Opens profile photo
Follow
Click to Follow JanelleCShane
Janelle Shane
@JanelleCShane
I write aiweirdness.com. My book "You Look Like a Thing and I Love You" is out now! Research Scientist in optics. she/her. wandering.shop/@janellecshane
Boulder, COjanelleshane.comJoined April 2014

Janelle Shane’s Tweets

Sometimes you get approximately what you asked for, sometimes you get mythical beings
Quote Tweet
Sitting at the airport waiting for Kevin’s plane and poking the latest version of the Midjourney AI. In some ways, this is a huge step forward. It got great at understanding “cartoon.” In other ways…well, I wanted a Labrador fisherman character holding a fish.
Show this thread
Image
2
35
I just finished a preprint and let me just say that a gaslamp murder mystery where the gas is the swirling atmosphere of Jupiter is GENIUS
Quote Tweet
promo time in case these are the Last Days of Twitter™️! The Mimicking of Known Successes: 👩🏾‍🤝‍👩🏽 f/f romance 🕵🏽‍♀️ murder mystery 🪐 on Jupiter 🎓 academia 🚅 long train journeys 🫖 tea March 7 from @TorDotComPub Pre-order links👇🏽 publishing.tor.com/themimickingof
Show this thread
Image
1
35
Last week, a Redditor fine-tuned an AI image model on the work of one illustrator, sparking a debate about the ethics of reproducing a living artist's style. I talked to that artist to see how she felt about it, and the person who made it.
28
1,128
Show this thread
There’s also this tool for finding people who’ve put their mastodon handles in their Twitter bios
Quote Tweet
This is cool! Note to anyone using this that some of the results here might be email addresses due to formatting issues but I've already found a few people I didn't know had Fediverse accounts I want to go follow twitter.com/luca/status/15…
15
Show this thread
I think I remember someone tagged me in a drawing youtube channel that did a couple of the #botober drawing prompts - but I was so busy at the time I forgot to watch it and now I can't find it in my mentions. Or did I imagine the whole thing?
3
10
Another failure: scorpion costume. #dalle2 seems to revert to generic "insect" features when trying to put together a scorpion costume. "yes i know you wanted to be a scorpion but this is what they had. you look lovely."
The cat is wearing a black and yellow striped t-shirt with a black and yellow hood with cow horns on it. The cat is also wearing striped double butterfly wings.
A cat in a hooded t-shirt. The hood has orange ears and black antennae with pom-poms on them. The hem of the tshirt is orange, and standing along the cat's spines are mesh plates like stegosaurus spines.
The cat is wearing a black hooded t-shirt with double pom-pom antennae on the hood. The cat has two wide yellow bee wings.
5
62
Show this thread
When I try to get DALL-E2 to dress my cat up as other animals, one common failure mode seems to be... this. "studio photo of a relaxed tortoiseshell cat wearing a giraffe costume"
Slumped in a giraffe-patterned hoodie is an extremely short-necked giraffe.
8
133
Show this thread
Here's an example of a site I'm pretty sure is at least partially AI-generated, which came up in search results for "do woodchucks like peanut butter" I used the GPTrue or False browser plugin to analyze it, and it's coming up with 0.1% chance that the article is human written.
Site reads in part "According to some researchers, it is found that groundhogs are like peanuts and also they can eat them. groundhogs are Mainly herbs, plant people eat a variety of plants, including from people’s gardens. But they can also eat things that we consider insects, such as worms, other insects, and snails."
GPTrue or False browser plugin reports that there is a 0.1% chance the text is human written
5
28
Show this thread
I'm also curious how many of these low-quality search results are AI generated with something like GPT-3, as opposed to being stolen and automatically reworded. i.e. the sites that end up referring to the New York Times as Novel York Instances
2
28
Show this thread
Informal poll: how often are you seeing AI-generated articles in your search results, on what types of searches (if any) do you tend to see them, and have you altered your search methods at all to avoid them?
Quote Tweet
I’ve been tinkering with AI writing tools and here is my biggest takeaways: 1. Amount of low quality content used to flood Google results and drive ad impressions will approach infinity. 2. Internet users will flock to where the humans are. Wikipedia, Reddit, Quora, etc.
Show this thread
Image
Image
Image
22
22
Show this thread
I have my PhD in "AI" & spent my career in tech as an engineer & then researcher. The skills required to understand how a system harms society aren't the same as those required to build it. Ppl don't need to know how face recognition works to know that it harms them.
Quote Tweet
I beg you, if you work on the Ethics of AI, please learn something about AI.
81
5,279
Show this thread
An AI like GPT-3’s goal is to sound like every other lifehack listsicle from the mid 2010s. In training it’s rewarded for blending in, not for new ideas.
Quote Tweet
24 hours ago, I tried an experiment - I tweeted a thread with 15 productivity hacks. It’s become one of my highest performing tweets of all time, with over 1 million impressions and 23,648 engagements. The truth is - the entire thread was written by an AI. Let me explain… 1/ twitter.com/AliAbdaal/stat…
Show this thread
Image
2
84
I'm giving a talk on Saturday at ! I got to preview the 100% online space during my tech check and somehow ended up in a room that was full of oatmeal? it was pretty cool
Quote Tweet
We're hard at work to give the space a fresh coat of paint before showtime! Even if you came last year, the space will be full of fun new easter eggs and tracery grammars to explore. ...also one of us may have taken "fresh coat of paint" too literally and added some color schemes
A partial screenshot of the social space with the settings modal open. The "Amber CRT" theme is selected, and the color scheme mimics an old amber monitor, using a black background and orange text.
24