Pierre Dulac @_dulacp

👨‍💻 CTO @ https://t.co/fu4jEQamOP | Ex-Google | Where there's a will, there's a workaround. dulacp.com Paris, Ile-de-France Joined June 2009

Tweets

1K
Followers

421
Following

1K
Likes

6K

ZML @zml_ai

2 years ago

github.com/zml/zml

28 134 820 315K 567

View Details

Pierre Dulac @_dulacp

2 years ago

@steeve ~110 tok/sec on a H100 (fp8 though)

0 0 2 33 0

View Details

Ethan @torchcompiled

3 years ago

When you think about it your DNA is like the latent encoding of a VAE and your upbringing is the stochastic component

2 3 48 61K 17

View Details

Pierre Dulac @_dulacp

2 years ago

@Ligne14_RATP trouvé dans la rame ca40, ça peut dérouter qq touristes :)

1 0 1 406 0

View Details

@PostManHuMain @sainingxie DDPM stands for Denoising Diffusion Probabilistic Models, in the original paper (arxiv.org/abs/2006.11239) they use a UNet to predict the noise at each time-step, whereas in DiT/Sora they use a Visual Transformer (ViT) to replace the UNet part.

0 0 4 213 0

View Details

Pierre Dulac @_dulacp

3 years ago

@icreatelife I used ardi.studio website, a service co-developed with an architect in Paris. I usually need to generate 4 variations to get one interesting.

1 0 1 70 1

View Details

Jim Fan @DrJimFan

3 years ago

Today 6 years ago, "Attention is All You Need" went on Arxiv! Happy birthday Transformer! 🎂 Fun facts: - Transformer did not invent attention, but pushed it to the extreme. The first attention paper was published 3 years prior (2014) and had an unassuming title: "Neural Machine Translation by Jointly Learning to Align and Translate", from Yoshua Bengio's lab. It is a combination of RNN + "context vectors" (i.e. attention). Many of you likely haven't heard about this paper, but it's one of the greatest milestones in NLP and has been cited 29K times (compared to Transformer's 77K). - Neither Transformer nor the original attention paper talked about the general-purpose sequence computer. Instead, both were conceived as solutions to one narrow & specific problem: machine translation. It's remarkable that AGI (some day soon) can trace its origin to the humble Google Translate. 😅 - Transformer was published at NeurIPS 2017, one of the top AI conferences worldwide. Yet it didn't even get an Oral presentation, let alone awards. There were 3 best papers at NeurIPS that year. Combined, they have 529 citations as of today.

74 902 4K 1.7M 2K

View Details

Pierre Dulac @_dulacp

3 years ago

@blair_fix @gilitrading What @gilitrading is trying to tell you is that: given your sample size, if your samples were from a random distribution you would have a decent chance to get R^2 > 20%. Check out this simulation with 100 data points: stats.stackexchange.com/a/39131

1 0 5 345 1

View Details

Lior Alexander @LiorOnAI

4 years ago

OpenAI just announced a new experimental model that knows when and how to browse the internet within ChatGPT!

33 77 628 156K 114

View Details

Pierre Dulac @_dulacp

4 years ago

@winsontang Thanks for the support 🙌

0 0 0 6 0

View Details

Pierre Dulac @_dulacp

4 years ago

@karpathy "Unsupervised Sentiment Neuron" 2015 paper by OpenAI, might also deserve a highlight. This paper hinted that the simple task of "predicting the next character" on a massive dataset could be enough to learn disentangled concepts like sentiment analysis. Mind-blowing findings.

0 0 0 0 0

View Details

Nassim Nicholas Taleb @nntaleb

4 years ago

Idiots are arguing ag the $8 using "free speech" arguments. English uses the same word for free as *gratis* (short for free of charge) & free as *liber*; other languages don't. Go tell the Le Monde newspaper they must be free (gratuit) because of free speech (liberté d'exp).

133 342 3K 0 76

View Details

Chris Albon @chrisalbon

4 years ago

2022: “WOW you can write a prompt and an AI will draw it!” 2028: “You want to write a prompt? First you need to hire 10-15 promptOps Engineers to build out your PromptFlow pipelines which sends promptjobs to your PromptLake from the PromptQueue using the EventPrompt stream”

168 1K 14K 0 541

View Details

Paul Graham @paulg

4 years ago

When you have an idea for a new kind of startup, ask yourself how likely it is that no one has thought of this idea before. If you conclude that it's unlikely, start looking for the reasons this idea is harder than it looks.

84 188 2K 0 262

View Details

Peyman Milanfar @docmilanfar

5 years ago

The Al logarithms are broken

42 47 483 0 32

View Details

Hjörtur Ingvi Jóhannsson @HjorturIngvi

5 years ago

@natesolon I think Karpov is a great example of a player who doesn’t fall apart in a bad position. Of course in a much different class than an ordinary gm.

0 1 2 0 0

View Details

Pierre Dulac @_dulacp

5 years ago

@nntaleb IMHO, the hash function is mostly used as a simple random game that people need to play by picking a random nonce, computing r_{t+1} = hash(r_t, transactions, nonce), they need r{t+1} the be lower than a given threshold, otherwise they just try another random nonce until it works

0 0 0 0 0

View Details

Félix Cantournet @La_Rainette

6 years ago

You: I code in C because I want to control exactly what the CPU does The CPU when you write if x { foo() } :

56 649 3K 0 315

View Details

Emanuel Derman @emanuelderman.bsky @EmanuelDerman

6 years ago

Human Affairs

32 231 911 0 207

View Details

François Chollet @fchollet

6 years ago

I'd guess I'd summarize it as, being a "deep learning expert" in 2021 is like being a "medicine expert" in 1800. You know a lot less than you think, and most of what you think you know is wrong. Just keep learning and experimenting, and don't play stupid status games.