David Picard @davidpicard.eurosky.social · 4hI think it's overly optimistic with the timing. I'm waiting for the last step anytime now 😬 020 2
David Picard @davidpicard.eurosky.social · 10hGuys, I don't think we have the same definition of "outstanding"... 170 7
David Picard @davidpicard.eurosky.social · 16hThe Swedes are in da place! In today's seminar, we're learning 3D with @davnords.bsky.social 181 8
David Picard @davidpicard.eurosky.social · 16hI nearly missed the time when the trees in front of the lab turned red! But here there are, with a different viewpoint than usual 🍂🍁 0151 15
David Picard @davidpicard.eurosky.social · 20/10/2026Also, pretty image! The website for the project is gorgeous, check it out: arijit-hub.github.io/projects/din... Code and models are available: github.com/arijit-hub/d... 040 4
David Picard @davidpicard.eurosky.social · 20/10/2026If you sample from a fixed DINO feature, you can do DINO inversion. Not sure how useful this is, but it's neat nonetheless because it show that the self-condition can be used to steer the generation process 160 6
David Picard @davidpicard.eurosky.social · 20/10/2026Doing so is much faster than the other uses of a pretrained feature extractor, and leads to a massive improvement compared to the baseline (i.e., without DINO) both in latent and pixel flow-matching. 130 3
David Picard @davidpicard.eurosky.social · 20/10/2026During Sampling, you start with the usual Gaussian noise for the image and zeros for the self-condition, and then you rollout the predicted DINO to the next sampling step The predicted features are fairly stable and act like an anchor guiding the image ODE 130 3
David Picard @davidpicard.eurosky.social · 20/10/2026The training is also very simple and is similar to the self-conditioning in RIN (Jabri ICML'23). You do a 1st forward with 0, get your predicted DINO features (detached) and pass them as additional condition to the 2nd forward+backward pass Since there is only 1 ODE, there is only 1 noise schedule 230 3
David Picard @davidpicard.eurosky.social · 16/10/2026Today has a great "Magritte sky", with the deep blue and the white fluffy clouds. 2100 10
David Picard @davidpicard.eurosky.social · 14/10/2026I'm not saying LLMs are made in hell, but the similarity... 2143 14
David Picard @davidpicard.eurosky.social · 08/10/2026What a weird acknowledgment. I've seen companies doing that in the past, but that's a first for a theoretical CS paper by academics. I don't think there is any opinion in that paper. Is it because of Anthropic? (paper: arxiv.org/abs/2506.07977 lowers the complexity of 3SUM and related problems) 160 6
David Picard @davidpicard.eurosky.social · 02/10/2026Big pb with the US servers or what? I keep having that and my networks works perfectly fine. 010 1
David Picard @davidpicard.eurosky.social · 02/10/2026Il faut regarder par catégorie. Le gros de l'exp vient de CS.LG, CS.CV et CS.CL c'est à dire l'IA moderne. 130 3
David Picard @davidpicard.eurosky.social · 01/10/2026outlook (the webapp) is just unusable. It has more bugs than a first year student's assignment. How is this company valued trillions of $? I just cannot understand that. It's pure crap. the worst email app I've ever used, and I used to host my own webmail... 140 4
David Picard @davidpicard.eurosky.social · 01/10/2026One more year, one more card from my mom (Her comment was "it's not a Trabant, but you get the idea". Do I really look like a Trabi? 🙄) 140 4
David Picard @davidpicard.eurosky.social · 01/10/2026The convolution is back in the office! (Yes, this is a straight sheet of paper) 2100 10
David Picard @davidpicard.eurosky.social · 25/09/2026Who else has this visualization engraved in their brain? 360 6
David Picard @davidpicard.eurosky.social · 21/09/2026🏫🎒 First class of the year! Excited to present new material, I hope it'll be mutual. 1201 20
David Picard @davidpicard.eurosky.social · 20/09/2026Drag'n drop figures for illustrating 1-NN. This would take so long without an LLM, I would not do it. I'm like a kid in a candy store with these interactive figures. I like interactive demos soooo much. 3112 11
David Picard @davidpicard.eurosky.social · 17/09/2026Guys, do you have the slightest idea who you are sending this to? I know I have a common name, but I'm pretty sure an LLM would not have included me in the list of recipients after only 1 google search. 220 2
David Picard @davidpicard.eurosky.social · 10/09/2026Excuse-me Claude, but... Where did you get that from? 140 4
David Picard @davidpicard.eurosky.social · 10/09/2026Si seulement ça pouvait percoler dans un peu plus de têtes... (au passage: je me demande si les 100 jours, c'est un truc qui se fait dans d'autres pays) 010 1
David Picard @davidpicard.eurosky.social · 10/09/2026Every year this cartoon becomes more relevant. (the author does not seem to have a bluesky account). 0143 14
David Picard @davidpicard.eurosky.social · 08/09/2026This sentence is terrifying for any institutional organisation relying on this company. It should break immediately many many contracts with somewhat important information. 3614 61
David Picard @davidpicard.eurosky.social · 08/09/2026Remember folks, if it leaves your computer, it's no longer "your data" 39820 98
David Picard @davidpicard.eurosky.social · 01/09/2026Don't get me started on the Poincaré was wrong thread, I don't have time for that! In my notes, I am mostly contrasting those 2 quotes. 010 1
David Picard @davidpicard.eurosky.social · 29/08/2026For reference, this is the real setup from another angle 010 1
David Picard @davidpicard.eurosky.social · 29/08/2026Let's try this. The table is actually split in 2 separated by close to a meter and the camera is positioned to make it appear seamless. 120 2
David Picard @davidpicard.eurosky.social · 26/08/2026I'm sorry but, waaaat? Only 1 week of R-AC discussion but 6 weeks to release the decisions? How? That doesn't make any sense. 080 8
David Picard @davidpicard.eurosky.social · 07/08/2026So apparently our roof has become a tinder hotspot for black redstart (rougequeue noir en français). They're very noisy during court, it seems. Surprisingly curious too. 160 6
David Picard @davidpicard.eurosky.social · 06/08/2026La célèbre station balnéaire de Charme et Chênes 😎 🤭 040 4
David Picard @davidpicard.eurosky.social · 02/08/2026Spam complètement invraisemblable reçu ce matin 💀 Comment peut-on en 2026 1) écrire ce genre de foutaises de boomers au moment où le pays prend en pleine poire la réalité du climat qu'il a construit, 2) avoir le culot de l'envoyer à un chercheur d'une école du ministère de la transition écologique ? 1114 11
David Picard @davidpicard.eurosky.social · 31/07/2026Sur la question du pourquoi les gens et les gouvernements successifs ne prennent pas les décisions qui s'imposent, je rappelle que cet article à presque 20 ans : www.pnas.org/doi/10.1073/... La réponse est alors claire : un égoïsme important doublé d'un niveau catastrophique en calcul de proba. 020 2
David Picard @davidpicard.eurosky.social · 25/07/2026Everything is burning in Europe. May people remember that next time they have to vote. (Link to the map, in French: www.lemonde.fr/les-decodeur...) 0172 17
David Picard @davidpicard.eurosky.social · 24/07/2026Trying to appeal to my gigantic ego, I see... But no, sorry, it doesn't work. 040 4