text
stringlengths
4
378
likes
stringlengths
0
4
reply
stringlengths
0
309
The business reason why open source AI platforms will win.
1K
Open-source models will destroy ChatGPT and Gemini. The story of open-source Large Language Models is the story of Linux. Windows and Mac won consumers, but Linux became the Internet's operating system. The same will happen with ChatGPT, Gemini, and open-source models. Closed,…
Okay, same article, bigger portrait.
710
At the end of a year dominated by news of AI
A new family of antibiotics discovered with graph deep learning.
4.9K
Big
The new compounds can kill methicillin-resistant Staphylococcus aureus (MRSA) which causes 10,000 deaths per year in the US. Blog post from MIT:
346
Winter is coming in quantum computing ? "Meta’s head of AI research Yann LeCun recently made headlines after pouring cold water on the prospect of quantum computers making a meaningful contribution in the near future. Speaking at a media event celebrating the 10-year anniversary…
An interview of me in Wired with the unequaled Steven Levy. "How Not to Be Stupid About AI, With Yann LeCun" It’ll take over the world. It won’t subjugate humans. For Meta’s chief AI scientist, both things are true. Excerpts: - Steven Levy: In a recent talk, you said, “Machine…
Congratulations Dr Zeming @ebetica Lin
112
I defended my thesis! Big thanks to my committee
Real-time , low-latency, speech-to-speech translation that preserves the voice and expression of the speaker.
2.4K
SeamlessStreaming is an AI translation model that can deliver state-of-the-art results on streaming translation with <2 seconds of latency. One core piece of our latest Seamless Communication research work by teams at FAIR. More on this project
If the practice of medicine could be learned purely from text, training physicians would not require years of residency.
1K
Pretty much everyone in my family is a medical doctor. I often hear there's great intuition in practicing medicine effectively -- a gut feeling about whether a patient is healthy, from many subtle signals. Like riding a bike, medicine won't be automated by AI trained on text.
Here ya go.
470
LeCun's idea that "the good AI's can protect us from the bad AI's" is actually one of our best bets for ensuring a good future imo It is what Dan Hendrycks calls a "Leviathan", a collective of good AI's to prevent rogue AI's from acting poorly. It also means we need models to be…
Tired: military-industrial complex. Wired: national security-effective altruism complex.
421
And then there are people who are simultaneously pro and anti....
221
Pro-AI:
Fun chatting with @EricTopol and @lxbrun about AI and healthcare.
141
@RamziJelassi
AI and AR/VR/smart glasses are converging. A blog post by Meta CTO @boztank . https://about.fb.com/news/2023/12/metas-2023-progress-in-ai-and-mixed-reality/…
61
2023 was a wild year for technological progress. At Meta, our two biggest long-term bets on the technologies of the future both make great leaps forward but more excitingly, they began to converge. I wrote about it here:
Exactly.
552
In AI, the ratio of attention on hypothetical, future, forms of harm to actual, current, realized forms of harm seems out of whack. Many of the hypothetical forms of harm, like AI "taking over", are based on highly questionable hypotheses about what technology that does not…
The whole genesis, history, and inspirations for PyTorch, in which @soumithchintala demonstrates that the free exchange of ideas and code makes software progress faster. Unlike with technical papers, there is no default citation mechanism for code. But citing sources of…
470
PyTorch's design origins, its connection to Lua, its intertwined deep connection to JAX, its symbiotic connection to Chainer The groundwork for PyTorch originally started in early 2016, online, among a band of Torch7's contributors.
Not surprising. But ouch!
669
Good work from
Technologies that empower humans by increasing communication, knowledge, and effective intelligence always cause opposition from people, governments, or institutions who want control and fear other humans. The arguments against open source AI today mirror older arguments against…
1.5K
@primalpoly
The emergence of superhuman AI will not be an event. Progress is going to be progressive. It will start with systems that can learn how the world works, like baby animals. Then we'll have machines that are objective driven and that satisfy guardrails. Then, we'll have machines…
Nice segment on CBS about AI research for the 10th anniversary of Meta's Fundamental AI Research lab (FAIR): the necessity of open research, AI risks and benefits, the future, AI assistants and domestic robots.
137
Amid fears of an artificial intelligence "arms race," some researchers are emphasizing open research.
In-depth interview on CBS Saturday Morning with Brook Silva-Braga, where we discuss the present and future of AI, the benefits and the risks. (and why AI isn't going to kill us but will make all of us smarter)
337
We are hiring!
373
Just realized that the last 10 things, I was impressed by or worked with (projects, papers, frameworks) are all from
You're welcome.
440
Meta is the best thing that has happened to AI
Starting soon
36
This panel will be FUN! Starts in 20min - 3:45pm CST. Stop by or call-in (register to get zoom link)
Testing on the training set.
453
Quick lesson in the dangers of data contamination. Years ago, I came up with an acronym for remembering the periods of the Paleozoic era — “Catastrophic Overthrow Started Different Colder Period”. I was curious if ChatGPT could guess what it stood for. 1/4
Not sure what imaginary universe some AI folks live in.
1.1K
NEW: I spoke to
Like or comment if you agree.
1.3K
@ylecun
It would be a useful exercise to imagine where the AI industry would be today if AI research labs in industry had remained closed, had not open sourced their code, had patented everything and enforced their patents. Imagine a world without PyTorch, and with enforced patents on…
3K
AI is nothing without open source,
Gigantic
1.2K
Poster sessions
Auto-Regressive LLMs have a role to play: turning abstract ideas into token sequences (words, actions, code...). But abstract ideas should be elaborated through planning/reasoning in representation space. AR-LLMs go directly from prompt to answer, skipping the step of reasoning…
1K
i'm skeptical of
Ego-Exo4d: a large dataset of videos of people doing stuff.
733
Together with the Ego4D consortium, today we're releasing Ego-Exo4D, the largest ever public dataset of its kind to support research on video learning & multimodal perception — including 1,400+ hours of videos of skilled human activities. Download
Meta party at #neurips2023 Giant selfie.
619
VICReging with Randall at NeurIPS. https://arxiv.org/abs/2303.00633
212
With Ravid and Ido in front of our NeurIPS poster "reverse engineering Self-Supervised Learning" https://arxiv.org/abs/2305.15614
479
Selfying with the NeurIPS crowd at the Meta booth.
1.4K
"Between extinction and renaissance: what AI can do for us" A fireside chat with my former Meta colleague Jerome Pesenti hosted by the Transatlantic Leaders Forum last month in New York. We talk about AI, why current LLM suck (yet are useful), and what the next steps in AI might…
EU AI Act: it's not over yet. Regulating foundation models is a bad idea that was added late in the text and rightfully fought against by Macron's government.
1.1K
If it were true that raw intelligence was sufficient for a human to want to dominate others and succeed at it, then Albert Einstein, Richard Feynman, Leonard Euler, Niels Abel, Kurt Gödel and other scientists would have been both rich and powerful, and they were neither. 1.…
Excellent NYT article about parisian AI unicorn Mistral, the effervescent AI ecosystem in Paris, the rapidly increasing strength of the open source AI community, the support of open source AI by Meta, and the opposition to it by Google and OpenAI.
371
Ways to get Auto-Regressive LLMs to produce correct answers almost all the time: - test on the training set. - write the answer in the prompt.
NYU Courant Institute authors at #NeurIPS2023
31
Established in 1987 to advance research in artificial intelligence and machine learning, the 36th
Open source AI models are on a path to overtake proprietary models.
1.6K
This is perhaps one of the most important charts on AI for 2024. It was built by the amazing researcher team at
Former AT&T colleagues at #NeurIPS 2002: Front row: Patrick Haffner, Corina Cortes, me, Larry Jackel. Standing: Chris Burges, Yoshua Bengio, Bernhard Schölkopf, Sara Solla, Isabelle Guyon.
422
. @clmt demonstrating his FPGA implementation of convolutional networks doing real-time image recognition at #NeurIPS 2010. Yes, before ConvNets were cool, we had them running on FPGAs.
423
A picture of former AT&T colleagues at #NeurIPS 2006: Léon Bottou, Dan Lee, Michael Littman, Chris Burges, Patrice Simard, Larry Jackel, and me.
333
Je suis signataire de cette tribune.
133
Baisse des compétences en mathématiques et en sciences : « Nous lançons un appel pour que soit mise en œuvre une stratégie nationale ambitieuse »
Audiobox!
771
Starting today you can try our new foundation research model for audio generation. The demo includes Zero shot TTS, Text to sound effects, Infilling and more! Try Audiobox
Thank you.
889
In my humble opinion, among all the big names of
"La société face aux défis de l'intelligence artificielle" Un débat public organisé par la revue européenne Le Grand Continent hébergé à l'ENS. Participants: Anne Bouverot, Marc Mézard, Martin Tisné, Jamal Atif, et moi-même. Ce débat s'inscrit dans la continuité de l'entretien…
What is good and not-so-good about the EU AI Act from @togelius
161
At least from this press release, it seems that the EU AI Act came out less bad than feared. We seem to have avoided any need for licensing or similar for foundation models, and open-source distribution is permitted and seemingly even encouraged.
The EU AI Act negotiations ended. One contentious issue was the regulation of foundation models, particularly open source ones. Kudos to the French, German, and Italian governments for not giving up on open source models. Juicy part: "The legislation ultimately included…
Purple Llama.
256
Announcing Purple Llama — A new project to help level the playing field for building safe & responsible generative AI experiences. Purple Llama includes permissively licensed tools, evals & models to enable both research & commercial use. More details
Telling all the Llama-2 things!
151
This talk by Angela Fan on Llama2 is so good. 30 min, she just tells you all the things.
With @jpineau1 on the Boz To The Future podcast for the 10th anniversary of FAIR.
40
New episode of my podcast out today with AI luminaries
Today, I was made a Chevalier de la Légion d'Honneur by President Macron at the Élysée Palace. Président Macron gave a wonderful speech.
5.5K
Nah, that's merely what the overly optimistic young guns, AGI startup fund raisers, and self-deluded technologists. For most "serious" folks, the correct statement is: "Human-level AI is always 10-20 years away"
1.4K
AGI is always 3-5 years away
MMCR
23
@tedyerxa
EfficientSAM ! You liked SAM (Segment Anything Model)? You're gonna love EfficientSAM! A smaller, faster, and almost as good version of SAM.
749
New paper!
SeamlessExpression: Speech-to-speech translation that preserves the voice and expression of the speaker.
304
SeamlessExpressive, a new AI translation model by research teams at Meta, enables high-quality speech translation that maintains the speaker's vocal style, tone and unique expressions in translated outputs. Try the demo with your own voice
The Cake first appeared in my talk at the NYU future of AI Symposium in early 2016. Here is a video from my 2016 NIPS keynote. The point was that there is much more knowledge to extract about the world using self-supervised learning from high-bandwidth sensory input such as…
270
This also happened in 2016. Whole world was behind end to end RL as the solution to general intelligence after AlphaGo. And
By "not any time soon", I mean "clearly not in the next 5 years", contrary to a number of folks in the AI industry. Yes, I'm skeptical of quantum computing, particularly when it comes to its application to AI.
1.4K
What this NYT article tells us about the genesis of the new AI industry is that there was a lot of drama at Google, DeepMind, OpenAI, and Anthropic. Comparatively, the creation of FAIR was straightforward and drama free.
88
I'm on the cover of the weekly Le Point, as some sort of tiny éminence grise pulling the strings behind @elonmusk , @sundarpichai , @kaifulee and a larger-than-life Sam Altman. The title reads "AI, the battle to control our brains" I just want to say that I have no intention of…
1.3K
Haha! It wasn't quite like that. Many senior members of the CV community were actually very friendly and curious about ConvNets (if skeptical). There was a true intellectual debate. Everyone had to have a "theory" of how vision worked. For me, it was end-to-end gradient-based…
1.3K
Old photo of
The number of industry labs that practice open research and encourage scientists to publish can be counted on the finger of one hand (if that). FAIR is famous for that. This allows scientists to transition to a position in academia if they so decide at some point in their career.…
398
@rsalakhu
Débat public sur l'IA, mardi 5 décembre à l'ENS.
120
Notre prochain mardi réunira Jamal Atif,
Ross Girschick ( @inkynumbers ) is leaving FAIR for AI2, following @sainingxie who joined the NYU faculty (yay!), and @georgiagkioxari who went to Caltech, and Hervé Jégou and @AlexDefosse who went to non-profit Kyutai. It's a loss for FAIR, but I'm happy for them. There is…
SeamlessExpressive: speech-to-speech translation that preserves the voice, the tone, and the expression.
465
Yesterday we introduced SeamlessExpressive — a new model that preserves unique vocal styles & expression for speech translation, built on our SeamlessM4T v2 foundation model. More details on the family of Seamless Communication models
New open source projects, datasets, and demos released for FAIR's 10th anniversary.
328
@AIatMeta
Seamless: speech to text, text to speech, text to text, and speech to speech, transcription and translation in 100 languages. From FAIR.
2.1K
Today we're sharing the next milestone in our Seamless Communication research — a new family of AI translation models that preserve expression and deliver near-real time streaming translations. More on this new work
Ego-Exo4D: huge dataset of videos showing human experts doing stuff. From FAIR.
429
Today, we announce Ego-Exo4D, the largest and most diverse multi-view dataset, showing human experts around the world performing a core set of skilled activities, w/ unprecedented multi-modality, novel new video-language resources, and rich annotations
Basically, I'm a "foundation model" for celebrities. I can be fine-tuned into Harry Potter, Johnny Depp, Snoop Dogg, Taylor Swift, or a slightly overweight version of Mark Zuckerberg.
1.1K
Due to popular demand, I've fused
FAIR is turning 10 The creation of FAIR was announced by Mark Zuckerberg, Mike Schroepfer and me at the NeurIPS conference in early December 2013.
411
10 years of FAIR. 10 years of advancing the state of the art in AI through open research. We're celebrating the 10th anniversary of Meta's Fundamental AI Research team and continuing that legacy by sharing our work on three exciting new research projects today. Details below
FAIR is soon turning 10 years old. 10 years of published AI research. 10 years of open source AI software. Over 900 GitHub repos. https://github.com/facebookresearch…
420
As we approach the 10th anniversary of FAIR,
Cute. Very simple.
161
Optical illusions with diffusion models. There are so many good gifs on this page but honestly I would like several million more.
Okay, this has got to be the absolute best use of text-to-image technology ever
1.2K
Very excited to share LEDITS++ bringing textual image editing to a new level. LEDITS++ is - fast
More drama from Sama and Ilya?
3.3K
Leveraging open source LLMs Mistral-7B and Llama2-70B.
580
We’re thrilled to announce two online LLMs we’ve trained: pplx-7b-online and pplx-70b-online! Built on top of open-source LLMs and fine-tuned to use knowledge from the internet. They are now available via Labs and in a first-of-its-kind live-LLM API.
AI systems are white boxes that we design As such, they are way easier to align than humans and animals which are black-boxes with all kinds of hard-wired objectives that are immutable and may or may not be "aligned", whatever you mean by that.
445
Introducing AI Optimism: a philosophy of hope, freedom, and fairness for all. We strive for a future where everyone is empowered by AIs under their own control. In our first post, we argue AI is easy to control, and will get more controllable over time.
A talk about SSL for PDE solving and identification by @garridoq_ as a preview of our upcoming NeurIPS paper.
49
#AI4Science
VICReg-style SSL for PDE identification and solving. At NeurIPS 2023.
32
We will be presenting our work about Self-Supervised learning on PDEs at NeurIPS in two weeks, but if you can't wait or want more details, come check out our talk tomorrow ! twitter.com/AI4scienceTalk…
Debunking LLM doomerism.
199
@balesni
"Grade A Prime-Cut Bullshit"
67
This is some Grade A Prime-Cut Bullshit Tucker Carlson says it is 100% certain that US government has alien spaceships. Why cover it up? ETI tech is so advanced it would end war & the Pentagon makes too much money on the Military Industrial Complex. And it would solve all our… twitter.com/TuckerCarlson/…
The AI doomers are the "useful idiots" of the proprietary AI industry.
937
@tegmark
The UAE Minister of AI @OmarSAlolama points to a historical precedent of premature technology regulation motivated by fear: the ban of the printing press in 1515 by Sultan Selim I led to the decline of the Ottoman Empire. “We overregulated a technology, which was the printing…
Funny.
186
We are entering a new era where mobile chip performance is measured not by GHz, but by Llama2's token/sec
Good essay on the fallacy of assuming that LLMs are intelligent because they are fluent.
504
In this new Vox piece written with
A version of Llama-2 fine-tuned for medicine from EPFL.
517
Check out our Meditron-7B & 70B
Haha
291
AI alarmists, stop HALlucinating.
What children can do that LLMs can't.
517
@yudapearl
MMCR: a new learning criterion for Self-Supervised training of joint embedding architectures based on concepts and results from statistical physics. By a team from Stanford, NYU, FAIR, and MIT.
548
Excited to begin announcing our
Equivariance to translations implies the use of convolutions *if* we insist on using linear operators. There are lots of non-linear operators that are equivariant.
165
This is the 2nd post in our series on equivariant neural nets. It explains conventional CNNs from a representation theoretic viewpoint and clarifies the mutual relationship between equivariance and spatial weight sharing.
We've barely scratched the surface of the space of deep learning architectures. It's a high dimensional space, so the volume is almost entirely contained in the surface. But we've scratched a tiny subset of the surface.
2.1K
"Simplifying Transformer Blocks" ranks easily among my favorite research papers that I've read this year. Here, the authors look into how the standard transformer block, essential to LLMs, can be simplified without compromising convergence properties and downstream task…
At least a few of these predictions will come true.
326
Six predictions for AI in 2024: - A hyped AI company will go bankrupt or get acquired for a ridiculously low price - Open-source LLMs will reach the level of the best closed-source LLMs - Big breakthroughs in AI for video, time-series, biology and chemistry - We will talk much…
Rescheduled.
154
POSTPONED - AI in Healthcare: The Explainability Dilemma - NEW DATE Our guests are the leading experts in their respective fields, and we understand that these things can happen! Due to a last minute change in the agenda of one of our speakers, the live discussion - AI in…
Popping this up: a response to a question about what I consider reasoning & planning, why current Auto-Regressive LLMs can't do it, why that would require AI systems with world models, and why we still have a lot of progress to do towards AI systems that can learn and reason.
721
@TEDchris
Former French minister @cedric_o responds to @tegmark 's ignominious accusations of corruption surrounding the EU AI Act, and the opposition from the French, German, and Italian governments to regulating foundation model technology. Cédric says, like me, that he has no issue with…
406
@tegmark
But seriously folks, this a short and juicy tirade in which I say: (0) there will be superhuman AI in the future (1) they will be under our control (2) they will not dominate us nor kill us (3) they will mediate all of our interactions with the digital world (4) hence, they will…
6.3K
Can we just make
Haha
740
Can we just make
Reposting this answer to a question from @geoffreyhinton about whether I think LLMs "understand" what they say. I point out what I think is missing from current architectures to reach cat-level intelligence (never mind human level): world models and planning/reasoning abilities.
649
@geoffreyhinton