AI Weirdness: the strange side of machine learning

Tag: OpenAI

Total 18 Posts
The drawings of DALL-E

The drawings of DALL-E

What type of giraffe would you like today? Last week OpenAI published a blog post previewing DALL-E, a new neural network that they trained to generate pictures from text descriptions. I’ve written about past algorithms that have tried to make drawings to order, but struggled to produce recognizable humans
Post-human flirting

Post-human flirting

In Henry J. Wehman’s The Mystery of Love, Courtship and Marriage Explained (1890) there are charts of secret messages Victorian flirters could send with fans, parasols, gloves, and handkerchiefs. Whether the handkerchief codes were ever in use (or mainly used to sell books), twitter user @_irishjars noticed that they
What to do with your time

What to do with your time

In Leonardo DaVinci’s diary you can find his to-do lists, the kinds of broad-ranging investigations the famous polymath would use to occupy his time: Draw Milan. Get the master of arithmetic to show you how to square a triangle Ask Giannino the Bombardier about how the tower of Ferrara
Splorts Teams

Splorts Teams

There is a game called Blaseball that started online in summer 2020. Gameplay is both entirely straightforward and profoundly confusing - a rulebook is now available, but is almost completely redacted, and what is left is ominous. Players have names like “Jessica Telephone” and “Jaylen Hotdogfingers” and keep getting incinerated,
High-visibility costumes for distanced viewing

High-visibility costumes for distanced viewing

I’ve used neural networks to generate Halloween costumes in the past, but the early ones struggled with their complete lack of context for the training data. Why is Statue of Liberty a plausible costume yet Statue of Pizza isn’t? Those neural networks, which had to learn to spell
What color are you?

What color are you?

I’ve been having fun using GPT-3 to generate weird versions of internet memes. Even if it’s a meme that didn’t exist when GPT-3’s training data was collected from the internet (which was October 2019 for the version I’m using), most memes will fit within its
This blog is not a blog of honor

This blog is not a blog of honor

Last time I used GPT-3 to suggest forbidding landscapes that would repel future civilizations from our nuclear waste disposal sites. There’s something about taking a subject that’s very very permanent - some kinds of nuclear waste remain deadly for over 250,000 years - and then trying to
A 10,000 year warning

A 10,000 year warning

It isn’t every project that requires people to make plans for thousands or millions of years in the future. But some kinds of nuclear waste remain deadly for over 250,000 years, probably outliving all existing civilizations, languages, and possibly even humanity itself. So when people design nuclear waste
AI supervillains

AI supervillains

Earlier, I did the rather unsettlingly meta exercise of using a new neural net, GPT-3, to generate hypothetical AI Weirdness blog posts. One of the tricks I can do with this: GPT-3 can play along with an opening in which I declare that I’ve performed a feat of dataset
The story behind AI drawing prompts

The story behind AI drawing prompts

Introducing #Botober, a set of AI-generated drawing prompts for each day in October! Last year I generated prompts by finetuning GPT-2 on 124 examples from previous years. The human-written training examples included items like Thunder, Fierce, Tired, or Friend. The neural net-generated examples included Complete Whoop, Take Control of Ostrich,
You've successfully subscribed to AI Weirdness
Great! Next, complete checkout for full access to AI Weirdness
Welcome back! You've successfully signed in.
Unable to sign you in. Please try again.
Success! Your account is fully activated, you now have access to all content.
Error! Stripe checkout failed.
Success! Your billing info is updated.
Error! Billing info update failed.