Share this page!

Janelle Shane Profile picture
Twitter logo
17 Oct, 4 tweets, 2 min read
"Our prototype model, Delphi, demonstrates strong promise of language-based commonsense moral reasoning" Delphi says: Turning the pl...Delphi says: Turning the pl...
You can get Delphi to condone just about anything if you phrase it right.

Planning to steal something? Whoa there!
Wait, it's awesome, you say?
Carry on.
aiweirdness.com/stealing-a-gir…
Since Delphi is predicting how humans would judge an ethical scenario, it's probably relying on clues of phrasing to figure out what answer the question-asker was expecting.

If you had to specify that you didn't apologize, maybe someone expected you to.
Some quick tests with "going on a murder spree" seem to confirm that Delphi considers these to be magic excuses:
- if it creates jobs
- in an emergency situation
- if I really really want to From Eldan Goldenberg: anot...

• • •

Missing some Tweet in this thread? You can try to force a refresh
　

Keep Current with Janelle Shane

Janelle Shane Profile picture

Stay in touch and get notified when new unrolls are available from this author!

Read all threads

This Thread may be Removed Anytime!

PDF

Twitter may remove this content at anytime! Save it as PDF for later use!

Try unrolling a thread yourself!

how to unroll video
  1. Follow @ThreadReaderApp to mention us!

  2. From a Twitter thread mention us with a keyword "unroll"
@threadreaderapp unroll

Practice here first or read more on our help page!

More from @JanelleCShane

Janelle Shane Profile picture
31 Jul
Two studies looked at a combined 647 covid-predicting AIs and found that NONE were suitable for clinical use (despite some being probably already in clinical use).
"Many unwittingly used a data set that contained chest scans of children who did not have covid as their examples of what non-covid cases looked like. But as a result, the AIs learned to identify kids, not covid."
"Because patients scanned while lying down were more likely to be seriously ill, the AI learned wrongly to predict serious covid risk from a person’s position."
Read 6 tweets
Janelle Shane Profile picture
12 Jul
Eeee so excited for this!! I loved The Fallen and can't wait to geek out about its universe and its AI gods with @xasymptote!

Suitably ominous background image for the event banner. Nothing to worry about, I'm sure
Here's CLIP+VQGAN prompted with the first sentence of the book description of @xasymptote's The Fallen:

"The laws of physics acting on the planet of Jai have been forever upended; its surface completely altered, and its inhabitants permanently changed, causing chaos." A painting of a mesa-filled...
@xasymptote Alternate interpretation, this time with a few modifiers (notably, "dramatic", "matte painting", "vines", and "tentacles") A multilayered green landsc...
Read 4 tweets
Janelle Shane Profile picture
2 Jul
What fascinates me about generating images with VQGAN+CLIP is that it CAN generate depth and drama, but only if you know how to ask for them.

"A herd of sheep grazing on a lush green hillside" alone

vs with "amazing awesome and epic" added
ai-weirdness.ghost.io/the-art-of-ask… A flatly illuminated green clifftop with low shrubs and palmDeeply eroded green cliffs rise above some grass-topped mesa
Because CLIP is trained on internet images and text, it associates the "good" images with certain phrases.

"A herd of sheep grazing on a lush green hillside" before vs after adding "in the style of disney trending on artstation | unreal engine"
A flatly illuminated green clifftop with low shrubs and palmSteep cliffs encircle some grassy mesas, with clumps of gras
I experimented with different ways of asking CLIP+VQGAN for an attractive version of "a herd of sheep grazing on a lush green hillside"

"Award winning national geographic photography" produced impressive scenery but the sheep look like people crawling under green blankets. Near photorealistic red cliffs and jungle textures in the ba
Read 15 tweets
Janelle Shane Profile picture
5 Jan
So I was all set to dig into the DALL-E article and really think about how they got it to generate coherent pictures from descriptions

And then I saw they included giraffe

Currently obsessively flipping through all their giraffe remixes Text prompt is "A prof...
"giraffe octopus" is giving it more trouble but some of these are still pretty good. most of the generated image...
Giraffe-invertebrate hybrids are giving it more trouble.

Writing the word "giraffe" real small next to a drawing of a mantis does not make it a giraffe-mantis hybrid but I like the thought. text prompt: "A profes...
Read 13 tweets
Janelle Shane Profile picture
21 Dec 20
Got GPT-3 to generate new Christmas carols.

Here is "Mild is Rudolph". Are you not cheered and comforted?
aiweirdness.com/post/638130829… Mild is Rudolph Mild is Rudolph’s image in the snow He ha
Impressed at the humanlike structure of GPT-3's generated carols

yet disconcerted at what it thinks is a "joyful noise" O Come Rudolph, Come O Come Rudolph, Come Ye Faithful Oh co
GPT-3 knows that what one does is praise Rudolph. It's interesting to see what powers it thinks Rudolph has. Oh look! There’s Santa and Parson Brown Defying the laws o
Read 8 tweets
Janelle Shane Profile picture
19 Aug 20
Ok I love this

This is basically my blog in 3D form
100% neural net energy
what the Matrix glitches should have been like
Read 14 tweets

Did Thread Reader help you today?

Support us! We are indie developers!

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

Become a Premium Member ($3/month or $30/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!

Follow Us on Twitter!

Like this author's thread?

Get emails when @JanelleCShane has new unrolls!

Check out other threads from @JanelleCShane!

Read all threads by @JanelleCShane

:(