TimDarcet

1/ This week we released DINOv2: a series of general vision encoders pretrained without supervision. Good out-of-the-box performance on a variety of domains, matching or surpassing other publicly available encoders.

6. The Influence of Kaiming He Xie deeply admires Kaiming He, with whom he worked closely at FAIR. He learned that Kaiming's superpower is his extreme focus and ability to build an impenetrable baseline ("scaffold") before aiming for breakthroughs [02:34:30]. Kaiming taught him that the real research "signals" come from the failures and surprises during the experimental process, not from armchair theorizing [02:08:12]. 7. "Impact" vs. "Understanding" Xie dislikes the word "impact," finding it too ego-driven and aggressive. Instead, he views research through the lens of philosopher Hannah Arendt: the goal of research is "understanding." Publishing a paper is about sharing a profound realization with the world to find resonance and intellectual kinship, rather than just changing the world by force [01:31:27]. 8. Intelligence is More Than Human Language Xie cautions against human arrogance in defining intelligence solely through language. Drawing on evolutionary biology, he argues that building an AI with the survival and physical reasoning skills of a squirrel is actually a much harder problem than building an AI that can write code or pass exams [06:13:05]. 9. The Power of "Research Taste" Xie equates research to filmmaking, where "taste" dictates everything from the problem you choose to solve down to the formatting of the paper. Influenced by Kaiming He (who even gifted him the Buddhist Diamond Sutra to teach him to look past superficial "forms" to find true substance), Xie believes good taste means avoiding crowded, hyped areas in favor of fundamental, eternal problems like representation learning [02:45:03]. 10. The Degradation of Open Academic AI Research Xie expresses concern over how major industrial AI labs have become increasingly closed off, shifting from open academic exploration to secretive commercial competition. This shift strips researchers of their autonomy and turns them into easily replaceable cogs in a massive engineering machine, heavily motivating his decision to start a new, more open research-driven company [05:21:26].

@gabriberton @francoisfleuret I did not say "we could use MLPs if we had infinite compute", I said "we could use MLPs" Granted, you'd have to slightly twist the word to be able to feed text to an MLP

@TimDarcet @francoisfleuret We always knew that right? Sufficiently large MLPs can approximate any function but are too data-hungry and inefficient to be realistically used

If you have an explanation of why the transformer is so successful, here is a rapid sanity check: if it works for a huge MLP ("depth!", "SGD!", "magic of ml!") it's a very insufficient explanation.

@francoisfleuret In vision when ViTs came people were like wow does that mean we can use any arch?? can we use an MLP???? omg it works MLP transformers CNN they alll work omgggg And then everyone went back to the arch they used before, out of habit and well-tuned recipes

@francoisfleuret Counterpoint transformers are just big MLPs and we could probably also use big MLPs

🔎A closer look at Omnilingual No Language Left Behind, the encoder-decoder system presented as part of @AIatMeta new Omnilingual Machine Translation work!🌍 Many say encoder-decoder is dead in the age of decoder-only LLMs but we show it’s not! 📄:ai.meta.com/research/publi… 🧵1/n

Happy to announce that today we released OmniSONAR (tinyurl.com/3j9rn2u8) and OmniMT. In OmniSONAR, we have been able to really push the edge on largely mutlilingual embedding models, where representations across all languages are aligned like never before! 🧵1/n

Happy to share 🌍Omnilingual Machine Translation🌍 In this work @AIatMeta we explore translation systems supporting 1,600+ languages. We show how our models (1B to 8B) can outperform baselines of up to 70B while having much larger language coverage. 📄:ai.meta.com/research/publi…

Our live tissue clearing paper is out in @naturemethods! We achieved optical clearing of mammalian brain tissues without compromising normal neuronal function. Big congrats to @Shigenori774 and our wonderful collaborators! 🎉 nature.com/articles/s4159… (1/10)

@vmfunc @UnderscoreTalk (positively)

@vmfunc @UnderscoreTalk You look exactly like how I imagine everyone who has an anime profile pic to look like

the full recording for my talk @UnderscoreTalk is available and translated! persona, the free internet, the age of agentic hackers, KYC, ID verification, and censorship

@difficultyang ughhh steam does not want my money why can't I get to the buy pageeee

Red alert red alert slaythespire reddit has gone full skonger this is not a drill

@giffmana I didn't know you had it there too

This is such a classic Google move.

26 years ago I created OG Torch. Now I am on X. Samy: whhhatt, you want to call it Torche? Me: yes, why not? Samy: but people will know what it means in French? <after 1 day of name brainstorming> Samy: ok, remove the "e", let's call it Torch. I think it was a good name.

🔸Je lance un appel solennel à tous les journalistes qui me suivent (et il y en a beaucoup). Faîtes passer le mot, s'il vous plaît, n'invitez plus Luc Julia pour venir parler d'IA, vraiment 🙏 Il y a d'excellents spécialistes en France alors qu'on doit se taper Julia à toutes les sauces. C'est pénible, il est complètement dépassé, has-been et fait preuve d'un rassurisme ridicule. Et vous mes abonnés, je compte vraiment sur vous pour partager ce post, qu'il soit visible au maximum. Cet homme fait de la désinformation sur tous les médias où il passe. Préservez-vous. Pitié !

@miniapeur Beta, mu, pi and sigma are the other way around for me

Mathematician training:

@jm_alexia When at Meta? Or before? Also I think you meant Alexandr, with an e it's a fashion designer

In case you didint know, Alexander Wang, the chief AI officer of Meta, did the same as Sam Altman. He sold LLMs for use in the US army. So FYI both Meta and OpenAI are part of the same coin sadly.

Out of all the fascinating concepts in Dune, I really, really did not expect that the Butlerian Jihad would be relevant to real life one day

very underrated technical deep learning channel btw it’s like grade AAA lectures but somewhat chill?

@LagosJanos Ha well played, I had had not predicted it!

@TimDarcet try this >>>t = (1, 2, [30, 40]) >>> t[2] += [50, 60]

Fun question: what does this give out?

@giffmana Yep! And/or never paid much thought to the inner workings

@TimDarcet this is actually a standard gotcha about closures, not just in python. It's also super common in js. The solution pattern is to do `lambda i=i: i`. This tells me you haven't written much functionall code :)