JFPuget 🇺🇦🇨🇦🇬🇱 @JFPuget

Machine Learning at @Nvidia, 6x Kaggle Grandmaster CPMP. ENS Ulm alumni. ML PhD. Ex ILOG CPLEX, IBM. Views are my own. Blocking ad hominem attacks. bsky.app/profile/jfpuge… France Joined March 2012

Tweets

31K
Followers

17K
Following

2K
Likes

33K

JFPuget 🇺🇦🇨🇦🇬🇱 @JFPuget

8 hours ago

Are people rediscovering bidirectional encoders, as was BERT from the start?

TuringPost @TheTuringPost

2 days ago

Are people rediscovering bidirectional encoders, as was BERT from the start?

8 34 281 61K 205

Download Image

2 0 7 1K 0

Je n'ai pas lu le jugement de Mr le president Sarkozy mais une chose me frappe. Les memes qui denoncent une justice trop laxiste, par exemple Mr Sarkozy et Mme Le Pen, sont les memes qui hurlent car la peine de Mr Sarkozy est trop lourde. Les juges appliquent la loi. Si celle ci…

0 1 13 1K 1

Filip Piekniewski🌻 🐘:@[email protected] @filippie509

2 days ago

Wait so they actually need to hire people? Can't replace them with PhD level AI that is writing 90% of all code about now? Also why can't CNBC ask such simple questions instead of just pumping totally illogical nonsense.

7 16 121 5K 8

Download Image

Jürgen Schmidhuber @SchmidhuberAI

3 days ago

@madiator Nonsense. AlexNet (2012) was similar to DanNet (2011) which won 4 image recognition contests before AlexNet (first superhuman performance in 2011). See people.idsia.ch/~juergen/DanNe…

9 21 340 92K 78

JFPuget 🇺🇦🇨🇦🇬🇱 @JFPuget

2 days ago

Have been saying this all along: The prompt is often more important than the system's architecture.

Santiago @svpino

3 days ago

Have been saying this all along: The prompt is often more important than the system's architecture.

30 15 118 39K 93

0 0 6 1K 1

JFPuget 🇺🇦🇨🇦🇬🇱 @JFPuget

2 days ago

Interesting meet between two world that were disjoint till now: - GPU accelerated computation (e.g. AI, Graphics) - Mathematical optimization

NVIDIA AI @NVIDIAAI

3 days ago

Interesting meet between two world that were disjoint till now: - GPU accelerated computation (e.g. AI, Graphics) - Mathematical optimization

5 3 44 7K 4

Download Image

1 2 22 2K 6

Kaggle @kaggle

3 days ago

🏈 The 8th annual @NFL Big Data Bowl is here! This year, we’re excited to launch two competitions in collaboration with NFL. Participants will use NFL player tracking (Next Gen Stats) to generate insights that enhance the game. Pro tip: you can compete in both 😎 More info 👇

2 26 99 27K 50

Tanishq Mathew Abraham, Ph.D. @iScienceLuvr

5 days ago

SimpleFold: Folding Proteins is Simpler than You Think "we introduce SimpleFold, the first flow-matching based protein folding model that solely uses general purpose transformer blocks. Protein folding models typically employ computationally expensive modules involving…

29 137 1K 275K 965

Download Image

JFPuget 🇺🇦🇨🇦🇬🇱 @JFPuget

4 days ago

I lost a bit track of llm activity because I was busy in a Neurips comp (no llm there). But one thing strikes me: models seem more and more specialized by task. Like in last Qwen announces. The era of one super model, like GPT4, seem over. Even if a system has a single entry…

5 1 40 4K 2

JFPuget 🇺🇦🇨🇦🇬🇱 @JFPuget

4 days ago

Very happy to get a top 10 gold medal in Ariel Neurips competition on @kaggle . I teamed with @kagglingdieter which was great, as usual. We learned a lot, but lacked a couple of weeks to finish our work! kaggle.com/competitions/a…

5 5 57 4K 7

Download Image

Das Lab @RDasLab

4 days ago

The results are in: top codes in Stanford #RNA 3D Folding @kaggle are competitive with CASP16-leading humans Vfold, beat AlphaFold 3. Top team’s trick was template-based modeling, not #DeepLearning. Congrats: john, odat, Eigen, + all 1706 participants! kaggle.com/competitions/s…

1 13 73 22K 23

Download Image

Cameron!! @cameronmattis

6 days ago

i can't believe this shit actually works

703 11K 188K 9.0M 32K

Download Image

elvis @omarsar0

6 days ago

Everything that AI hypermarketers claimed was going to be dead by now is very much alive: • CS degree — do it if you can • Coding — learn it, period • RAG & prompt engineering — roughly all you need • Evals — no pain, no gain • Wrapper startup — pay close attention to your…

27 70 498 111K 364

Kaggle @kaggle

6 days ago

We’re excited to see what you’ll build to recognize behaviors in mice, uncover new insights into animal social structures, and push behavioral science research forward. 👉 kaggle.com/competitions/M…

Ann Kennedy @Antihebbiann

a week ago

1 9 24 17K 8

Download Image

1 6 34 10K 8

JFPuget 🇺🇦🇨🇦🇬🇱 @JFPuget

6 days ago

Most have seen the news already, but maybe not the official NVIDIA statement. And no, I didn't know about this before today!

NVIDIA Newsroom @nvidianewsroom

6 days ago

Most have seen the news already, but maybe not the official NVIDIA statement. And no, I didn't know about this before today!

45 183 830 133K 73

Download Image

1 0 6 2K 0

Peyman Milanfar @docmilanfar

7 days ago

it's called encoder-decoder for a reason

12 84 992 44K 98

Download Image

JFPuget 🇺🇦🇨🇦🇬🇱 @JFPuget

7 days ago

Very interesting share from my team: @ChrisDeotte @The0Viel @0verfit @Giba1 ) Lesson from competitive ML on @kaggle. Kaggling is not all you need to do to become great data scientists or ML engineers But it is extremely useful for improving your modeling skills. Hope that…

Gilberto Titericz Jr @Giba1

a week ago

8 117 707 174K 844

3 9 114 14K 62

JFPuget 🇺🇦🇨🇦🇬🇱 @JFPuget

7 days ago

That's good news for anyone working on swe-bench. Contamination was a real issue in swe-bench verified. I hope this new benchmark lives to the promise!

Bing Liu @vbingliu

a week ago

That's good news for anyone working on swe-bench. Contamination was a real issue in swe-bench verified. I hope this new benchmark lives to the promise!