Abdoulaye Diack
banner
diack.bsky.social
Abdoulaye Diack
@diack.bsky.social
1.8K followers 1.3K following 290 posts
Program Manager ML & AI @ Google Research | Ex-Google Brain. Speaker (FR/EN) Abdoulaye.ai Opinions are my own. He/His A lot of my retweets and likes are for bookmarking purposes. 🌍 Accra, Ghana (cover photo: Dar Es Saalam, circa 2015)
Posts Media Videos Starter Packs
Pinned
We recently published an open source dataset of building footprints and height estimations for the Global South. It covers ~2 billion buildings and 8 years of data derived from Sentinel 2 imagery.

Blog: shorturl.at/S8ejc
Paper: arxiv.org/abs/2310.11622
Dataset: shorturl.at/FHuQE
very elegant way to put it... "strong programming skills (even in cases where support from an LLM is not available)"
Kind of a big deal. We can now monitor and track air pollution by the hour. ESA Sentinel-4 can beam back hourly data on nitrogen dioxide hotspots & sulfur dioxide plumes. This is a massive leap for public health & air quality forecasting. See Northern Italy 😱

www.esa.int/Applications...
Credit: ESA
That's also my assumption. Btw congrats on the "for you" algo, it's so good!
Reposted by Abdoulaye Diack
Reposted by Abdoulaye Diack
A monohedral tiling of the plane by "spandrelized" squares.
Each unit square includes a circular arc of a 1/2-radius circle centered at each vertex.
Adams, Colin. "Spandrelized Tilings." Amer. Math. Monthly 132, no. 3 (2025): 199-217.

doi.org/10.1080/0002...
#MathSky #Mathematics #Geometry #Tiling
Thanks, NVIDIA, for dropping DGX just after I bought a gaming laptop as an AI playground
the winner also focused on data quality instead of quantity
this is the most interesting and hilarious thing i've read in a while
While many teams relied on deep learning, the winning team (jaejohn) surprised everyone with a highly optimized pipeline that revived classic template-based modeling. 👇
www.kaggle.com/competitions...
1st Place Solution | Kaggle
Hybrid TBM + DRfold2 Approach
www.kaggle.com
moral of the story:
AI isn't always the answer
Necessity really is the mother of invention (i recommend reading the solution and the comments)
If you're GPU poor, don't count yourself out. Ingenuity can still win.
Competitor enters a major AI competition (RNA folding)
GPU poor so can't train an AI
Builds a "classic" eng pipeline instead. (90s tech)
Wins and beat everyone using DL 💀
Their winning "hybrid" model had an AI in it. Their original one did not and had a higher score
So they won despite the AI 😂
While many teams relied on deep learning, the winning team (jaejohn) surprised everyone with a highly optimized pipeline that revived classic template-based modeling. 👇
www.kaggle.com/competitions...
1st Place Solution | Kaggle
Hybrid TBM + DRfold2 Approach
www.kaggle.com
Reposted by Abdoulaye Diack
🚀 Global MMLU Lite is now live on Kaggle Benchmarks!

Developed by @cohereforai.bsky.social, it spans 16 languages with both Culturally Sensitive & Agnostic samples - helping researchers uncover cultural & linguistic biases in multilingual evaluation.
Global MMLU Lite Leaderboard | Kaggle
Understanding and Addressing Cultural and Linguistic Biases in Multilingual Evaluation.
www.kaggle.com
Crazy that these are no longer CGI vids.
A year ago these things could just shuffle along slowly and would fall down if you looked at them wrong
Sunflower, Uganda's first multilingual large language model, is designed to translate, summarize, and generate text for 31 Ugandan languages and others across East Africa.
sunflower.sunbird.ai
Sunflower
Multilingual LLM created by Sunbird AI
sunflower.sunbird.ai
Reposted by Abdoulaye Diack
This is so cool. When you look at representational geometry, it seems intuitive that models are combining convex regions of "concepts", but I wouldn't have expected that this is PROVABLY true for attention or that there was such a rich theory for this kind of geometry.
🕳️🐇Into the Rabbit Hull – Part II

Continuing our interpretation of DINOv2, the second part of our study concerns the *geometry of concepts* and the synthesis of our findings toward a new representational *phenomenology*:

the Minkowski Representation Hypothesis
Something tells me I'm going to miss this period of relative calm on bluesky down the road
Reposted by Abdoulaye Diack
I am recruiting PhD students to start in 2026! If you are interested in robustness, training dynamics, interpretability for scientific understanding, or the science of LLM analysis you should apply. BU is building a huge LLM analysis/interp group and you’ll be joining at the ground floor.
Life update: I'm starting as faculty at Boston University
@bucds.bsky.social in 2026! BU has SCHEMES for LM interpretability & analysis, I couldn't be more pumped to join a burgeoning supergroup w/ @najoung.bsky.social @amuuueller.bsky.social. Looking for my first students, so apply and reach out!
Reposted by Abdoulaye Diack
Small models work great for GLAM but there aren't enough examples!

With @wjbmattingly.bsky.social I'm launching small-models-for-glam on @hf.co to create/curate models that run on modest hardware and address GLAM use cases.

Follow the org to keep up-to-date!
huggingface.co/small-models...
Reposted by Abdoulaye Diack
Excited to share SamudrACE, the first 3D AI ocean–atm–sea-ice #climate emulator! 🚀 Simulates 800 years in 1 day on 1 GPU, ~100× faster than traditional models, straight from your laptop 👩‍💻 Collaboration with @ai2.bsky.social and GFDL, advancing #AIforScience with #DeepLearning.
tinyurl.com/Samudrace
SamudrACE: A fast, accurate, efficient 3D coupled climate AI emulator
A fast digital twin of a state-of-the-art coupled climate model, simulating 800 years in 1 day with 1 GPU. SamudrACE combines two leading…
medium.com
Congratulations to all the recipients of the 2025 Google Academic Research Award! This year's awards support research in AI for privacy and security, digital trust and safety, and quantum neuroscience. Learn more about the awardees and their projects here: blog.google/outreach-ini...
Congratulations to the 2025 Google Academic Research Award recipients
Google.org announces the recipients of the 2025 Google Academic Research Awards.
blog.google
Reposted by Abdoulaye Diack
Very excited to be able to talk about something I've been working on for a while now - we're working with Commonwealth Fusion Systems, IMO the leading fusion startup in the world, to take our work on AI and tokamaks and make it work at the frontier of fusion energy. deepmind.google/discover/blo...
Google DeepMind is bringing AI to the next generation of fusion energy
We’re announcing our research partnership with Commonwealth Fusion Systems (CFS) to bring clean, safe, limitless fusion energy closer to reality with our advanced AI systems. This partnership...
deepmind.google
Reposted by Abdoulaye Diack
Chapter 3, and with it the first 176 pages, is now live! (mng.bz/lZ5B)
Reposted by Abdoulaye Diack
Gemma 27B variant discovered a new cancer pathway treatment that has been validated

Scientists setup an environment and context, the model made a novel inference

blog.google/technology/a...
Reposted by Abdoulaye Diack
Google is open-sourcing RISC-V based NPU?

Coral NPU: A full-stack platform for Edge AI

A full-stack, open-source platform designed to address the core performance, fragmentation, and privacy challenges that limit powerful, always-on AI with low-power edge devices and wearables.
Coral NPU: A full-stack platform for Edge AI
research.google
Congrats to the researchers at Yale & my colleagues on Cell2Sentence! A massive open-weight model with source-available code, trained on public data, to help accelerate cancer research.
blog.google/technology/a...
How a Gemma model helped discover a new potential cancer therapy pathway
We’re launching a new 27 billion parameter foundation model for single-cell analysis built on the Gemma family of open models.
blog.google