Skip to main content
Back to blog
ā€¢2 min read

DALL-E 2 Is Wild

AItechthoughts

so openai dropped this thing called dall-e 2 a few months ago and i've been watching the demos and i genuinely don't know how to process it.

you type words. it makes pictures. like, good pictures.

what it does

input: "a corgi wearing a space suit, digital art"

output: an actual picture of a corgi in a space suit that looks like a real artist made it

what.

why this feels different

i've been paying attention to AI stuff casually for a while. cool projects here and there. but this feels... different.

it's not just that it works. it's that it works in a way that feels creative. which is supposedly the thing humans are good at. the thing that machines can't do.

except now maybe they can?

the discourse

everyone on twitter is either:

  1. "this is incredible! art is democratized!"
  2. "this is terrifying! artists will lose their jobs!"
  3. "this is overhyped, it can't do X specific thing"

i'm somewhere in the middle. impressed and slightly unsettled.

technical curiosity

as a CS student, i want to understand how it works. from what i can tell:

  • it's a diffusion model (learned that term today)
  • trained on millions of image-text pairs
  • learns to go from noise to image based on prompts
  • the details are proprietary but the papers are fascinating

i understood maybe 40% of the research paper. which is 40% more than i expected.

implications

if AI can generate images from text, what else can it do?

  • code? (github copilot exists already)
  • music?
  • writing?
  • everything?

there's a future where AI is a tool that amplifies human creativity. there's also a future where AI replaces a lot of what humans do.

probably it's both. probably it's complicated.

my hot take

the technology is cool. how we use it matters more than how it works. and we (humans, society, whatever) are historically bad at predicting how we'll use new technology.

so i'm going to stay curious. keep learning. try to understand what's coming.

whether i like it or not, this is going to be relevant to my career.

also i spent an hour generating pictures of cats in various historical outfits. for research.