|
@
fchollet
Mountain View, CA
|
|
Deep learning @google. Creator of Keras, neural networks library. Author of 'Deep Learning with Python'. Opinions are my own.
|
|
|
12.868
Tweetovi
|
603
Pratim
|
181.454
Osobe koje vas prate
|
| Tweetovi |
|
François Chollet
@fchollet
|
7 h |
|
If your media treats politics as a constant spectacle, it's not surprising that politics would evolve into a literal reality show
|
||
|
|
||
|
François Chollet
@fchollet
|
14 h |
|
For clarification, the meeting is an online call. You can join from anywhere.
|
||
|
|
||
|
François Chollet
@fchollet
|
15 h |
|
We're having a Keras community meeting this Friday. If you want to make a Keras-related announcement or short presentation (~2 min) at the meeting, please send it to me by email and I'll include you.
|
||
|
|
||
|
François Chollet
@fchollet
|
24 h |
|
Meaning is contextual
|
||
|
|
||
|
François Chollet
@fchollet
|
24 h |
|
Modeling the statistical structure of language absent any context is a bit like modeling the latent space of all airplane cockpits. It will enable you to generate new realistic airplane cockpits via interpolation, but it misses the purpose & the broader context of this interface
|
||
|
|
||
|
François Chollet
@fchollet
|
4. velj |
|
The common thread here is that language is always an interface: the study of language only makes sense in the context of the relationship between an agent and its environment (including other agents), an agent and itself, or many agents as a collective.
|
||
|
|
||
|
François Chollet
@fchollet
|
4. velj |
|
- Language as operating system for the mind (a tool to formulate complex thoughts, to store thoughts, index & retrieve thoughts or memories)
- Language as artistic medium (poetry, etc)
- Language as DNA for society & civilization (tool to store and propagate knowledge, culture)
|
||
|
|
||
|
François Chollet
@fchollet
|
4. velj |
|
So what are the functions of language? On top of my head:
- Language as communication channel (receive and convey information about the world)
- Language as affordance (saying something = action towards a goal)
- Language as social lubricant (chitchat, etc)
|
||
|
|
||
|
François Chollet
@fchollet
|
3. velj |
|
Software at Google is always built from source, it is never downloaded from an external source.
|
||
|
|
||
|
François Chollet
@fchollet
|
3. velj |
|
You can actually select TF 2 at startup: colab.research.google.com/notebooks/tens…
If installing from pypi I would suggest using the latest version (2.1 right now):
!pip install tensorflow --quiet --upgrade
|
||
|
|
||
|
François Chollet
@fchollet
|
3. velj |
|
PyPI downloads for TensorFlow (and its closest competitor, added for scale). Notice how it starts jumping after the release of TF 2.0 late last year (the short gap afterwards is the holiday break)
Up and to the right 📈 pic.twitter.com/Tlltw40hMO
|
||
|
|
||
| François Chollet proslijedio/la je tweet | ||
|
Zack Akil
@ZackAkil
|
2. velj |
|
Play against my first home made #ReinforcementLearning game agent (zackakil.github.io/deep-tic-tac-t…) I built using @TensorFlow + #Keras + #TensorFlowJS + @vuejs!
Genuinely amazed that it seems to work 😅
Big thanks my teachers @Sentdex , @aureliengeron, @gusthema, and @polonglin 👏👏👏 pic.twitter.com/DRACkNmnVY
|
||
|
|
||
|
François Chollet
@fchollet
|
3. velj |
|
Much like an airplane, language has a function. No matter how realistic you make your image of a plane, if it misses the function, you still haven't mastered human flight. pic.twitter.com/UX8LDecrvS
|
||
|
|
||
|
François Chollet
@fchollet
|
3. velj |
|
I'm looking forward to seeing AI agents that:
- Have a model of the world
- Assume others do too
- Use some form of language to update models on each side
... As well as many others things you use language for.
It doesn't matter if their syntax is terrible.
|
||
|
|
||
|
François Chollet
@fchollet
|
3. velj |
|
Statistical mimicry is to language what a plastic toy picturing a bus is to public transportation innovation. It may look right, but it can't perform its function. A horse-drawn cart would be cruder, but could actually move people from A to B.
It's cargo-cult AI.
|
||
|
|
||
|
François Chollet
@fchollet
|
3. velj |
|
If an algorithm has "mastered natural language", I would expect it to be able to do some of the things language is for -- communicating information, receiving information, acting on the world... Not merely output something that statistically sounds like language.
|
||
|
|
||
|
François Chollet
@fchollet
|
2. velj |
|
Refreshingly honest pic.twitter.com/zFn6t5Jrx9
|
||
|
|
||
| François Chollet proslijedio/la je tweet | ||
|
Ed Hawkins
@ed_hawkins
|
30. sij |
|
The last 2019 years of global temperature change climate-lab-book.ac.uk/2020/2019-year… pic.twitter.com/dWYiwY1zIn
|
||
|
|
||
|
François Chollet
@fchollet
|
2. velj |
|
I think really good fiction has the power to remake reality
|
||
|
|
||
| François Chollet proslijedio/la je tweet | ||
|
Ava
@noampomsky
|
1. velj |
|
I feel like a lot of US citizens don’t quite get how hard it is to get an immigrant visa even if you are super qualified—many friends/family have struggled w the H1-B lottery or spent months applying for an O1 or got denied entry even though they have MIT degrees/great jobs/etc
|
||
|
|
||