Gabriel Goh @gabeeegoooh Twitter profile

Last Seen Profiles

@wildflower_hour

@thisisformative

@azeemghumman

@SpinnaSpence

@zowipromotes

@Sinvicta

@kissfoundation

@XBLOODYDEMON_

@letskyfalls

@GraysonJAllen

@HCRInstitute

@DionThaBeast

@JosephAHastings

@MissJCousins

@HaleyPaige7

@pixel_heroes

@CheyneyWBB

@globejimmcbride

@um_fess

@Praise1k

@_harmonyxo

@SeguretatSCC

@waspsstudios

@CottarelliCPI

@ClimatSuisse

@MuskelboyX

@BelmontUniv

@BrandonStoner25

@carmino25

@ChetMason32

@pigwichkc

@Kaminskientomo

@1DARKONTOP

@meliarobin

@EvoPalaeoLab

@bitcoinkatee

Gabriel Goh

@gabeeegoooh

4 years

Epidemic calculator: a visual calculator for modeling possible paths of COVID19 (powered by a SEIR infection model)

91

1K

2K

Gabriel Goh

@gabeeegoooh

6 months

OpenAI is nothing without its people

22

43

797

Gabriel Goh

@gabeeegoooh

6 months

❤️

Sam Altman

@sama

6 months

i love the openai team so much

5K

4K

73K

7

18

465

Gabriel Goh

@gabeeegoooh

6 months

if you get the chance, do try decoding your Stable Diffusion generations with this decoder. You should see some improvements in text, small faces and straight lines. Made with @_tim_brooks , @DrYangSong , @model_mechanic , @txhf , @neonbjb

5

33

194

Gabriel Goh

@gabeeegoooh

4 years

Source code for epidemic calculator please use anyway you wish.

GitHub - gabgoh/epcalc: Epidemic Calculator

Epidemic Calculator. Contribute to gabgoh/epcalc development by creating an account on GitHub.

github.com

29

40

141

Gabriel Goh

@gabeeegoooh

3 years

@ericjang11

1

0

102

Gabriel Goh

@gabeeegoooh

3 years

A remarkable idea by @csvoss

Brendan Dolan-Gavitt

@moyix

3 years

The latest generation of adversarial image attacks is, uh, somewhat simpler to carry out

182

6K

18K

3

9

69

Gabriel Goh

@gabeeegoooh

7 years

attention is like a recommender system for neurons

1

5

57

Gabriel Goh

@gabeeegoooh

6 months

@sama ❤️

0

30

Gabriel Goh

@gabeeegoooh

4 years

@danrabinowitz this is a good idea!

3

1

29

Gabriel Goh

@gabeeegoooh

10 months

early version of scaling law, from Dunbar in 92

3

7

19

Gabriel Goh

@gabeeegoooh

7 years

@hardmaru My favorite quote from the slides

5

3

24

Gabriel Goh

@gabeeegoooh

4 years

A little late retweeting this, but it was super fun working with @nicklovescode optimizing splines to match curve detectors via gradient descent. here's one of the outtakes

Nick

@nickcammarata

4 years

Proud to share Curve Detectors Last year at OpenAI we decided to closely study 10 curve neurons, dedicating a week to the project. We found a lot more depth than we expected, and we’re writing three papers on them. Here’s the 40+ page part one!

1

34

154

1

5

23

Gabriel Goh

@gabeeegoooh

7 years

fascinating paper on running PCA on monkey neuron brain activations

1

3

21

Gabriel Goh

@gabeeegoooh

7 years

This neural singing synthesizer gave me goosebumps. (demo) (paper)

A Neural Parametric Singing Synthesizer

We present a new model for singing synthesis based on a modified version of the WaveNet architecture. Instead of modeling raw waveform, we model features produced by a parametric vocoder that...

arxiv.org

2

5

20

Gabriel Goh

@gabeeegoooh

4 years

neural nets discover the most fascinating algorithms

Chris Olah

@ch402

4 years

In InceptionV1, Black & White detectors detect the absence of color. They're implemented by having many color features inhibit them.

3

26

160

1

3

18

Gabriel Goh

@gabeeegoooh

3 years

whoa

Aran Komatsuzaki

@arankomatsuzaki

3 years

When you generate images with VQGAN + CLIP, the image quality dramatically improves if you add "unreal engine" to your prompt. People are now calling this "unreal engine trick" lol e.g. "the angel of air. unreal engine"

49

405

3K

2

1

13

Gabriel Goh

@gabeeegoooh

5 months

i'm tired of people calling LLMs "glorified autocompletes". They work just fine when you train them with the tokens reversed too.

1

0

14

Gabriel Goh

@gabeeegoooh

3 years

The Kronecker product: (i,j),(k,l) = x.shape, y.shape a = np.kron(x,y).reshape(i,k,j,l).transpose(1,3,0,2) b = x[None,None,:,:]*y[:,:,None,None] a == b

3

2

13

Gabriel Goh

@gabeeegoooh

7 years

Thanks @hardmaru and team!! This dataset looks awesome

GitHub - googlecreativelab/quickdraw-dataset: Documentation on how to access and use the Quick,...

Documentation on how to access and use the Quick, Draw! Dataset. - googlecreativelab/quickdraw-dataset

github.com

0

2

12

Gabriel Goh

@gabeeegoooh

7 years

a bad day is kinda like an adversarial example - it's the accumulation of lots of insignificant annoyances

0

1

12

Gabriel Goh

@gabeeegoooh

7 years

Two thoughtful takes on Evolutionary Strategies (1) as optimization on a smoothed function

1

5

10

Gabriel Goh

@gabeeegoooh

2 years

@ericjang11 no axis labels. no distracting grid lines. not even a title. Just a single plot, centered and impeccability antialiased, that is sloping gracefully downwards, always.

2

0

10

Gabriel Goh

@gabeeegoooh

7 years

the next step in vector embeddings - word2vec from wetware? 🤔

0

10

Gabriel Goh

@gabeeegoooh

6 months

@giffmana straight parallel lines, perfect checkerboards, large perfect circles. all surprisingly hard for diffusion models.

2

0

6

Gabriel Goh

@gabeeegoooh

8 years

Fresh new Hinton paper. Read it while it's still warm.

Using Fast Weights to Attend to the Recent Past

Until recently, research on artificial neural networks was largely restricted to systems with only two types of variable: Neural activities that represent the current or recent input and weights...

arxiv.org

0

6

Gabriel Goh

@gabeeegoooh

7 years

(2) And as Nesterov's gradient-free method

1

6

Gabriel Goh

@gabeeegoooh

4 years

@KyleCranmer @spott_andrew I haven't put up the svelte source code, but will be happy to if theres interest!

6

0

6

Gabriel Goh

@gabeeegoooh

7 years

RL for image captioning - an alternative to beam search?

0

6

Gabriel Goh

@gabeeegoooh

3 years

Thread on a delightful discovery by @ch402 - region neurons react to the maps of the region they represent!

Chris Olah

@ch402

3 years

When we make assumptions about what features exist in neural networks, they often prove us wrong. It turns out that 4% of CLIPs final neurons (8% on a liberal interpretation) are focused on geography. I certainly wouldn't have guessed that in advance!

7

86

433

1

0

5

Gabriel Goh

@gabeeegoooh

8 years

How did I miss this 13 year old ted talk? "Use your Electricity for more than Light"

0

5

Gabriel Goh

@gabeeegoooh

3 years

butterfly wings and peacock feathers are the closet thing we have to feature visualization in nature

0

5

Gabriel Goh

@gabeeegoooh

7 years

oops - that link was meant to go here

0

5

Gabriel Goh

@gabeeegoooh

4 years

@orderwithchaos R_t is the reproduction number after the intervention. sorry it was confusing!

1

0

5

Gabriel Goh

@gabeeegoooh

7 years

@Mybridge @dehhmesquita @anorangeduck @theflofly @pmigdal @jesseengel @distillpub @OpenAI @MSFTResearch @fb_engineering An outstanding list! (i'm biased however, my article's number 2)

0

4

Gabriel Goh

@gabeeegoooh

4 years

@sgouws @sindero i believe this is corrrect. since the reproduction number is fixed, increasing the duration of infection just spreads out the infections over a longer period. However there was a typo T^{-1}_{inf} should be T_{inf}

1

0

4

Gabriel Goh

@gabeeegoooh

7 years

@debasishg @ch402 the name eigenfeatures was inspired by this famous paper on eigenfaces - that reference may help too.

1

3

Gabriel Goh

@gabeeegoooh

3 years

something i've not fully understood: what is the role of time in Pearl's formalism of causality?

0

3

Gabriel Goh

@gabeeegoooh

3 years

@dribnet @golan @kyttenjanae @cyrildiagne @NeuralBricolage @videodrome @soficrespo91 @PINguAR @JanelleCShane @quasimondo @mschf @zzznah @nettieb @aman_gif @graycrawford @annaridler @cogconfluence @Artbreeder @junyanz89 👋

0

3

Gabriel Goh

@gabeeegoooh

4 months

was this video a part of your AI origin story

yes

4

no

16

cool video

9

2

0

4

Gabriel Goh

@gabeeegoooh

8 years

@doctorow :O Glad you found it interesting

0

2

3

Gabriel Goh

@gabeeegoooh

7 years

@dribnet yes - matlab is the OG deep learning language, it is my #1 too. nothing better than a folder with undocumented 200 .m files

0

3

Gabriel Goh

@gabeeegoooh

8 years

a fun new (?) convex set - stable matrices satisfying the pac man condition.

Smerity

@Smerity

8 years

I'm in love with this clear + intuitive discussion of RNNs as dynamical systems w @mrtz + Ma + Recht. Also, Pacman.

2

21

76

0

3

Gabriel Goh

@gabeeegoooh

3 years

.

0

2

Gabriel Goh

@gabeeegoooh

7 years

@pchampin @danbri @chaals @evamen killing us softly with it's song ...

0

2

Gabriel Goh

@gabeeegoooh

7 years

@debasishg @ch402 [1/2] Thanks! See 9.3.2 of for an example on closed form gd

2

1

2

Gabriel Goh

@gabeeegoooh

7 years

@dribnet gan man style?

0

2

Gabriel Goh

@gabeeegoooh

4 years

@gtjarruda @sgouws @sindero This is correct!

0

2

Gabriel Goh

@gabeeegoooh

7 years

@hardmaru @dennybritz but "Boltzmann Machine" was his magnum opus imo. That name is sexy.

0

1

Gabriel Goh

@gabeeegoooh

8 years

sorry :( @jjeffries1

0

1

Gabriel Goh

@gabeeegoooh

8 years

@dribnet much appreciated :) I was amazed it worked

0

1

Gabriel Goh

@gabeeegoooh

7 years

@yoavgo @dustinvtran dropout?

0

1

Gabriel Goh

@gabeeegoooh

8 years

@mfortki if you look carefully, you may notice other genitalia being represented. you got the math down though.

0

1

Gabriel Goh

@gabeeegoooh

2 years

@pmddomingos But if you could find a vector pointing to the minimum, the problem becomes trivial. You just needs to follow it in a straight line till you're there

1

0

1

Gabriel Goh

@gabeeegoooh

7 years

@fbz It appears we've come a full circle from the Jacquard Loom.