Arthur @[email protected] @arthurlec

Vous avez entendu parler du "syndrome de Paris" ? Les touristes japonais qui, dégoûtés par Paris (parce qu'ils croyaient arriver dans Amélie Poulain), sombreraient dans la dépression. Et bien en fait, ça a tout de la légende urbaine.

#Communiqué 🗞️ L’Académie norvégienne des sciences et des lettres attribue le #AbelPrize2024 au Français Michel Talagrand, qui a effectué sa carrière de chercheur au @CNRS. 👏 👉 cnrs.fr/fr/presse/prix… #AbelPrize #TalentsCNRS 🏆 @abel_prize @CNRS_INSMI

I'm super excited to release DiPaCo, a new kind of mixture of experts, that can scale engineering-wise to data centers across the entire world! A few words about it in this thread 🧵

AK @_akhaliq

a month ago

Google presents DiPaCo Distributed Path Composition Progress in machine learning (ML) has been fueled by scaling neural network models. This scaling has been enabled by ever more heroic feats of engineering, necessary for accommodating ML approaches that require high

2 24 161 76K 92

Download Image

Phenomenal evening yesterday @parisai 🇫🇷 The French tech is on the rise and @instadeepai is playing its part! 🚀 Kudos to my fellow speakers, Camille Bouget of @ScientaLab, Gabriel Synnaeve (@syhw) of @AIatMeta and David Pfau (@pfau) of @GoogleDeepMind, for their inspiring…

Diffusion models are surprisingly good at solving algorithmic tasks. With @francoisfleuret and @EvannCourdier, we use discrete diffusion to find shortest paths in mazes represented as images. 1/5

1/7 For the past decade, our team at Meta Reality Labs (previously CTRL-labs) has been dedicated to developing a neuromotor interface. Our goal is to address the Human Computer Interaction challenge of providing effortless, intuitive, and efficient input to computers.

The statistical physics of generative diffusion models is getting hot! As shown in our paper and analyzed here, class separations is a symmetry breaking phase transition. On the other hand, memorization is the result of a transition similar to condensation in glassy systems!

Giulio Biroli @GiulioBiroli

2 months ago

Our paper on dynamical regimes of diffusion models is out arxiv.org/abs/2402.18491 ! A great collaboration with @MarcMezard T. Bonnaire @ValentinDeBort1 analysing generative AI using statistical physics concepts and methods, in particular symmetry breaking & glass transitions.

4 146 745 86K 383

Download Image

Our paper on dynamical regimes of diffusion models is out arxiv.org/abs/2402.18491 ! A great collaboration with @MarcMezard T. Bonnaire @ValentinDeBort1 analysing generative AI using statistical physics concepts and methods, in particular symmetry breaking & glass transitions.

I am really excited to reveal what @GoogleDeepMind's Open Endedness Team has been up to 🚀. We introduce Genie 🧞, a foundation world model trained exclusively from Internet videos that can generate an endless variety of action-controllable 2D worlds given image prompts.

New paper: how to do flow matching on discrete data. Flows give a simple generative framework and better performance than discrete diffusion models. Discrete flows are easily combined with continuous flow matching for multimodal models. arxiv.org/abs/2402.04997 A thread 1/7

"Incremental work". "Limited Novelty". Last year 2359 papers were accepted at CVPR. Impossible that all are breakthroughs. We need to reconsider novelty as the main point driving our reviewing system. The field advances in grand part to rigorous analysis of existing ideas.

The hardest part about working in AI is finding a project that makes sense. The AI landscape is full of BS, this is a status that, if you do not believe in: "we do it because it sells well" or "we do it because we can", makes it hard to find a meaning to your job. It's sad.

for(float e,i,a,w,x,g,h;i++<90.;o+=.01-.02/exp(max(s,e)*3e3)/h){vec3 p=vec3((FC.xy-.5*r)/r.y*g+2.,g);p.zy*=rotate2D(.5);e=p.y;h=e+p.x*.3;p.z+=t;for(a=.6;a>.001;a*=.7)p.xz*=rotate2D(5.),x=(p.x+p.z)/a+t+t,e-=w=exp(sin(x)-3.)*a,h+=abs(dot(sin(p.xz/a*.3)*a,r/r));g+=e=min(e,h*.5-1.);}

Have you ever done a dense grid search over neural network hyperparameters? Like a *really dense* grid search? It looks like this (!!). Blueish colors correspond to hyperparameters for which training converges, redish colors to hyperparameters for which training diverges.

Open source is one of the biggest drivers of progress in software - AI would be unrecognizable without it. However, it is under existential threat from both regulation and well-funded lobby groups. The community needs to defend it vigorously. 🧵

Breaking news: opening of a junior professor tenure-track position in “Statistical Physics and Machine Learning” at the Physics Department of Ecole Normale Supérieure Paris @LP_ENS_ @ENS_ULM More information at phys.ens.fr/fr/article/pos…

Join us for our next meetup on Tues 5 March, feat. talks on research and applications of AI technology from @instadeepai @GoogleDeepMind @ScientaLab @Meta Register/more information--> paris.ai

● In less than 48 hours, over 200+ engineers and designers have reached out to join this project. Over the next six weeks, we are going to collectively build the world's first open-source Language Model Computer (LMC). We will also publish a family of standard protocols to…

killian @hellokillian

4 months ago

so fucking cool. we could make an open-source vers of this pretty quick with a Raspberry Pi + Open Interpreter. fully offline, <$50 for parts. Whisper, Mixtral, Linux, OI, 1 button. what's a good local TTS? coqui? who wants to help? I'm in Seattle, DM me

192 186 3K 571K 1K

wow. seeing this on my timeline below that leaked Rabbit R1 tiktok where it has a 20 second delay @vikhyatk is on the 01 core team. we're building an open-source Rabbit R1. in 5 weeks, a model like this will be running locally in your hands:

vik @vikhyatk

3 months ago

Created a Huggingface Space for moondream1 to make it easier to try out! huggingface.co/spaces/vikhyat…

33 55 372 207K 241

Download Video

Enough’s been written about the break-neck pace of publishing in CS. I want to highlight a related & also problematic aspect: citation patterns. Look at where papers are published, which papers they cite, and where those cited papers are published. You’ll find a closed system.…