iiinova — where your digital self keeps evolving

Yacine @yacinemahdid.bsky.social · 20/05/2025

That’s how I start all my YouTube tutorial about the latest deep learning architecture.

120

Yacine @yacinemahdid.bsky.social · 04/05/2025

Started coding a while game with the kids, whew this is fun!

020

Yacine @yacinemahdid.bsky.social · 01/04/2025

I just asked ChatGPT to help me set up the boilerplate for a Python script that make use of their API. 1. Secret is pasted straight into file, no environment management. 2. The code is for a deprecated API. What a vibe.

000

Yacine @yacinemahdid.bsky.social · 01/04/2025

2025 will be the year of linear attention, I feel it.

020

Yacine @yacinemahdid.bsky.social · 01/04/2025

There is an exhilarating feeling in finally understand a whole line of research after a few weeks of study. It’s like a flash of every paper, formula and code seen that just come flooding all at once in its correct form.

110

Yacine @yacinemahdid.bsky.social · 31/03/2025

Most foundational models use softmax attention, which scales quadratically with input length—a major bottleneck. Linear attention has existed since 2020, yet large-scale models rarely use it. Why? minimax-01 finally makes linear attention work at scale. Deep dive here: 📌 youtu.be/iRuvGU-Sk3c

041

Yacine @yacinemahdid.bsky.social · 17/03/2025

I'm back for the weekly deep-learning study session! ✨ Sorry for the month break, was a bit overwhelmed with lots of things at work. I'll try to move around the schedule a bit so that more people in different time zones can attend. 📸 PS: I gave a talk at a conference in February!

100

Yacine @yacinemahdid.bsky.social · 10/03/2025

Lots of confusion out there about what AI Engineering is about. What's an agent, what's a workflow, what's an agentic system, etc. I made this tutorial on the topic packed with information from the latest research from HuggingFace. Check it out over here: youtu.be/UMYKjT9exb4 Enjoy! 🌹

youtu.be

Introduction to AI Agents - Theory and Code

YouTube video by Deep Learning with Yacine

011

Reposted by Yacine

Valeriy M., PhD, MBA, CQF @predict-addict.bsky.social · 25/02/2025

Class imbalance is not a problem per se, the real problems are: *) using incorrect and/or incomplete metrics such as accuracy, ROC AUC and so on.

121

Yacine @yacinemahdid.bsky.social · 25/02/2025

This is the kind of research we need more of:

000

Yacine @yacinemahdid.bsky.social · 25/02/2025

The state of AI/consciousness discourse:

000

Reposted by Yacine

Joseph Howley @illdottore.bsky.social · 12/02/2025

you fucked up a perfectly good computer is what you did. look at it. it's got innumeracy

Screenshot of a twitter post showing that the latest openAI commercial model is better than previous models at doing arithmetic but still cannot reliably produce the correct answer of multiplication problems with values greater than 11 x 11. It's supposed to be impressive I think

1083161624

Yacine @yacinemahdid.bsky.social · 10/02/2025

Wouldn’t it be funny that we never reach AGI because of short term incentive to keep going with Transformers. Then we patch the whole thing left and right to keep the illusion of general intelligence with massive injection of capital. Literally yeeting the AI field in a local minima and digging.

000

Yacine @yacinemahdid.bsky.social · 09/02/2025

This is so well put, must read!

010

Yacine @yacinemahdid.bsky.social · 29/01/2025

Open AI is getting absolutely cooker right now. Crazy how we went from the darling of AI to a company researchers loathe. not a good vibe.

010

Yacine @yacinemahdid.bsky.social · 06/01/2025

The one thing I dislike about current v. of OpenAI is how surface level they are in their research coms. They are hinting big breakthrough, but man look at the landscape. Every competitors around is stacked with billions and PhD. Whatever they are trying to win, won’t be achieved by secrecy.

100

Reposted by Yacine

Vicki @vickiboykis.com · 05/01/2025

how i'd learn machine learning in 2025 if i had to start from scratch: 1. find a log that i initially planned to turn into a table leg 2. make it into a puppet that can walk and talk 3. have the puppet, through a series of adventures, turn into a real boy and realize the true value of friendship

1337630

Reposted by Yacine

Shahab Bakhtiari @shahabbakht.bsky.social · 03/01/2025

Just came across this interesting blog post on the job market for new PhD grads in AI: kyunghyuncho.me/i-sensed-anxiety-and-frustration-at-neurips24 The argument feels pretty reasonable. Here is my take: (1/6) #MLSky #NeuroAI 🧠📈

2122

Reposted by Yacine

Valeriy M., PhD, MBA, CQF @predict-addict.bsky.social · 01/01/2025

Different ways to represent uncertainty. #conformalprediction

112

Reposted by Yacine

Surya Ganguli @suryaganguli.bsky.social · 31/12/2024

Our new paper! "Analytic theory of creativity in convolutional diffusion models" lead expertly by @masonkamb.bsky.social arxiv.org/abs/2412.20292 Our closed-form theory needs no training, is mechanistically interpretable & accurately predicts diffusion model outputs with high median r^2~0.9

513431

Reposted by Yacine

Nithur @nithur.bsky.social · 30/12/2024

Do this in 2025:

2359

Reposted by Yacine

Simon Willison @simonwillison.net · 31/12/2024

Here's my end-of-year review of things we learned out about LLMs in 2024 - we learned a LOT of things simonwillison.net/2024/Dec/31/... Table of contents:

The GPT-4 barrier was comprehensively broken
Some of those GPT-4 models run on my laptop
LLM prices crashed, thanks to competition and increased efficiency
Multimodal vision is common, audio and video are starting to emerge
Voice and live camera mode are science fiction come to life
Prompt driven app generation is a commodity already
Universal access to the best models lasted for just a few short months
“Agents” still haven’t really happened yet
Evals really matter
Apple Intelligence is bad, Apple’s MLX library is excellent
The rise of inference-scaling “reasoning” models
Was the best currently available LLM trained in China for less than $6m?
The environmental impact got better
The environmental impact got much, much worse
The year of slop
Synthetic training data works great
LLMs somehow got even harder to use
Knowledge is incredibly unevenly distributed
LLMs need better criticism
Everything tagged “llms” on my blog in 2024

28648148

Yacine @yacinemahdid.bsky.social · 31/12/2024

100%, this is messed up and will have massive consequences. We’re going to see more and more heavily gated web communities.

000

Reposted by Yacine

Eugene Vinitsky 🍒 @eugenevinitsky.bsky.social · 30/12/2024

It is roughly 10 lines of code to go from 1 GPU to N GPUs with pytorch DDP. Pointing this out so that everyone is aware and doesn't shy away from scaling their code

61056

Yacine @yacinemahdid.bsky.social · 31/12/2024

I’m calling it, AI companies will wreck the internet-search equilibrium of the last two decades. With the arm race of collecting information to train models and keep them fresh, we’re going to see web scraping going out of the gray zone into the firmly black camp. Can’t be sustainable.

110

Yacine @yacinemahdid.bsky.social · 30/12/2024

That’s some gent moves right there. Unsung heroes of the modern age.

010

Reposted by Yacine

YY Ahn @yyahn.bsky.social · 29/12/2024

www.youtube.com/watch?v=UGO_... A super cool explanation of recent LLM mechanistic interpretability studies from anthropic etc. from Welch Labs (an awesome channel btw). see also "The Dark Matter of Neural Networks?" by @colah.bsky.social transformer-circuits.pub/2024/july-up...

youtube.com

The Dark Matter of AI [Mechanistic Interpretability]

YouTube video by Welch Labs

1388

Reposted by Yacine

Ted Underwood @tedunderwood.com · 30/12/2024

Something I don't understand is: why can't LLMs write novel-length fiction yet? They've got the context length for it. And new models seem capable of the multi-hop reasoning required for plot. So why hasn't anyone demoed a model that can write long interesting stories? I do have a theory ... +

4820030

Reposted by Yacine

Itai Yanai @itaiyanai.bsky.social · 28/12/2024

Research involves day science and night science, requiring you to become good at both. Day science proceeds in logical steps and specializes in hypothesis testing, but to generate new ideas we pop out to the creative night science world, where leaps among ideas are made possible.

1486

Reposted by Yacine

Valeriy M., PhD, MBA, CQF @predict-addict.bsky.social · 28/12/2024

🚀 **CatBoost Leads in Tabular Data Modeling!** 🚀 A comprehensive study analyzing 300 diverse datasets has reaffirmed CatBoost's dominance in tabular data tasks.

131

Yacine @yacinemahdid.bsky.social · 29/12/2024

LLMs are starting to feel like a thinly veiled conspiracy from cloud providers to make unreal amount of cash.

100

Reposted by Yacine

Ido Aizenbud @idoai.bsky.social · 26/12/2024

Human cortical pyramidal neurons are larger, with more elaborate branching, and distinct nonlinear biophysical properties compared to rat cortical pyramidal neurons. Are they more functionally complex? Could that boost the human brain’s computational power? and is that what makes us human? (1/11)

720361

Reposted by Yacine

Scott McGrath @smcgrath.phd · 29/12/2024

🧪 While LLMs have done well in medical skill comparisons with multiple choice questions, a new study finds GPT-4 scored lower than human doctors in free-text assessments of primary care cases, highlighting its limitations in complex clinical reasoning. 🩺🖥️ #MLSky

bmjopen.bmj.com

ChatGPT (GPT-4) versus doctors on complex cases of the Swedish family medicine specialist examination: an observational comparative study

Background Recent breakthroughs in artificial intelligence research include the development of generative pretrained transformers (GPT). ChatGPT has been shown to perform well when answering several s...

78420

Reposted by Yacine

Man in Business Suit Levitating @dfeldman.org · 25/12/2024

You’re still arguing about tabs vs. spaces? May I present…

Code written with box characters used on old old software to make fake UIs

15652621266

Reposted by Yacine

Itai Yanai @itaiyanai.bsky.social · 25/12/2024

It’s crazy how talking with a science buddy is simultaneously what’s most likely to move the project forward and the thing that’s least prioritized in our schedule.

411732

Yacine @yacinemahdid.bsky.social · 25/12/2024

Been practicing fundamental math these days and boy is it fun. Feels like a child playing with toys at Christmas. 🎄

000

Reposted by Yacine

Phillip Isola @phillipisola.bsky.social · 10/12/2024

At NeurIPS this year my lab is sharing a few papers and talks. They are all about the following question: how to characterize the geometry of deep learning problems, and in particular how to measure *distance*? Each paper/talk gives a rather different answer, detailed below:

3555

Reposted by Yacine

Glen Berseth @glenberseth.bsky.social · 23/12/2024

#NeurIPS2024 wrapped up last week. I put together a curated reading list for #DeepRL and #reinforcementlearning work. (represents my interests). Talks and workshops: third-crowd-c77.notion.site/NeurIPS2024-... Curated reading list fracturedplane.notion.site/NeurIPS2024-... #Holidayreading

fracturedplane.notion.site

NeurIPS2024 Related RL papers | Notion

Deep RL papers

06914

Reposted by Yacine

Nedjma Ousidhoum @nedjmaou-nlp.bsky.social · 24/12/2024

@camachocollados.bsky.social and I taught ML for #nlp last semester. Here is a list of resources that we shared with the students (sorted by theme) in this blogpost: github.com/nedjmaou/CMT... [Not all students were familiar with coding so it also includes resources for beginners.]

github.com

GitHub - nedjmaou/CMT122-2425-Resources: Resources for CMT122 students (2024-2025).

Resources for CMT122 students (2024-2025). Contribute to nedjmaou/CMT122-2425-Resources development by creating an account on GitHub.

0286

Yacine @yacinemahdid.bsky.social · 23/12/2024

In the long run, Mamba will win over transformer architecture. The law of cool name in AI dictates it.

000

Yacine @yacinemahdid.bsky.social · 13/12/2024

Learning deep learning math can be daunting. So here’s a very simple framework I used to learn the basics: 📌 youtu.be/QzaggHKDhYU Btw it works for all type of maths.

youtu.be

How to Master Deep Learning Math?

YouTube video by Deep Learning with Yacine

000

Reposted by Yacine

bryson @brysonm.bsky.social · 28/11/2024

using AI as a blanket label for an enormous swath of technologies has always been a mistake

2281

Yacine @yacinemahdid.bsky.social · 27/11/2024

Deep learning has to be one of the most successful rebranding ever. Going from no funding to companies raising ridiculous amount based on vague promise for AGI. Insane marketing.

120

Yacine @yacinemahdid.bsky.social · 26/11/2024

A few people asked me how to get started with research as an undergrad. So here is a step by step tutorial on how I did it: youtu.be/EKPm8jVkhJI

youtu.be

How to do Scientific Research in Undergrad?

YouTube video by Deep Learning with Yacine

000

Yacine @yacinemahdid.bsky.social · 21/11/2024

Just finished a deep dive on the segment anything model by the FAIR team at Meta: youtu.be/X2b_BD_LgwU Very interesting methods in this paper which I feel can be reused in many other field, not just computer vision!

youtu.be

Segment Anything Model (SAM) - Foundational Model Deep Dive

YouTube video by Deep Learning with Yacine

010

Yacine @yacinemahdid.bsky.social · 20/11/2024

The jon snow jokes on one of my latest video were unexpected 😂

000

Reposted by Yacine

tittwiht @tittwiht.bsky.social · 12/11/2024

every time I wipe windows off of one of my devices in order to install linux I feel a deep sense of peace and calmness, as if I have banished a demon or evil spirit

1832215