az-mtl.bsky.social - Profile | ThreadSky | a Reddit-style client for Bluesky

Without fundamental scientific research, there is no data to train any AI models. All the AI breakthroughs in drug discovery (e.g. AlphaFold), medicine, materials and climate change happen thanks to data collected from scientists. No, engineering and scaling is not enough. 😑😑😑

submitted 16 hours ago • 0 comments

SigLIP 2 just came out. 👀 #vlms #clip arxiv.org/abs/2502.14786

submitted 1 day ago • 0 comments

Let’s play “can you find my car?”. #montreal #snowstorm ❄️❄️❄️

submitted 4 days ago • 0 comments

Something special and majestic about the Antwerp train station. First pic looks like it was taken from Harry Potter movies at Hogwarts 😄. #antwerp #anvers #antwerpen

submitted 5 days ago • 0 comments

Rough post-snowstorm day 😭. #montreal #snowstorm

submitted 5 days ago • 0 comments

Quelle expérience d'avoir passé sur le mythique Tout le monde en parle! J'ai vraiment adoré chaque seconde, regardez mon entrevue ici : ici.radio-canada.ca/tele/tout-le... (Et wow, le niveau de glamour est incroyable! I could get used to this...)

submitted 6 days ago • 3 comments

Anyone has a good resource on demystifying GeLU vs GLU vs GeGLU? I’m not up to date with the newest activation functions 😅.

submitted 6 days ago • 0 comments

The coolest thing with diffusion models is when you visualize the denoising steps. Art in motion 🙂‍↔️.

submitted 9 days ago • 0 comments

Throwback Thursday. #hungary #budapest #magyarorszag ❤️🤍💚

submitted 9 days ago • 0 comments

It’s hard to believe this is real life. As a scientist having been involved in biomedical research, it’s just sad and frustrating to see an anti-vax and conspiracy theorist lead the Health department of the most influential country. ☹️

submitted 9 days ago • 0 comments

As we’re preparing to get a huge snowstorm tomorrow in Montreal, I just wish I was living in the West Coast instead 🥲🥲🥲. #vancouver

submitted 10 days ago • 0 comments

😀

submitted 13 days ago • 0 comments

I’m late to the party, but absolutely hooked on Metroid Dread for the Switch. I’m struggling even on rookie mode, but gameplay and visuals are amazing 🤩.

submitted 13 days ago • 0 comments

I still need to learn how to say no 😭.

submitted 14 days ago • 0 comments

The best thing you can do as an ML practitioner in 2025 is to let go of the pressure to keep up with every new paper, focus on what genuinely excites you rather than chasing trends, and embrace the eternal imposter syndrome as part of the journey. 🥲🥲🥲

submitted 15 days ago • 0 comments

Throwback Friday 😅. #brussels #belgium #notAI

submitted 15 days ago • 0 comments

I just finished writing up my take on reasoning models: magazine.sebastianraschka.com/p/understand... Here, I 1. Discuss the advantages & disadvantages of reasoning models 2. Of course, describe and discuss DeepSeek R1 3. Describe the 4 main ways to building & improving reasoning models

submitted 17 days ago • 1 comment

I was hoping that recent events would lead to a mass exodus from X. Many have left, but most of the ML and LLM people have not. I have lost a lot of respect for the ML community.

submitted 18 days ago • 9 comments

What to know about DeepSeek youtu.be/0eMzc-WnBfQ?... In which we attempt to figure out MoE, o1, scaling, tech reporting, modern semiconductors, microeconomics, and international geopolitics.

submitted 18 days ago • 1 comment

Since everyone wants to learn RL for language models now post DeepSeek, reminder that I've been working on this book quietly in the background for months. Policy gradient chapter is coming together. Plugging away at the book every day now. rlhfbook.com/c/11-policy-...

submitted 21 days ago • 2 comments

People: Oh wow you’re doing multimodal alignment! Must be so complex! Us: Yeah we train MLPs to match vision embeddings to text embeddings 😅😅😅.

submitted 21 days ago • 0 comments

You don’t see this often, fair play 😅.

submitted 23 days ago • 0 comments

Throwback Thursday! #strasbourg #alsace #france #notAI

submitted 23 days ago • 0 comments

Today, we are publishing the first-ever International AI Safety Report, backed by 30 countries and the OECD, UN, and EU. It summarises the state of the science on AI capabilities and risks, and how to mitigate those risks. 🧵 Full Report: assets.publishing.service.gov.uk/media/679a0c... 1/21

submitted 24 days ago • 8 comments

Mila peeps: we are opening an internship position in our Applied Research Team. It consists of improving multimodal alignment in foundation VLMs (could lead to publication). Open to external as well (but you need to be based in 🇨🇦). HMU if you want more info.

submitted 24 days ago • 0 comments

The Illustrated DeepSeek-R1 Spent the weekend reading the paper and sorting through the intuitions. Here's a visual guide and the main intuitions to understand the model and the process that created it. newsletter.languagemodels.co/p/the-illust...

submitted 26 days ago • 1 comment

So many people from the AI community secretly support this clown, which is disheartening.

submitted 26 days ago • 0 comments

Last time I was at the Biodôme was 20 years ago when I was a kid. As an adult you get to appreciate more the ecosystem, plants, panels (and not just the animals).

submitted 27 days ago • 0 comments

I know a lot of AI peeps won’t care much about this because as long as they keep their GPUs, they’re happy. But AI4Science is nothing without the science. Expect a huge blow for climate change, medicine and fundamental research in the next 4 years. Science censorship is real.

submitted 29 days ago • 0 comments

Throwback Thursday 😃. Only a few days to go before the Roland Garros 2025 lottery draw opens. #tennis #paris #rolandgarros #nadal

submitted 30 days ago • 0 comments

submitted 32 days ago • 437 comments

100% Same. I need my safe bubble in times like these.

submitted 32 days ago • 0 comments

Social media as we knew it is dead. We are now trapped in the oligarchy of tech giants that spread hate and misinformation. The anti-science propaganda also in full swing. 😢😢😢

submitted 32 days ago • 0 comments

"Masculine Energy"

submitted 34 days ago • 291 comments

Absolutely crazy how GPT4o still hallucinates academic papers. I asked it to give me examples of papers for a niche task, it gave 5 non-existing papers (with credible titles), with hallucinated list of authors and summaries. 😭

Without fundamental scientific research, there is no data to train any AI models. All the AI breakthroughs in drug discovery (e.g. AlphaFold), medicine, materials and climate change happen thanks to data collected from scientists. No, engineering and scaling is not enough. 😑😑😑

SigLIP 2 just came out. 👀 #vlms #clip arxiv.org/abs/2502.14786

Let’s play “can you find my car?”. #montreal #snowstorm ❄️❄️❄️

Something special and majestic about the Antwerp train station. First pic looks like it was taken from Harry Potter movies at Hogwarts 😄. #antwerp #anvers #antwerpen

Rough post-snowstorm day 😭. #montreal #snowstorm

Quelle expérience d'avoir passé sur le mythique Tout le monde en parle! J'ai vraiment adoré chaque seconde, regardez mon entrevue ici : ici.radio-canada.ca/tele/tout-le... (Et wow, le niveau de glamour est incroyable! I could get used to this...)

Anyone has a good resource on demystifying GeLU vs GLU vs GeGLU? I’m not up to date with the newest activation functions 😅.

The coolest thing with diffusion models is when you visualize the denoising steps. Art in motion 🙂‍↔️.

Throwback Thursday. #hungary #budapest #magyarorszag ❤️🤍💚

It’s hard to believe this is real life. As a scientist having been involved in biomedical research, it’s just sad and frustrating to see an anti-vax and conspiracy theorist lead the Health department of the most influential country. ☹️

As we’re preparing to get a huge snowstorm tomorrow in Montreal, I just wish I was living in the West Coast instead 🥲🥲🥲. #vancouver

😀

I’m late to the party, but absolutely hooked on Metroid Dread for the Switch. I’m struggling even on rookie mode, but gameplay and visuals are amazing 🤩.

I still need to learn how to say no 😭.

The best thing you can do as an ML practitioner in 2025 is to let go of the pressure to keep up with every new paper, focus on what genuinely excites you rather than chasing trends, and embrace the eternal imposter syndrome as part of the journey. 🥲🥲🥲

Throwback Friday 😅. #brussels #belgium #notAI

I just finished writing up my take on reasoning models: magazine.sebastianraschka.com/p/understand... Here, I 1. Discuss the advantages & disadvantages of reasoning models 2. Of course, describe and discuss DeepSeek R1 3. Describe the 4 main ways to building & improving reasoning models

I was hoping that recent events would lead to a mass exodus from X. Many have left, but most of the ML and LLM people have not. I have lost a lot of respect for the ML community.

What to know about DeepSeek youtu.be/0eMzc-WnBfQ?... In which we attempt to figure out MoE, o1, scaling, tech reporting, modern semiconductors, microeconomics, and international geopolitics.

Since everyone wants to learn RL for language models now post DeepSeek, reminder that I've been working on this book quietly in the background for months. Policy gradient chapter is coming together. Plugging away at the book every day now. rlhfbook.com/c/11-policy-...

People: Oh wow you’re doing multimodal alignment! Must be so complex! Us: Yeah we train MLPs to match vision embeddings to text embeddings 😅😅😅.

You don’t see this often, fair play 😅.

Throwback Thursday! #strasbourg #alsace #france #notAI

Today, we are publishing the first-ever International AI Safety Report, backed by 30 countries and the OECD, UN, and EU. It summarises the state of the science on AI capabilities and risks, and how to mitigate those risks. 🧵 Full Report: assets.publishing.service.gov.uk/media/679a0c... 1/21

Mila peeps: we are opening an internship position in our Applied Research Team. It consists of improving multimodal alignment in foundation VLMs (could lead to publication). Open to external as well (but you need to be based in 🇨🇦). HMU if you want more info.

The Illustrated DeepSeek-R1 Spent the weekend reading the paper and sorting through the intuitions. Here's a visual guide and the main intuitions to understand the model and the process that created it. newsletter.languagemodels.co/p/the-illust...

So many people from the AI community secretly support this clown, which is disheartening.

Last time I was at the Biodôme was 20 years ago when I was a kid. As an adult you get to appreciate more the ecosystem, plants, panels (and not just the animals).

I know a lot of AI peeps won’t care much about this because as long as they keep their GPUs, they’re happy. But AI4Science is nothing without the science. Expect a huge blow for climate change, medicine and fundamental research in the next 4 years. Science censorship is real.

Throwback Thursday 😃. Only a few days to go before the Roland Garros 2025 lottery draw opens. #tennis #paris #rolandgarros #nadal

100% Same. I need my safe bubble in times like these.

Social media as we knew it is dead. We are now trapped in the oligarchy of tech giants that spread hate and misinformation. The anti-science propaganda also in full swing. 😢😢😢

"Masculine Energy"

Absolutely crazy how GPT4o still hallucinates academic papers. I asked it to give me examples of papers for a niche task, it gave 5 non-existing papers (with credible titles), with hallucinated list of authors and summaries. 😭

My only resolution of 2025 is to fill up this thing, what am I missing? 👀

Throwback Thursday! #munchen #munich #deutschland #germany #notAI