I created the Google Cardboard almost ten years ago but quit VR soon after to focus on Machine Learning.
VR is based on a ridiculous misunderstanding.
Let me explain why in this thread.
VR takes the problem the wrong way. It tries to immerse us by fooling our sensory system.
VR is lying to our bodies. That’s why most VR experiences are unpleasant after a couple of minutes.
Reading a book is decoding black symbols on a white page. Our brains decode these abstract symbols, creating a virtual world inside our brains. There is no need for brain implants; we can already read the “Matrix”.
The VR promise is to bring “a sense of presence” to computing.
But the “sense of presence,” seeing things as if they were there, adds little value beyond the wahoo effect it generates.
The misunderstanding is between "immersion" and a "sense of presence."
Someone asked me recently why the Paris AI ecosystem is so 🔥 these days.
On the surface, it looks like Paris became a major
#AI
center overnight, but it didn't.
It took time and didn't appear by accident.
Here is a story that started more than 10 years ago.
👇
Our perceptions fade away when we are immersed. For instance, while reading a book in the tube.
On the other end, VR uses our body's senses as the medium, distracting us from the story, the message, and the real thing we want to be immersed in.
I just found this video of a project I did 5 years ago.
The screen is not displaying what the webcam sees but a prediction of what the webcam will see next.
#NeurIPS2018
is over! Time to wrap up!
In this thread I'll share what I found the most interesting in the field of ML and creativity.
Everything below worth your time IMO!
The depth map allows generating a 2.5D model (almost 3D, but we are only adding depth to the picture plane)
The normal map is then applied to the 2.5D model.
The best thing about ML is simplicity.
This paper explains how to create a speech to text system that works on all languages, without labelled data.
It's fascinating to see that this task can be achieved with a simple concept.
New work: "Unsupervised speech recognition"
TL;DR: it's possible for a neural network to transcribe speech into text with very strong performance, without being given any labeled data.
Paper:
Blog:
Code:
Very proud to work with
@NASA
and release today a new
@googlearts
experiment.
We used NLP to make a semantic maps of the concepts extracted from NASA images metadata.
Check it out:
A normal map can be seen as a derivative of the depth map, but in practice, it adds a lot of information about how light should be reflected or not with much better precision than the depth map.
More info if you are curious:
Putting Google Brain under Deepming is a smart move, but it won't be enough to solve the main issue that Google is facing.
The problem never was research.
1/N
Here a tutorial about Mixture Density Network.
The notebook is highly inspired from
@hardmaru
MDN tutorial, but I adapted it so it can generate mixtures for multidimensional outputs.
@ojoshe
Music don't try to mimic reality the same way a book don't try to imitate reality.
But it's a direct channel to your emotions.
Sound engineering, 3d sound system, or better image quality are just nice to have.
#Dalle
,
#StableDiffusion
, etc.. are amazing to imagine things like avocado chairs, but they can't draw specific objects.
This is a major blocker for those who want to use this tech for advertising a real product.
But this will change.
👇 This thread to explain how.
Today I'm thrilled to launch ✨
A free tool to remove objects and defects from any picture.
💻 Try it:
🤖 Code:
⬆️ PH:
🥰 100% free & open-source thanks to
@clipdropapp
Extraordinary clip from 20 years ago. David Bowie sharing what he feels about what the internet is going to turn into. And he’s not wrong.
Paxman: it’s just a tool tho isn’t it?
Bowie: No it’s not.
We need to get rid of anthropomorphism to reveal it's real power.
ML can do things that no human can do, from signals we can't sense.
2 ex:
Infer a user identity from keystroke:
Infer a user pose by analyzing radio signal:
Hey! I'm curating a list of videos done using ML. Can you please add your findings by replying to this thread?
Please don't be shy and feel free to add your own work.
I'm also curious about the tools used to produce them.
🔥🔥 Today we,
@clipdropapp
are launching our 💡image relighting
#AI
application 💡
The app allows you to apply professional lights to your portrait images 📸 in real time ⚡
Try it now! it is free 🙂
#photography
#MachineLearning
I'm now COO of :-)
I have the privilege to work with the 2 very best co-founder you can think of:
@cyrildiagne
&
@jblanchefr
I can't be more excited by the idea of helping people through machine learning. There are so many things waiting to be done!
Annoyed by shaky videos? Check out our work on stabilizing shaky handheld videos *without cropping*!
Neural Re-rendering for Full-frame Video Stabilization
Project:
Very few people knows that just after the
#deepdream
storm that
@zzznah
initiated, he continued to refine this technique and produced crazy beautiful artwork while the world was invaded by puppies eyes.
I feel incredibly lucky and grateful to have this one at home. 😊
Joelle Pineau talked about reproducibility in reinforcement learning. She gave a a checklist if you want to summit to
#NeurIPS2019
and contribute to solve the reproducibility crisis.
Let's face it: the Walkman and all it's mp3/offline/online based sequels are the very best AR devices you can think of.
The layer added to reality is so powerful, it can transform the worse commute into the best moment of your day.
In 2015,
@ylecun
created FAIR Paris.
(Facebook AI Research, now Fundamental AI Research).
Yann Le Cun is recognized as one of the 3 inventors of deep learning as we know it today.
Deep learning was at least a third French thing from the beginning!
(The other two are Canadians.)
In 2023, building a product is no more spending time in Figma to polish the "UX"
The "UX" is embedded in the model itself.
But a good "UX" won't emerge magically from random data or random refining. It needs to be opinionated.
4/N
"Out Painting" is a technique to extend a picture by guessing what is around.
Here are some examples.
The right image is created by an algorithm from the small image on the left.
How to do that?
It's an entirely new field that is almost unexplored.
There is only one way, the maker way:
Do, try stuff, iterate, learn, fail, learn, repeat, retry.
But always with the product you want to build in mind.
5/N
I created the Google Cardboard and still is a VR enthusiast.
But it's important to keep in mind what is real and what is not.
I'll put my money on a trip to Iceland rather than on a new headset for sure.
In 2013
@Xavier75
created Ecole 42.
This unusual school accepts students from many different backgrounds and trains them to code based on peer-to-peer learning.
(The concept was imported into the state by Lambda School and some others later.)
Each dot is a sentence from Ulysses, James Joyce.
Organized by similarity: Bert + t-SNE.
Color is the sentence position in the book.
It fascinating to see that NLP is able to discover some inner structure in this massive master piece.
Following
@karpathy
recommendation, I just read "Understand"
I loved it.
While most SF authors optimize their writings for video adaptation, Ted Chiang uses the written language medium to convey much more than any video would.
Any recommendations for other books like this?
Large Language Models (LLM) like GPT3 are great for completing a sentence or a paragraph. This is exactly what they have been trained for.
A surprising consequence, they are quite good at question answering.
But what if they can use a calculator? Or consult Wikipedia?
👇
When building the primary Dataset, you must first consider the product you are creating.
The loss of the model should come after that.
Every choice that falls into research when you write a paper needs to be made with the product in mind at every step.
3/N
A novel where the characters have to reprogram themselves, genuinely change their tastes, personality and behavior to escape an AI which is enslaving them.
Stable Diffusion can teleport any product anywhere!
Which one is a real picture, and which ones are fake?
Given the pace of research, do you think you'll be able to tell the difference in one week? ;-)