I'm working on super tiny neural networks that perform better than the big ones, it's kinda insane.
14.09.2025 18:54 โ ๐ 18 ๐ 1 ๐ฌ 3 ๐ 2@alexiajm.bsky.social
AI Researcher at the Samsung SAIT AI Lab ๐ฑโ๐ป I build generative models for images, videos, text, tabular data, NN weights, molecules, and now video games!
I'm working on super tiny neural networks that perform better than the big ones, it's kinda insane.
14.09.2025 18:54 โ ๐ 18 ๐ 1 ๐ฌ 3 ๐ 2Exciting news! We're thrilled to announce the appointment of Professor Hugo Larochelle as Mila's new Scientific Director! A deep learning pioneer and former head of Google's AI lab in Montreal, Hugo's leadership will be pivotal in advancing AI for the benefit of all.
mila.quebec/en/news/hugo...
Researcher in 2026
>Use LLMs to generate 20 papers
>Spend 2 weeks clicking approve/reject
>Bill: 20k$ in API credits
>Spend 2 weeks verifying them, 1 is good
>Publish paper
>GitHub issues: Data leakage, bugs; ignore
>post on X: Human coding is obsolete! AI invented new science๐คฏ
๐๐ฅIntroducing Ctrl-Crash: controllable video generation for autonomous driving! SOTA models struggle to generate physically realistic car crashes. We propose an image2video diffusion model with bounding box and crash type control.
Website: anthonygosselin.github.io/Ctrl-Crash-P...
๐งต->
ILYA: "PRETRAINING IS DONE. WE ARE NOW IN THE POST TRAINING ERA."
13.12.2024 22:49 โ ๐ 40 ๐ 2 ๐ฌ 3 ๐ 1I'm finally starting to train video-game generative models! ๐ฎ The data processing took a long time.
12.12.2024 22:17 โ ๐ 10 ๐ 0 ๐ฌ 0 ๐ 0the problem is i want to play every video game and there are only finite hours in a lifetime
01.12.2024 20:57 โ ๐ 20234 ๐ 2440 ๐ฌ 398 ๐ 135On the Xet team at @huggingface.bsky.social we're always looking for ways to move bytes to computer near you as fast as possible.
To do this, we're redesigning the upload and download infrastructure on the Hub. This post describes how, check the thread for details ๐งต
huggingface.co/blog/rearchi...
These offer some GPU access for small groups. All of Mila has access to some of their servers, but its priority queues and getting lots of GPUs is hard.
I remember waiting for a week to get the 4 GPUs needed for training my video diffusion models at the time (from arxiv.org/abs/2205.09853).
This year, I started telling authors that I will likely increase my score if they fix X, Y, Z. If they do what I ask, they get more points instantly.
I wish my own reviewers did that instead of giving 3/10 and then not answering after addressing all their concerns. Do better people!
Any Canadian equivalent?
23.11.2024 15:27 โ ๐ 2 ๐ 0 ๐ฌ 1 ๐ 0A bit more detail for others. Almost anyone can get an Explore allocation by submitting a simple form, which can get you about 5000 GPU hours.
allocations.access-ci.org/project-types
The coocoos are growing.
ETA: AI religion in 5 years.
hello
22.11.2024 18:42 โ ๐ 3 ๐ 0 ๐ฌ 1 ๐ 0