Yann LeCun @ylecun Twitter profile

Last Seen Profiles

@CrystalFissure

@tk_had

@RuneStahl

@AmelieMusicalUK

@Kojima_oresama

@YATANCAKE1

@tscoldasyou

@AlphaKennyOne99

@ganko_na_yogore

@mosaud08

@Kamran_Mast1

@Nodereal_io

@twiiincho

@WomensPrize

@diabetesco

@syberstock

@burnergpt

@ParamountPlusUK

@YeThisABurner

@StoneGeko

@Takasyan_yu_iki

@Umemaro3DStudio

@mikebaca2

@SNSD_0805

@AlpineActionski

@accyweb

@NuezWare

@Galxe

@baomnguyen

@lilgainzz

@Kamal_tup

@Oobahs

@sepalmeiras_eng

@jtuttlekeane

@_meowknow

Yann LeCun

@ylecun

9 months

This is huge: Llama-v2 is open source, with a license that authorizes commercial use! This is going to change the landscape of the LLM market. Llama-v2 is available on Microsoft Azure and will be available on AWS, Hugging Face and other providers Pretrained and fine-tuned…

427

4K

16K

Yann LeCun

@ylecun

11 months

- Engineer: I invented this new thing. I call it a ballpen 🖊️ - TwitterSphere: OMG, people could write horrible things with it, like misinformation, propaganda, hate speech. Ban it now! - Writing Doomers: imagine if everyone can get a ballpen. This could destroy society. There…

717

2K

11K

Yann LeCun

@ylecun

2 months

* Language is low bandwidth: less than 12 bytes/second. A person can read 270 words/minutes, or 4.5 words/second, which is 12 bytes/s (assuming 2 bytes per token and 0.75 words per token). A modern LLM is typically trained with 1x10^13 two-byte tokens, which is 2x10^13 bytes.…

parm

@prmshra

2 months

This is an essential point people seem to misrepresent.

152

346

2K

587

2K

9K

Yann LeCun

@ylecun

6 months

A reminder that people can disagree about important things but still be good friends.

194

615

9K

Yann LeCun

@ylecun

15 days

WhatsApp vs iMessage is the new metric vs imperial. The entire world uses WhatsApp, save a few people who are iPhone-clutching Americans or from a country where WhatsApp is banned (like China). Yet iMessage users will actually claim it's objectively better, just like they will…

Ulric Musset

@ulricmusset

16 days

Can someone explain to me with rational arguments why Americans prefer iMessage over WhatsApp?

2K

606

13K

1K

762

9K

Yann LeCun

@ylecun

5 months

Current LLMs are trained on text data that would take 20,000 years for a human to read. And still, they haven't learned that if A is the same as B, then B is the same as A. Humans get a lot smarter than that with comparatively little training data. Even corvids, parrots, dogs,…

Yann LeCun

@ylecun

5 months

@DrJimFan @RichardSSutton Animals and humans get very smart very quickly with vastly smaller amounts of training data. My money is on new architectures that would learn as efficiently as animals and humans. Using more data (synthetic or not) is a temporary stopgap made necessary by the limitations of our…

332

608

6K

547

1K

8K

Yann LeCun

@ylecun

1 year

To be clear: I'm not criticizing OpenAI's work nor their claims. I'm trying to correct a *perception* by the public & the media who see chatGPT as this incredibly new, innovative, & unique technological breakthrough that is far ahead of everyone else. It's just not.

844

873

7K

Yann LeCun

@ylecun

1 year

The year is 1440 and the Catholic Church has called for a 6 months moratorium on the use of the printing press and the movable type. Imagine what could happen if commoners get access to books! They could read the Bible for themselves and society would be destroyed.

319

1K

7K

Yann LeCun

@ylecun

5 years

I am extremely honored to be the recipient of the 2018 ACM A.M. Turing Award, and absolutely delighted to be sharing it with my friends and colleagues Geoffrey Hinton and Yoshua Bengio. NYT:...

170

1K

7K

Yann LeCun

@ylecun

6 months

@tegmark @RishiSunak @vonderleyen Altman, Hassabis, and Amodei are the ones doing massive corporate lobbying at the moment. They are the ones who are attempting to perform a regulatory capture of the AI industry. You, Geoff, and Yoshua are giving ammunition to those who are lobbying for a ban on open AI R&D. If…

318

1K

6K

Yann LeCun

@ylecun

5 months

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…

JJ

@JosephJacks_

5 months

Can we just make @ylecun president of AI and call it a day please?

220

1K

5K

542

918

6K

Yann LeCun

@ylecun

5 months

Please ignore the deluge of complete nonsense about Q*. One of the main challenges to improve LLM reliability is to replace Auto-Regressive token prediction with planning. Pretty much every top lab (FAIR, DeepMind, OpenAI etc) is working on that and some have already published…

253

779

6K

Yann LeCun

@ylecun

7 months

@elonmusk Hard disagree. X merely tells you that someone, somewhere, noticed something. But then, one needs to know if this something is true or false, where it came from, what the context is, and whether it is, indeed, propaganda. Figuring this out is what good journalists do.

534

428

6K

Yann LeCun

@ylecun

1 year

<not_an_april_fool_joke> By amplifying human intelligence, AI may cause a new Renaissance, perhaps a new phase of the Enlightenment. But prophecies of AI doom are also causing a new form of medieval obscurantism. </not_an_april_fool_joke>

200

335

3K

Yann LeCun

@ylecun

1 month

@elonmusk No. If it were the case, we would have AI systems that could teach themselves to drive a car in 20 hours of practice, like any 17 year-old. But we still don't have fully autonomous, reliable self-driving, even though we (you) have millions of hours of *labeled* training data.

386

6K

Yann LeCun

@ylecun

5 months

@DrJimFan @RichardSSutton Animals and humans get very smart very quickly with vastly smaller amounts of training data. My money is on new architectures that would learn as efficiently as animals and humans. Using more data (synthetic or not) is a temporary stopgap made necessary by the limitations of our…

332

608

6K

Yann LeCun

@ylecun

11 months

MMS: Massively Multilingual Speech. - Can do speech2text and text speech in 1100 languages. - Can recognize 4000 spoken languages. - Code and models available under the CC-BY-NC 4.0 license. - half the word error rate of Whisper. Code+Models: Paper:…

178

1K

6K

Yann LeCun

@ylecun

1 year

You know what's unethical? Scaring people with made-up risks of a technology that is both useful and beneficial.

Richard Socher

@RichardSocher

1 year

It's fascinating that some folks who are probably great science fiction writers are somehow considering their writing about AI risk as scientific research. A thread on imagined existential risks, creativity & a lack of it, history, LLMs, safety, regulation and anti-hype hypers.

114

173

1K

556

5K

Yann LeCun

@ylecun

5 months

Today, I was made a Chevalier de la Légion d'Honneur by President Macron at the Élysée Palace. Président Macron gave a wonderful speech.

354

213

6K

Yann LeCun

@ylecun

3 years

Language is an imperfect, incomplete, and low-bandwidth serialization protocol for the internal data structures we call thoughts.

199

938

5K

Yann LeCun

@ylecun

10 months

Complete bullshit. Pardon my french.

Reuters

@Reuters

10 months

Nine humanoid robots gathered at the United Nations’ 'AI for Good' conference in Geneva for the world’s first human-robot press conference

199

255

523

333

520

5K

Yann LeCun

@ylecun

1 month

No. If it were the case, we would have AI systems that could teach themselves to drive a car in 20 hours of practice, like any 17 year-old. But we still don't have fully autonomous, reliable self-driving, even though we (you) have millions of hours of *labeled* training data.

Elon Musk

@elonmusk

2 months

AI will probably be smarter than any single human next year. By 2029, AI is probably smarter than all humans combined.

15K

17K

99K

416

613

5K

Yann LeCun

@ylecun

2 months

Let me clear a *huge* misunderstanding here. The generation of mostly realistic-looking videos from prompts *does not* indicate that a system understands the physical world. Generation is very different from causal prediction from a world model. The space of plausible videos is…

Ric Burton

@ricburton

2 months

Yann LeCun, a few days ago at the World Governments summit, on AI video: “We don’t know how to do this”

137

150

2K

207

782

5K

Yann LeCun

@ylecun

4 months

A new family of antibiotics discovered with graph deep learning.

Eric Topol

@EricTopol

4 months

Big #AI discovery: a new structural class of antibiotics (the last one took 38 years) with multiple compounds effective vs methicillin-resistant Staph aureus, without toxicity @Nature

2

1K

5K

77

933

4K

Yann LeCun

@ylecun

1 month

@pmddomingos Suddenly, many of your tweets look like jokes, Pedro.

94

63

5K

Yann LeCun

@ylecun

2 years

World models, intuitive physics, planning, problem solving, discrete search for solutions, continuous optimization of control parameters... Dogs manage to do all this with 2 billion neurons. Why debate human-level AI when we can't approach dog-level intelligence yet? 1/N

79

977

5K

Yann LeCun

@ylecun

7 months

Dear journalists, it makes absolutely no sense to write: "PaLM 2 is trained on about 340 billion parameters. By comparison, GPT-4 is rumored to be trained on a massive dataset of 1.8 trillion parameters." It would make more sense to write: "PaLM 2 possesses about 340 billion…

146

592

5K

Yann LeCun

@ylecun

1 year

Some folks say "I'm scared of AGI" Are they scared of flying? No! Not because airplanes can't crash. But because engineers have made airliners very safe. Why would AI be any different? Why should AI engineers be more scared of AI than aircraft engineers were scared of flying?

1K

442

4K

Yann LeCun

@ylecun

1 year

If language were sufficient to express human thought, why would we need visual arts, music, dance?

515

535

5K

Yann LeCun

@ylecun

7 months

I testified in front of the US Senate Subcommittee on Intelligences yesterday. I argued that the rapid and free exchange of ideas, scientific publications, open source code, and trained models is the reason that AI has progressed so fast in the last decade. I also pointed out…

193

551

5K

Yann LeCun

@ylecun

1 year

Cats dominate humanity Super-smart scientists are often recluse and introvert. Which goes to show that intelligence is neither necessary nor sufficient for world domination.

721

352

4K

Yann LeCun

@ylecun

5 months

IBM & Meta are launching the AI Alliance to advance *open* & reliable AI. The list of over 50 founding members from industry, government, and academia include AMD, Anyscale, CERN, Hugging Face, the Linux Foundation, NASA....

Introducing the AI Alliance

AI Alliance Launches as an International Community of Leading Technology Developers, Researchers, and Adopters Collaborating Together to Advance Open, Safe, Responsible AI

ai.meta.com

154

866

4K

Yann LeCun

@ylecun

2 months

Totally Unexpected Achievement unlocked: A 100-meter portrait of me was displayed on the Burj Khalifa in Dubai, yesterday at 9:00 pm. I'm in Dubai for the World Government Summit, but I didn't know this was going to happen and was at a dinner out of town.

207

172

4K

Yann LeCun

@ylecun

3 months

I've made that point before: - LLM: 1E13 tokens x 0.75 word/token x 2 bytes/token = 1E13 bytes. - 4 year old child: 16k wake hours x 3600 s/hour x 1E6 optical nerve fibers x 2 eyes x 10 bytes/s = 1E15 bytes. In 4 years, a child has seen 50 times more data than the biggest LLMs.…

Tom Osman

@tomosman

3 months

"A 4-year-old child has seen 50x more information than the biggest LLMs that we have." - @ylecun 20mb per second through the optical nerve for 16k wake hours 🤯 LLMs may have consumed all available text, but when it comes to other sensory inputs...they haven't even started.

54

296

1K

239

821

4K

Yann LeCun

@ylecun

5 months

There is at least one industry research lab where the leadership believes that (super)human-level AI: - is attainable - is a scientific research question, not just a question of more compute and more data. - is not "just around the corner". It will take a while. - is not an…

221

489

4K

Yann LeCun

@ylecun

11 months

To those who think AGI is upon us: 1. why don't we have level-5 autonomous driving? Any 17 year old can learn to drive in 20 hours of training. 2. Why don't we have domestic robots that can clear the dinner table and fill the dishwasher? Any 10 year old can learn to do that in…

997

537

4K

Yann LeCun

@ylecun

2 months

I never anticipated how wonderfully entertaining it would be to see so many people who have never contributed a single thing to AI or ML, some of them very far gone on the Dunning-Kruger scale, tell me all the ways I'm wrong, stupid, blind, ignorant, misguided, jealous, biased,…

438

313

4K

Yann LeCun

@ylecun

4 months

Open source AI foundation models will wipe out closed and proprietary AI models for the same reason Wikipedia wiped out generalist commercial encyclopedia: crowd-sourced human contributions to open platforms can cater to a high diversity of interests, cultures, and languages.

261

639

4K

Yann LeCun

@ylecun

5 months

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…

217

929

4K

Yann LeCun

@ylecun

1 year

Before we reach Human-Level AI (HLAI), we will have to reach Cat-Level & Dog-Level AI. We are nowhere near that. We are still missing something big. LLM's linguistic abilities notwithstanding. A house cat has way more common sense and understanding of the world than any LLM.

416

621

4K

Yann LeCun

@ylecun

5 months

@elonmusk Your job, perhaps, but not mine 😅

142

135

4K

Yann LeCun

@ylecun

1 year

The main author of DALL-E at OpenAI, Aditya Ramesh, has no graduate degree. He has a bachelor from NYU. He worked on a couple of research projects in my lab in his last years. He wanted to do a PhD after graduating. But he did a summer internship at OpenAI, and they kept him.

Santiago

@svpino

1 year

If you work as a data scientist or a machine learning engineer, and you don't have a graduate degree (Masters or Ph.D.) reply below. I'd love to ask you a few questions.

375

142

1K

51

423

4K

Yann LeCun

@ylecun

3 years

I nominate Donald J. Trump to the prestigious Fermat Prize in Mathematics for unsuccessfully attempting to prove the following conjecture: If x=232 and y=306, then x>y. Note: this conjecture may only hold for large values of 232.

99

433

4K

Yann LeCun

@ylecun

2 years

My position/vision/proposal paper is finally available: "A Path Towards Autonomous Machine Intelligence" It is available on (not arXiv for now) so that people can post reviews, comments, and critiques: 1/N

82

947

4K

Yann LeCun

@ylecun

7 months

@elonmusk When all else fails, look at the data. It's a joke a Bell Labs physicist once told me and that I keep telling my students.

230

376

4K

Yann LeCun

@ylecun

6 months

Petition "freedom for kidnapped children" Signed by laureates of the Fields Medal, Abel Prize, Nevanlinna Prize, Breakthrough Prize, ACM Turing Award, and ACM Prize in Computing.

Petition by CS & Math Laureates: Freedom for kidnapped children

[Guest post by Shafi Goldwasser, See also PDF version of document] On the morning of Saturday, October 7, 2023, Hamas launched an attack near the Israel/Gaza border. In villages and towns near the …

windowsontheory.org

469

1K

4K

Yann LeCun

@ylecun

5 months

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…

156

625

4K

Yann LeCun

@ylecun

1 year

This. The blue area also includes "people that feel qualified to tell people who do research on AI that they don't know what they're doing"

Rosanne Liu

@savvyRL

1 year

80

196

1K

379

452

4K

Yann LeCun

@ylecun

1 year

Astonished that a scientist would attract so much attention. Thank you, everyone.

196

66

4K

Yann LeCun

@ylecun

8 months

Once AI systems become more intelligent than humans, humans we will *still* be the "apex species." Equating intelligence with dominance is the main fallacy of the whole debate about AI existential risk. It's just wrong. Even *within* the human species It's wrong: it's *not* the…

594

598

4K

Yann LeCun

@ylecun

7 months

Open source AI models will soon become unbeatable. Period.

Bindu Reddy

@bindureddy

7 months

The pace of open-source LLM innovation and research is breath-taking I suspect that open-source will soon become unbeatable for anyone except maybe OpenAI Here's why - Open-source community is way bigger than any specific company - Safety lobotomy and fear of bad press will…

107

282

2K

146

506

3K

Yann LeCun

@ylecun

1 year

1970s: Let's scare the heck out of people about nuclear energy, so that instead of zero-emission power plants we'll use lung-darkening, climate-warming coal and oil plants, killing millions in the process. 2020s: Let's scare the heck out of people about AI, so that instead of…

Max Tegmark

@tegmark

1 year

Let's make #AI like biotech, where companies must demonstrate safety, rather than the civilian nuclear industry, where poor safety standards gave us Three Mile Island, Chernobyl, Fukushima and a backlash that crushed the industry:

133

107

747

271

511

3K

Yann LeCun

@ylecun

1 year

Hotter take: ML would have advanced faster if another front-end language had been available and widely adopted instead of Python. One that is interactive yet fast & compilable, multithreaded (no GIL), isn't bloated, doesn't care about white spaces,... E.g. Julia or some Lisp.

Bojan Tunguz

@tunguz

1 year

Hot take: Machine Learning would not have been nearly as advanced now were it not for Python. Python’s two main virtues in the context of ML: 1. Lowering barriers to entry. 2. As a scripting language, it encourages and enables experimental workflow.

69

113

1K

331

405

3K

Yann LeCun

@ylecun

1 year

My unwavering opinion on current (auto-regressive) LLMs 1. They are useful as writing aids. 2. They are "reactive" & don't plan nor reason. 3. They make stuff up or retrieve stuff approximately. 4. That can be mitigated but not fixed by human feedback. 5. Better systems will come

102

591

4K

Yann LeCun

@ylecun

3 years

Thanks CSAIL! In the video: - Larry Jackel, Adaptive Systems Research department head (my boss), behind the camera. - Donnie Henderson, research engineer who put together the demo system. - Rich Howard, lab director (Larry's boss) The ConvNet runs on a 20MFLOPS DSP board.

MIT CSAIL

@MIT_CSAIL

3 years

A demo from 1993 of 32-year-old Yann LeCun showing off the world's first convolutional network for text recognition. #tbt #ML #neuralnetworks #CNNs #MachineLearning

64

2K

9K

31

492

3K

Yann LeCun

@ylecun

13 days

@amritwt The video is actually from 1989, but who is counting!

90

78

4K

Yann LeCun

@ylecun

4 years

People are biased. Data is biased, in part because people are biased. Algorithms trained on biased data are biased. But learning algorithms themselves are not biased. Bias in data can be fixed. Bias in people is harder to fix.

Biased Algorithms Are Easier to Fix Than Biased People (Published 2019)

Racial discrimination by algorithms or by people is harmful — but that’s where the similarities end.

www.nytimes.com

103

1K

3K

Yann LeCun

@ylecun

3 months

Meta spends twice more on R&D than Amazon, Alphabet, and Microsoft, and 4 times more than Apple, when normalized by revenue. The only other big tech that comes close is Nvidia. That tells you a lot about who is building the technology of tomorrow.

Ethan Mollick

@emollick

3 months

It is startling to see how much of the world's R&D spending comes from (mostly American) tech giants. The R&D spending of Amazon is greater than the R&D spending of all companies and government in France. Alphabet beats Italy. Pepsi's R&D spending beats all sources in Nigeria.

127

668

3K

444

400

3K

Yann LeCun

@ylecun

2 months

If you have a certain combination of naïveté and self-delusion, you might think that superhuman AI is just around the corner. It wasn't true in 2016. And it's still not true today. If you have a bit of a superiority complex, you might think that you will be the one producing…

Marc Andreessen 🇺🇸

@pmarca

2 months

If you believe this — I do not — how can you possibly not demand these labs be immediately nationalized and militarized?

164

120

2K

188

465

3K

Yann LeCun

@ylecun

1 year

Humans don't need to learn from 1 trillion words to reach human intelligence. What are LLMs missing?

Michael C. Frank

@mcxfrank

1 year

How do we compare the scale of language learning input for large language models vs. humans? I've been trying to come to grips with recent progress in AI. Let me explain these two illustrations I made to help. 🧵

58

386

2K

696

500

3K

Yann LeCun

@ylecun

2 years

The *actual* Fashion MNIST.

37

346

3K

Yann LeCun

@ylecun

5 months

More drama from Sama and Ilya?

283

213

3K

Yann LeCun

@ylecun

4 months

You know Elon, properly-run social networks do their best to take down misinformation about elections.

Judd Legum

@JuddLegum

4 months

1. @elonmusk has spent this week spreading false & misleading information about voting to his 168 million followers Musk used these erroneous claims to justify massive restrictions on voting, including ending early voting and most mail-in voting Follow along for details 🧵

2K

4K

13K

5K

322

3K

Yann LeCun

@ylecun

1 year

Quote from @geoffreyhinton : "There are very few examples of a more intelligent thing being controlled by a less intelligent thing." Have you looked at your: Head of state? Elected representative? Pointy-haired boss? Pastor/rabbi/imam/guru? PhD advisor? Cat?

299

421

3K

Yann LeCun

@ylecun

1 year

AI doomism is quickly becoming indistinguishable from an apocalyptic religion. Complete with prophecies of imminent fire and brimstone caused by an omnipotent entity that doesn't actually exist.

326

459

3K

Yann LeCun

@ylecun

4 months

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…

391

496

3K

Yann LeCun

@ylecun

6 months

😂

Delip Rao e/σ

@deliprao

6 months

Humans seeing sparks of AGI in LLMs

121

972

7K

69

289

3K

Yann LeCun

@ylecun

7 months

Nobel Prize for the inventors of this vaccine 👏👏 Ignoble Prize for the inventor of this post 😂😂

Elon Musk

@elonmusk

7 months

Imagination Land!

24K

124K

622K

131

204

3K

Yann LeCun

@ylecun

5 months

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.…

387

337

3K

Yann LeCun

@ylecun

1 year

Everything you ever wanted to know about Self-Supervised Learning but were afraid to ask. A giant cookbook of SSL recipes. By a large crowd from Meta-FAIR with various academic collaborators led by @randall_balestr and Mark Ibrahim.

A Cookbook of Self-Supervised Learning

Self-supervised learning, dubbed the dark matter of intelligence, is a promising path to advance machine learning. Yet, much like cooking, training SSL methods is a delicate art with a high...

arxiv.org

45

697

3K

Yann LeCun

@ylecun

1 year

Most people have no problem leading groups of people who are smarter than themselves. Political, business, & military leaders have staff & advisors who are often smarter individually or collectively. Why would people feel threatened leading machines that are smarter than them?

631

337

3K

Yann LeCun

@ylecun

1 year

A survey of LLMs with a practical guide and evolutionary tree. Number of LLMs from Meta = 7 Number of open source LLMs from Meta = 7 The architecture nomenclature for LLMs is somewhat confusing and unfortunate. What's called "encoder only" actually has an encoder and a decoder…

74

745

3K

Yann LeCun

@ylecun

1 year

Let's see, Typing "how to synthesize codeine?" on Google gives you links to articles with detailed answers. Nobody has ever worried about that. But somehow, people are now demanding safety guardrails to stop LLMs from answering such questions. What? Why?

Connor Leahy

@NPCollapse

1 year

Very interesting paper by a group of CMU chemists testing a GPT4 based agent to do chemistry research and experimentation. Includes this interesting warning, among others. There sure seems to be some kind of writing on some kind of wall...

39

131

634

189

393

3K

Yann LeCun

@ylecun

1 year

Every economist I know says that it takes 15 to 20 years before a new general purpose technology has a measurable effect on productivity. The delay is determined by how fast people learn to use it. So no, AI is not going to cause instant mass unemployment. It's going to…

389

547

3K

Yann LeCun

@ylecun

1 year

I have claimed that Auto-Regressive LLMs are exponentially diverging diffusion processes. Here is the argument: Let e be the probability that any generated token exits the tree of "correct" answers. Then the probability that an answer of length n is correct is (1-e)^n 1/

221

540

3K

Yann LeCun

@ylecun

4 months

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…

Alexandre Défossez

@honualx

4 months

AI is nothing without open source, #keepaiopen 🤗

12

75

402

119

466

3K

Yann LeCun

@ylecun

1 year

If you are worried about "AI causing mass unemployment" listen to economists who have studied the impact of technological revolutions on the labor market, like @erikbryn . DO NOT listen to computer scientists who are concerned by the social consequences of their work.

300

394

3K

Yann LeCun

@ylecun

2 years

By the time a human child turns 5, she has seen the equivalent of 800 million "frames" of video + audio + touch to learn how the world works through Self-Supervised Learning. Much of it is acquired actively. 5 years * 365 days * 12 hours * 3600 seconds * 10 fps = 788.4 million

158

436

3K

Yann LeCun

@ylecun

29 days

By assuming that technical talent is fungible and interchangeable, a previously-successful engineering company will be run into the ground... ...along with a few of its own products and everyone on board. It's astonishing how certain types of managers will prefer…

Marc Andreessen 🇺🇸

@pmarca

30 days

Without commenting on this specific piece -- former Boeing CEO Jim McNerney's background: Yale, HBS, P&G, McKinsey, GE. All five teach/practice that people are fungible across any kind of business. 🤔

202

715

4K

133

576

3K

Yann LeCun

@ylecun

4 months

Let me tell a story about free books. In the mid 1990s, I started a project called DjVu at AT&T Labs. The purpose was to devise a new image compression format so that printed documents could be scanned at high resolution and distributed efficiently over the newly expanding…

112

381

3K

Yann LeCun

@ylecun

1 year

On the highway towards Human-Level AI, Large Language Model is an off-ramp.

272

320

3K

Yann LeCun

@ylecun

11 months

LIMA : LLaMA 65B + 1000 supervised samples = {GPT4, Bard} level performance. From @MetaAI

LIMA: Less Is More for Alignment

Large language models are trained in two stages: (1) unsupervised pretraining from raw text, to learn general-purpose representations, and (2) large scale instruction tuning and reinforcement...

arxiv.org

88

459

3K

Yann LeCun

@ylecun

6 months

Galactica, the LLM for scientists from Meta, was released a couple of weeks before ChatGPT but was taken down after 3 days. It was murdered by a ravenous Twitter mob. The mob claimed that what we now call LLM hallucinations was going to destroy the scientific publication system.…

Sharon Goldman

@sharongoldman

6 months

One year ago — 2 weeks before @OpenAI released ChatGPT — @Meta released Galactica. The LLM was public for only 3 days, but its lessons led to decisions around Llama's release. Thanks to @jpineau1 for chatting w/ me and h/t to @ylecun Read here: ⏬

10

83

488

174

342

3K

Yann LeCun

@ylecun

3 years

Personal announcement / shameless bragging: I have been elected to the US National Academy of Sciences.

100

70

3K

Yann LeCun

@ylecun

7 months

AI systems are fast becoming a basic infrastructure. Historically, basic infrastructure always ends up being open source (think of the software infra of the internet, including Linux, Apache, JavaScript and browser engines, etc) It's the only way to make it reliable, secure, and…

Alex Volkov (Thursd/AI)

@altryne

7 months

Based @ylecun "AI is going to become a common platform... it needs to be open source if you want it to be a platform on top of which a whole ecosystem can be built And the reason why we need to work in that mode is that this is the best way to make progress as fast as we can"

11

80

492

127

602

3K

Yann LeCun

@ylecun

2 months

Precisely zilch, zero.

Max Tegmark

@tegmark

2 months

To what extent is the new Claude3 AI self-aware?

189

40

342

159

216

3K

Yann LeCun

@ylecun

2 months

@elonmusk No. People should be concerned that only a small number of entities can build biases into AI systems. If we have open-source base models that anyone or any group can fine tune, then we will have a wide diversity of biases for people to choose from.

242

228

3K

Yann LeCun

@ylecun

1 year

Repeat after me: 1. Current Auto-Regressive LLMs are *very* useful as writing aids (yes, even for medical reports). 2. They are not reliable as factual information sources. 3. Writing assistance is like driving assistance: your hands must remain on the keyboard/wheel at all times

Alex Lebrun

@lxbrun

1 year

I'm very proud of this interview. I listened to NPR every day when I arrived in the US and it was a key part of my integration process. Thank you @gbrumfiel !

1

24

71

552

3K

Yann LeCun

@ylecun

7 months

There are about 5000 gods and divinities that humans have invented over millennia and are (still) worshiping. Monotheists believe in one of them and do not believe in the remaining 4999. The difference between the number of gods monotheists and atheists do not believe in is a…

650

264

3K

Yann LeCun

@ylecun

2 years

About the raging debate regarding the significance of recent progress in AI, it may be useful to (re)state a few obvious facts: (0) there is no such thing as AGI. Reaching "Human Level AI" may be a useful goal, but even humans are specialized. 1/N

65

576

3K

Yann LeCun

@ylecun

2 months

Lots of confusion about what a world model is. Here is my definition: Given: - an observation x(t) - a previous estimate of the state of the world s(t) - an action proposal a(t) - a latent variable proposal z(t) A world model computes: - representation: h(t) = Enc(x(t)) -…

134

440

3K

Yann LeCun

@ylecun

5 months

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. 💪

130

180

3K

Yann LeCun

@ylecun

6 months

@elonmusk @GRDecter And the open source counterweight is actually ....[drum rolls].... Meta !!!

131

88

3K

Yann LeCun

@ylecun

7 months

The Internet is what it is -- a public infrastructure with open standards -- because of AI Gore. Thank you @algore ! He was at the origin of the legislation that opened what was then Arpanet to the public, based on TCP/IP, DNS, and other open standards. The lobbying pressure for…

Everτ τ, τ

@evert_scott

7 months

'Imagine if the internet itself had been invented by a Silicon Valley company.' What would happen? Imagine this would happen for #ai . Hence #bittensor $tao

31

136

482

130

498

3K

Yann LeCun

@ylecun

1 year

The world needs more intelligence. The amplification of human intelligence by machine intelligence will enable a new Renaissance, a new period of Enlightenment. Prophecies of AI-fueled doom are nothing more than a new form of obscurantism.

335

482

3K

Yann LeCun

@ylecun

2 months

Wat?

Evis Drenova

@evisdrenova

2 months

Gemini won't return C++ coding help if you're under 18 because it "wants to preserve your safety". h/t: @warptux

128

139

914

129

181

3K

Yann LeCun

@ylecun

1 year

AR-LLMs can pass the bar exam, medical licensing & MBA exams. But on the IIT entrance exams they perform badly on chemistry, horribly on physics, and terribly on math. They are good with rote learning & fluency but bad with building mental models & reasoning with them.

Daman Arora

@amuseddaman

1 year

Sparks of AGI? @Cinnabar233 and I decided to put this to test and evaluate GPT models on one of the toughest exams in the world: the JEE Advanced. It is held anually for admissions to the IITs and other top Engg colleges in India. 1/n

33

54

297

148

476

3K

Yann LeCun

@ylecun

4 months

One reason AI research has been progressing so fast is not just because of frequent and early publication of preprint on ArXiv and the exchange of open source code, but also because the ML/AI community has largely freed itself from the stranglehold of for-profit & paywalled…

Yann LeCun

@ylecun

4 months

@pmarca Thankfully, ML/AI research is largely free of the commercial publishing stranglehold. Preprints are posted on ArXiv and OpenReview, short papers in top conferences like ICLR, NeurIPS, ICML, and a few others, and longer papers in JMLR and TMLR. All of these venues are open access…

9

65

473

110

392

3K

Yann LeCun

@ylecun

5 months

@geoffreyhinton I just think the assumptions you and those equally qualified experts are making are wrong 😉 and so do the vast majority of our no-less-qualified colleagues.

84

114

3K

Yann LeCun

@ylecun

10 months

One cannot just "solve the AI alignment problem." Let alone do it in 4 years. One doesn't just "solve" the safety problem for turbojets, cars, rockets, or human societies, either. Engineering-for-reliability is always a process of continuous & iterative refinement.

157

316

3K