Paul-Alexis DRAY

#AI #NLP 𝙍𝙚𝙨𝙚𝙖𝙧𝙘𝙝 𝙎𝙘𝙞𝙚𝙣𝙩𝙞𝙨𝙩, building pr0duct |
Be kind & human, forget boring stuff, use 🍉 |
Create is the way to shape your future.

[Paul-Alexis DRAY @DrayPAD]

🔔📰

[Thomas Scialom @ThomasScialom]

QuestEval is conceptually very intuitive: the idea is to measure the amount of similarity between two texts A & B as follow:
i) generating questions on A
ii) asking the questions on B
iii) comparing the answers

You can also do it the other way around, see the full schema:

2/

[Sébastien Toth @TothSebastien]

@ShaanVP https://t.co/BHMS1120XB

We provide automatic meeting/calls notes for start ups.

We have customer care desks in our users since day 1.

Notes are generated 100% automatically to save agent times, help product teams & help leaders take decision on data, not intuitions

[Paul-Alexis DRAY @DrayPAD]

"Most engineers assume people can read their mind." 😏.

[Tiago Ramalho @tmramalho]

Research Engineers are the people training and tuning state-of-the-art models like GPT-3, DALL-E, Imagen, Alphafold, etc

I spent 100s of hours coaching and mentoring junior and mid-career REs.

Here’s how you become an RE at a top-tier institution (Google, Meta, OpenAI,…) 🧵

[Paul-Alexis DRAY @DrayPAD]

🙌👏📚 Currently attending to Ph.D. defense: « Natural Language Generation with Reinforcement Learning » of @ThomasScialom 👉 https://t.co/7KOAO6j2m3

[Paul-Alexis DRAY @DrayPAD]

Recent image x text model architectures.

[Rosanne Liu @savvyRL]

A quick thread on "How DALL-E 2, Imagen and Parti Architectures Differ" with breakdown into comparable modules, annotated with size 🧵
#dalle2 #imagen #parti

* figures taken from corresponding papers with slight modification
* parts used for training only are greyed out

[Paul-Alexis DRAY @DrayPAD]

fractal is a tough one, or maybe ... 😏

[Jesse Mu @jayelmnop]

I am announcing the Perverse Scaling Prize: a $1.14 USD prize for tasks which exhibit any of the following scaling curves

[Paul-Alexis DRAY @DrayPAD]

These are both part of the same coin. We need tireless creators, but we also need guys to critique and think about what's been created. It is and will be a perpetual fight between 2 children who love each other more than they think... 😏

[Marcin Junczys-Dowmunt (Marian NMT) @marian_nmt]

NLP community: Yay! Let's democratize NLP!

Also NLP community: Yay! Letters of condemnation!

[Yannic Kilcher 🇸🇨 @ykilcher]

@percyliang Could you please at least link the video in the letter so people can make up their own mind?

[Paul-Alexis DRAY @DrayPAD]

🥲 #PyTorch #NVIDIA #AI #DeepLearning #NLProc

[Paul-Alexis DRAY @DrayPAD]

JAX 👀

[Matthew Lynley @mattlynley]

After losing out to PyTorch, Google is quietly moving to roll out a new AI framework internally called JAX. It's expected to become the underpinning of Google's products, fixing some of TensorFlow's biggest pain points that frustrate Googlers internally https://t.co/ZZ9rKpFUpt

[Paul-Alexis DRAY @DrayPAD]

@ilyasut AGI almost here.

[Paul-Alexis DRAY @DrayPAD]

web3 is dead, long life to web 2+3 🖐️

[Paul-Alexis DRAY @DrayPAD]

Do you know LaMDA ? You should !

[Andrej Karpathy @karpathy]

1) What is LaMDA and What Does it Want? https://t.co/BZmYnDxXZR
2) Interview https://t.co/fgpHpdPTRa

What can be said with confidence imo is that things are about to get a lot weirder because models appear to follow smooth scaling laws and data+model size can still plenty grow.

[Paul-Alexis DRAY @DrayPAD]

@Rainmaker1973 Teach a man how to fish...

[Paul-Alexis DRAY @DrayPAD]

This 👇

[Techmeme @Techmeme]

A week with Dall-E 2, OpenAI's text-to-image AI tool that is in private research beta and feels like a breakthrough in the history of consumer tech (@caseynewton / The Verge)

https://t.co/qdp2t97fln
https://t.co/6IAhlwvPKo

[Paul-Alexis DRAY @DrayPAD]

@gdb Most of the time, easier to start again from scratch.

[Paul-Alexis DRAY @DrayPAD]

BIG bench is finally there !

[Jascha Sohl-Dickstein @jaschasd]

After 2 years of work by 442 contributors across 132 institutions, I am thrilled to announce that the https://t.co/wezEGzDEHt paper is now live: https://t.co/4Yg36EB9Ru. BIG-bench consists of 204 diverse tasks to measure and extrapolate the capabilities of large language models.

[Paul-Alexis DRAY @DrayPAD]

@remilouf Bien joué !

[Paul-Alexis DRAY @DrayPAD]

@wintonARK Hey @wintonARK you should try https://t.co/NwdLNl2AxC, you could have meetings like never before !

[Paul-Alexis DRAY @DrayPAD]

And you, where are you on the #AGI diagram ? 👀