Levent Sagun (@leventsagun) 's Twitter Profile
Levent Sagun

@leventsagun

🏳️‍🌈 research scientist at FAIR, based in paris (mostly)

ID: 389904534

calendar_today13-10-2011 05:12:40

313 Tweet

878 Followers

3,3K Following

Abeba Birhane (@abebab) 's Twitter Profile Photo

what is the most sensible report/guideline you recently read on the risks of AI and recommendations on how to manage/regulate it???

Eman Abdelhadi (@emanabdelhadi) 's Twitter Profile Photo

University admin were on the wrong side of civil rights. University admin were on the wrong side of Vietnam. University admin were on the wrong side of South Africa. University admin are on the wrong side of Palestine.

Zeerak (زیرک ظلعت) is on Zeerak@ mastodon|bsky (@zeeraktalat) 's Twitter Profile Photo

So excited that mine and @abebab's chapter on AI and decoloniality is now available! elgaronline.com/edcollchap/boo… In the chapter, we focus on the tension between AI and decolonization.

Dieuwke Hupkes (@_dieuwke_) 's Twitter Profile Photo

Overall, our conclusion on whether LLMs learned and successfully use tacit world models remains negative. To allow further exploration, we release our data, including our finetuning dataset and various within- and out-of-distribution test sets. github.com/facebookresear… [8/n]

Sasha Luccioni, PhD 🦋🌎✨🤗 (@sashamtl) 's Twitter Profile Photo

TL;DR? Stuffing generative models into absolutely everything comes with a significant cost to the planet, and we should use fine-tuned models in cases when tasks are well-defined. 👩🏼‍💻 Alternative titles we explored include "InferNO" and "Think before you GPT" 😂

MMitchell (@mmitchell_ai) 's Twitter Profile Photo

Dear tech industry, Instead of having a race for who can put the highest numbers on (awkward) benchmarks, can we have a race on who can implement the best mechanisms for data consent?

Özgür Sevgi Göral (@sevgigoral) 's Twitter Profile Photo

Güney Afrika’nın İsrail’e karşı Uluslararası Adalet Divanı’nda açtığı dava üzerine, Türkçe'de gördüğüm en etraflı değerlendirmelerden biri. Aynı zamanda, hukuk sahasının politik mücadeleler tarafından kullanımına dair de nefis tartışma. Özlem'le Duru'nun eline sağlık💜Duru Yavan

Karen Hambardzumyan (@mahnerak) 's Twitter Profile Photo

[1/7] 🚀 Introducing the Language Model Transparency Tool - an open-source interactive toolkit for analyzing Transformer-based language models. We can't wait to see how the community will use this tool! github.com/facebookresear…

[1/7] 🚀 Introducing the Language Model Transparency Tool - an open-source interactive toolkit for analyzing Transformer-based language models.

We can't wait to see how the community will use this tool!
github.com/facebookresear…
Adina Williams (@adinamwilliams) 's Twitter Profile Photo

Our responsible AI team is hiring 3 research scientist interns this cycle (2 in Montreal, one in NYC). We're seeking enrolled PhD students who are excited to spend their summer figuring out how to ensure vision and/or language models work for everyone! metacareers.com/jobs/532549086…

Dieuwke Hupkes (@_dieuwke_) 's Twitter Profile Photo

New deep-dive into evaluation data contamination 😍🤩. Curious how much contamination there really is in common LLM training corpora, how much that actually impacts benchmark scores and what is the best metric to evaluate that? Read our new preprint! arxiv.org/abs/2411.03923

New deep-dive into evaluation data contamination 😍🤩.

Curious how much contamination there really is in common LLM training corpora, how much that actually impacts benchmark scores and what is the best metric to evaluate that? Read our new preprint! 

arxiv.org/abs/2411.03923
Skyler Wang (@skylrwang) 's Twitter Profile Photo

In this 📢 NEW PUB 📢, Samuel Bell and I trace how machine learning and AI researchers frame the problem of spurious correlations through ways that deviate from the statistical definition of the problem. Follow this thread to find out more... 1/n arxiv.org/abs/2411.04696

Anaelia (Elia) Ovalle (@ovalle_elia) 's Twitter Profile Photo

📢New paper! When LLMs are aligned towards "harmlessness", who are they actually safe for? For gender minorities, pre-existing patterns of LLM exclusion don't just persist, they can amplify through alignment - measurable in both LLM outputs & implicit reward signals! 🧵1/10

📢New paper! When LLMs are aligned towards "harmlessness", who are they actually safe for?

For gender minorities, pre-existing patterns of LLM exclusion don't just persist,  they can amplify through alignment - measurable in both LLM outputs & implicit reward signals!

🧵1/10
Alican Uludağ (@alicanuludag) 's Twitter Profile Photo

10 Nisan'da Selahattin Demirtaş, 52 yaşına girecek. 4 Kasım 2016'da tutuklandığında yalnızca 43 yaşında, rakiplerine göre genç bir siyasetçiydi. Tam 8,5 yıldır cezaevinde tutuluyor. Demirtaş'ın büyük kızı, babası tutuklandığında 12 yaşındaydı, bugün 21 oldu. Küçük kızı, 9

10 Nisan'da Selahattin Demirtaş, 52 yaşına girecek. 4 Kasım 2016'da tutuklandığında yalnızca 43 yaşında, rakiplerine göre genç bir siyasetçiydi. Tam 8,5 yıldır cezaevinde tutuluyor.

Demirtaş'ın büyük kızı, babası tutuklandığında 12 yaşındaydı, bugün 21 oldu. Küçük kızı, 9
Krunoslav Lehman Pavasovic (@krunolehman) 's Twitter Profile Photo

1/ Happy to share my first accepted paper as a PhD student at Meta and École normale supérieure | PSL which I will present at ICLR 2026: 📚 Our work proposes difFOCI, a novel rank-based objective for ✨better feature learning✨ In collab with David Lopez-Paz, Giulio Biroli and Levent Sagun!

Polina Kirichenko (@polkirichenko) 's Twitter Profile Photo

Excited to release AbstentionBench -- our paper and benchmark on evaluating LLMs’ *abstention*: the skill of knowing when NOT to answer! Key finding: reasoning LLMs struggle with unanswerable questions and hallucinate! Details and links to paper & open source code below! 🧵1/9

Excited to release AbstentionBench -- our paper and benchmark on evaluating LLMs’ *abstention*: the skill of knowing when NOT to answer!

Key finding: reasoning LLMs struggle with unanswerable questions and hallucinate!

Details and links to paper & open source code below!
🧵1/9