Profile picture
Janelle Shane
@JanelleCShane
, 5 tweets, 3 min read Read on Twitter
new blog post: @runwayml's SPADE demo!

So much artistic potential, and no coding required.
aiweirdness.com/post/185617397… terrible block drawing of a car transformed into a highly detailed drawing of a sporty car
i am pleased to report that the @runwayml SPADE demo has a giraffe paintbrush mountain scene with boat and giraffes
as long as you're content to be Bob Ross and don't try to add floating islands or sheep-orbs, nvidia's gauGAN does a superior job being painterly.
plus it's an online demo!
nvidia.com/en-us/research… grey flat rocks, a grey flat seainteresting golden hour lighting. more clear to see these are sea cliffs
I should point out that the @runwayml and GauGAN demos both use @nvidia's SPADE algorithm - just different versions.
One difference: GauGAN has opinions about where mountains can be, and will put them below or above the horizon very grudgingly. floating islands and mountainous islands. also a floating boat.floating islands are very hazy. mountainous islands are highly pixelated
possible use case for neural net generated landscapes: sarcasm artsy ocean landscape. clouds spell out the word NO
Missing some Tweet in this thread?
You can try to force a refresh.

Like this thread? Get email updates or save it to PDF!

Subscribe to Janelle Shane
Profile picture

Get real-time email alerts when new unrolls are available from this author!

This content may be removed anytime!

Twitter may remove this content at anytime, convert it as a PDF, save and print for later use!

Try unrolling a thread yourself!

how to unroll video

1) Follow Thread Reader App on Twitter so you can easily mention us!

2) Go to a Twitter thread (series of Tweets by the same owner) and mention us with a keyword "unroll" @threadreaderapp unroll

You can practice here first or read more on our help page!

More from @JanelleCShane see all

Profile picture
Janelle Shane
@JanelleCShane
Decided to see what would happen if I trained a neural net to write names for self-aware AIs.
May I present: GPT-2 trained on AI names from Iain M. Banks's Culture novels
aiweirdness.com/post/185883998… Not Disquieting At All<br /> Surprise Surprise<br /> And That's That!<br /> Absently Tilting to One Side<br /> ASS FEDERATION<br /> A Small Note of Disrespect<br /> 8 Angry Doughnut Feelings<br /> Mini Cactus Cake Fight<br /> Happy to Groom Any Animals You Want
Impressed at how the neural net managed the Culture ship balance of enthusiasm and snark. New Arrangement<br /> I Told You So<br /> Spoiler Alert<br /> Bonus Points!<br /> Collateral Damage<br /> Friendly Head Crusher<br /> Scruffy And Determined<br /> Race To The Bottom
The neural net was particularly on point when generating the sassier Culture ship names Third Letter of The Week<br /> Well Done and Thank You<br /> Just As Bad As Your Florist<br /> What Exactly Is It With You?<br /> Let Me Just Post This<br /> Protip: Don’t Ask<br /> Beyond Despair<br /> Way Too Personal<br /> Sobering Reality Check<br /> Charming (Except For The Dogs)
Read 4 tweets
Profile picture
Janelle Shane
@JanelleCShane
oops that’s not what i meant #ProgrammingIn6Words
think ai is infallible? ask autocorrect
#ProgrammingIn6Words
copying human behavior means copying bias #ProgrammingIn6Words
Read 3 tweets
Profile picture
Janelle Shane
@JanelleCShane
There's an online game where you can take turns drawing a picture with an AI. The AI is brutal.
Game: bit.ly/2KwjbkM
cmon ai FOCUS
Read 4 tweets

Related threads

Profile picture
David Page
@dcpage3
New blog post: How does batch norm _really_ help optimisation?

We go on a tour of bad inits, degenerate networks and spiky Hessians - all in a Colab notebook:
colab.research.google.com/github/davidcp…

Summary 👇
1/ Early signs of trouble.

We learn that deep ReLU nets with He-init, but no batch norm, basically ignore their inputs! (Check out arxiv.org/abs/1902.04942 by Luther, @SebastianSeung for background.)

Easy to miss if you pool across channels:
2/ How to break your ResNet.

Next we go looking for trouble: we find degenerate networks near the training trajectory by (mis)using backprop. Batch norm comes to the rescue!
Read 4 tweets
Profile picture
Meenakshi Prabhune
@minu_pr
It's been 3 years since my first blog. I remember how terrified I was to hit "publish" on my first #blog post. Unleashing my words into the worldwide web was a scary thought. What if people hated my writing? Or worse, what if the article had mistakes? (1/5) #writingtips #SciComm
Importantly, this was supposed to be my first step towards transitioning into a career in #sciencewriting- a field that was a complete black box to my family and friends. Needless to say, everyone, including myself, was skeptical. (2/5)
Anyways, I tried, and learned 2 things over the years. 1) My writing didn't (positively or negatively) break the internet. At it's best and worst, readers moved on with their lives. (3/5)
Read 5 tweets
Profile picture
François Chollet
@fchollet
New blog post, where I describe a potential risk for the Bitcoin network medium.com/@francois.chol…
In short, a large fraction of the mining hardware produced in 2018 is going to be turned off. The hashrate, difficulty, & mining profitability will find a new equilibrium, but...
...there is going to be a lot of now-worthless hashpower on the market at extremely low prices, far more in fact than the hashpower that's actually online. Someone with a bit of capital available could buy up a fraction of that hardware and implement a 51% attack on the network.
Read 4 tweets
Profile picture
Sylvain Chabé-Ferret
@SylvainCF
[New Blog Post] Why We Cannot Trust the Published Empirical Record in Economics and How to Make Things Better

economistjourney.blogspot.com/2018/09/we-can…

That’s a big important one in which I discuss the evidence that there is a replication crisis in economics and propose solutions

\begin{thread}
I explain what the replication crisis is.
And the four contributors to it, the four horsemen of the apocalypse
Read 13 tweets
Profile picture
Arvind Narayanan
@random_walker
New blog post: How to constructively review a research paper
freedom-to-tinker.com/2018/05/15/how…
Let's start by acknowledging that three key review criteria — correctness/validity, impact/significance, and novelty/originality — differ greatly in terms of reviewers’ ability to judge them.
Unfortunately, "novelty" in peer review is often used as a euphemism for cleverness. If this tendency is left unchecked, publishing is at risk of devolving into a contest where academics impress each other.
Some suggestions for reviewers on how to constructively talk about correctness, impact, and novelty in a review. freedom-to-tinker.com/2018/05/15/how…
Read 4 tweets
Profile picture
G. Elliott Morris📈🤷‍♂️
@gelliottmorris
#New PA cong. map increases D chance of winning the House majority by 4%. They only need a 6.8% generic ballot margin now (down from 7.7) to be favored.

Diff. will be closer to 10% in Nov when uncertainty from polls is ⬇️. Redraw is a very big deal.

One major reason the new PA map is such a big gain for Democrats is that it solidifies, not just tips, PA districts in Philly burbs. Also creates one additional vulnerable R district. (Easily seen in this map made by @JMilesColeman)
My forecasting model will have new PA map in the near future. This is not just a simple switch; it’s a coding redesign, restructuring of input, & drawing new geographic files. Takes a while (for me, a student).

In the meantime, just add 3-4% to the model: bit.ly/2018_house
Read 4 tweets

Trending hashtags

#USS #truthwins #New #ByePhoenicia #ItEndsNow #Trump #CONSPIRACY #SayBraveThings #PatriotsFight #Barr #PAM #FISA #NutShellReport #brexit #NRA #JUSTICE #ClimateBreakdown #TeamPatriot #NoRedactions #USVIA #IDF #MORE #NEW #WeThePeople #Correction
Follow Us on Twitter!

Did Thread Reader help you today?

Support us! We are indie developers!

This site is made by just three indie developers on a laptop doing marketing, support and development! Read more about the story.

Become a Premium Member ($3.00/month or $30.00/year) and get exclusive features!

Become Premium

Too expensive? Make a small donation by buying us coffee ($5) or help with server cost ($10)

Donate via Paypal Become our Patreon

Thank you for your support!