_giorgio_

Backprop kept producing wins. That bought it time.

“Wait longer” is not a blank check. In 2026, with Meta-scale talent, data, and compute, serious ideas should show strong intermediate results, not just theory.

Time is necessary, but it is not evidence. More compute does not replace insight, but it does speed up falsification.

So no, skepticism is not naive. If a research program still cannot point to a clear empirical advantage after years, “it just needs more time” stops sounding like science and starts sounding like insulation from the scoreboard.

He has coauthored tens of papers on the same subject.

This means sponsorships, millions spent on training etc.

Often I don’t purchase anything because I have to sift through a billion different options.

Model identifiers are often unique to specific stores, because they carry laptop configurations made just for them.

Apple, AmazonBasics, and a few others, by contrast, understand the consumer and offer a very limited—though often configurable—selection.

LeCun has had every advantage imaginable — and the scoreboard remains empty.

He joined Facebook (now Meta) in December 2013. That's over 12 years of access to one of the largest AI labs in the world, near-unlimited compute, and some of the best researchers money can buy.

He introduced I-JEPA in 2023, nearly 3 years ago. It was supposed to represent a fundamental shift in how machines learn — moving beyond generative models toward a deeper, more structured world understanding.

And yet: I-JEPA hasn't decisively beaten existing models on any major benchmark. No Meta product uses JEPA as a core approach. The research community hasn't adopted it — the field keeps pushing on LLMs and diffusion models. There's been no "GPT moment" for JEPA, no single result that made its value obvious to everyone.

So the question becomes simple: how many years, how many resources, and how many failed proof-of-concepts does it take before we're allowed to judge whether an idea actually works?

If his ideas had real substance, we would have seen substantial results by now. He introduced I-JEPA in 2023, so almost three years ago at this point.

If he still hasn’t produced anything truly meaningful after all these years at Meta, when is that supposed to happen? Yann LeCun has been at Facebook/Meta since December 2013.

Your chronological sequence is interesting, but it refers to a time when the number of researchers and the amount of compute available were a tiny fraction of what they are today.

Thinking Machines looks half-dead already.

The giant seed round proves investors were willing to fund Mira Murati, not that the company had built anything durable.

Within months, it had already lost cofounder Andrew Tulloch to Meta, then cofounders Barret Zoph and Luke Metz plus researcher Sam Schoenholz to OpenAI; WIRED also reported that at least three other researchers left. At that point, citing it as evidence of real competitive momentum feels weak.

I can’t reconcile this dichotomy: most of the landmark deep learning papers were developed with what, by today’s standards, were almost ridiculously small training budgets — from Transformers to dropout, and so on.

So I keep wondering: if his idea is really that good — and I genuinely hope it is — why hasn’t it led to anything truly groundbreaking yet? It can’t just be a matter of needing more data or more researchers. You tell me :-D

People where using it to run webservers behind a Mullvad VPN?

It's quite bad not having a port forward when filesharing though.

I haven't tried Ghostty yet, but for a windows user the combo:

wt.exe https://apps.microsoft.com/detail/9n0dx20hk701?hl=en-US&gl=U...

pwsh.exe https://learn.microsoft.com/en-us/powershell/scripting/insta...

is really powerful, it has finally driven me away from conemu, which was really bad with claude (lots of fonts problems).

I can't believe that people can't simply accept gifts.

I have no idea what you're talking about.

I'm just saying that AI critics like to say that they don't like AI, and to prove their point they constantly move up their definition of "good enough", and when and AI reaches that objective, they change their definition of good enough.

Always moving targets.

They never surrender.

It doesn't work for international users.

these links are ok:

https://www.reddit.com/r/cars/comments/1r04f0x/official_ferr...

https://www.ferrari.com/en-EN/auto/ferrari-luce

On android, these don't work: Firefox Chrome Firefox focus :-(

Thanks anthropic

doesn't work.

He should have done more of these simple videos, instead he aimed for a bigger target... it has been two years, and still nothing.

Forever grateful for this series anyway.

Very good arxiv paper, I wish there where some updates on that.

All my notes are in latex.

Any way to use them, or do I have to go through markdown format?

What

Bad, bad architects.

They have a reason for this choice. I remember studying it at university—the professor said that when people have intimacy, there's no need for doors.

But what if you have a guest? And what if your poop stinks?

Incredibly low-IQ people.

Oriol Vinyals said that.

Kindly suggest some books about RL?

I've already studied a lot of deep learning.

Please confirm if these resoruces are good, or suggest yours:

Sutton et al. - Reinforcement Learning

Kevin Patrick Murphy - Reinforcement Learning, an overview https://arxiv.org/abs/2412.05265

Sebastian Raschka (upcoming book)

...

Scaling is not over, there's no wall.

Oriol Vinyals VP of Gemini research

https://x.com/OriolVinyalsML/status/1990854455802343680?t=oC...

They can't solve the ambiguity about YOLO because the article exists only because of that inaccuracy.

Who are you.

I'm not your tutor.

Yes it is a scam because it says that YOLO is ultralytics. That's the wrong point. There are hundreds of different YOLO implementations. Just search for some bookmarks or look in huggingface. Paperswithcode is dead now, hugging face built something similar.

A small percent difference means nothing at all.

Your camera will be very different from the images the models have been trained on, so you'll have to do fine tuning. So the only thing that matters is this phase.

Plus, there are models that run device, on servers etc.

YOLO IS NOT ULTRALYTICS

> The YOLO series is developed and maintained by Ultralytics.

This is totally fake. There are a lot of different YOLO versions, each with a different licence.

This article is a total scam, it should be removed.

Read the first comment. https://www.reddit.com/r/computervision/comments/1gxce90/yol...

Arduino merda.

Among the top 10 tech companies and beyond, they have the most successful open source program.

These projects come to my mind:

SAM segment anything.

PyTorch

LLama

...

Open source datacenters and server blueprints.

the following instead comes from grok.com

Meta’s open-source hall of fame (Nov 2025)

---------------------

Llama family (2 → 3.3) – 2023-2025 >500k total stars · powers ~80% of models on Hugging Face Single-handedly killed the closed frontier model monopoly

---------------------

PyTorch – 2017 85k+ stars · the #1 ML framework in research TensorFlow is basically dead in academia now

---------------------

React + React Native – 2013/2015 230k + 120k stars Still the de-facto UI standard for web & mobile

---------------------

FAISS – 2017 32k stars · used literally everywhere (even inside OpenAI) The vector similarity search library

---------------------

Segment Anything (SAM 1 & 2) – 2023-2024 55k stars Revolutionized image segmentation overnight

---------------------

Open Compute Project – 2011 Entire open-source datacenter designs (servers, racks, networking, power) Google, Microsoft, Apple, and basically the whole hyperscaler industry build on OCP blueprints

---------------------

Zstandard (zstd) – 2016 Faster than gzip · now in Linux kernel, NVIDIA drivers, Cloudflare, etc. The new compression king

---------------------

Buck2 – 2023 Rust build system, 3-5× faster than Buck1 Handles Meta’s insane monorepo without dying

---------------------

Prophet – 2017 · 20k stars Go-to time-series forecasting library for business

---------------------

Hydra – 2020 · 9k stars Config management that saved the sanity of ML researchers

---------------------

Docusaurus – 2017 · 55k stars Powers docs for React, Jest, Babel, etc.

---------------------

Velox – 2022 C++ query engine · backbone of next-gen Presto/Trino

---------------------

Sapling – 2023 Git replacement that actually works at 10M+ file scale

---------------------

Meta’s GitHub org is now >3 million stars total — more than Google + Microsoft + Amazon combined.

---------------------

Bottom line: if you’re using modern AI in 2025, there’s a ~90% chance you’re running on something Meta open-sourced for free.