 6:38 PM
August 13, 2025 at 12:23 AM
Mister, Miss, Mixxer. Because of woke, DJs now have their own gender.
August 8, 2025 at 5:13 PM
WHAT IS THIS Y AXIS?? 52.8 IS HIGHER THAN 69.1, WHICH IS ALSO THE SAME AS 30.8 FOR SOME REASON.
THIS CHART IS A HATE CRIME!!!!
August 7, 2025 at 6:13 PM
They're chaining me to a bed and doing this to me tonight.
August 2, 2025 at 5:49 AM
Monthly inflation in Argentina has reached 1.6%
July 14, 2025 at 7:27 PM
Tesla FSD when I ask it to drive me to Will Stancil's house
July 12, 2025 at 4:56 AM
People on Xitter have discovered that xAI's newly released Grok 4, when asked on the Israel/Palestine conflict, seeks out Elon's opinion
July 10, 2025 at 10:33 PM
lets fucking go (I should update my LinkedIn). Been talking to some AMD AI top brass and one said he'd give me connections. Maybe I'll get a job at AMD
July 4, 2025 at 10:59 PM
I am now a ROCm Star. Now to wait for my name to be added to the list.
June 22, 2025 at 6:18 PM
Apple just reinvented Aero. We must RETVRN to the Windows 7/Vista days.
June 10, 2025 at 8:10 AM
People on xitter found out you can tell Google search you are an animal and the AI overview will believe you
June 7, 2025 at 2:58 AM
Never kys yourself.
May 30, 2025 at 10:39 PM
After switching the encoder to a pretrained Resnet18, freezing layers and training for 1 epoch, my model can (kind of) drive, having learned from 81k frames in 44 laps of me playing
May 29, 2025 at 3:04 AM
I'm supposed to be writing a technical report, but I can't stop testing out my music LSTM (tech demo for my approach to language modeling audio). Only 18M parameters btw
May 24, 2025 at 3:30 AM
Small bikes are fun too.
May 11, 2025 at 3:17 AM
Audio language modeling has always involved people training models to VQ audio directly. But what if we quantized mel spectrograms, then trained a vocoder like iSTFTNet, and later our AR prior on mel spectrogram indices?
We can language model 44.1KHz audio with a single 1k codebook.
May 3, 2025 at 8:10 PM
I tried the thing with ChatGPT where you can make up animal names and it will create pics of what it thinks they look like
May 3, 2025 at 4:47 AM
A payroll company named Deel got sued for espionage. The lawyer claims that building "payroll engines" is an "extremely difficult task", and goes on about mainframes, manufacturing parts, then claims to process payroll on a global scale you need "significant advances in quantum computing"
April 28, 2025 at 12:58 AM
Today I learned that some researchers evaluated how different LLMs could run a vending machine business. Claude 3.5 forgot to check its inventory, believing the goods were not delivered, and over 77 days began escalating threats, ultimately sending 1-second notices and nuclear threats
April 22, 2025 at 1:06 AM
I grilled some short ribs earlier today.
April 21, 2025 at 1:44 AM
one of my favorite passtimes is telling chatgpt of events that just happened and having it straight up not believe me.
April 19, 2025 at 12:02 AM
Got access to an MI300X 🔥
April 17, 2025 at 7:52 PM
Damn, this is good stuff.
March 30, 2025 at 8:51 PM
I threw some art into GPT-4o's native image gen and had it edit and give laser eyes.
March 27, 2025 at 5:25 AM