OpenAI, Google will watermark AI-generated content to hinder deepfakes, misinfo

https://arstechnica.com/ai/2023/07/openai-google-will-watermark-ai-generated-content-to-hinder-deepfakes-misinfo/

23 points · 8 comments · view on lemmy.world

8 Comments

rikudou@lemmings.world · 19 pts · 3y

Sure, that'll help... While currently these companies are in the lead, even Google internally admits they've lost to open source.

Stable Diffusion is not as plug and play as Midjourney, DALL-E etc., but damn is it much more powerful. And no one is gonna watermark open source (because someone is gonna release a fork without the watermark like 3 and half seconds after the commit adding the watermark is pushed).

Sir_Kevin@discuss.online · 9 pts · 3y

It won't be long before AI content is indistinguishable from human made content. So now they intend to flag one from the other I do not know. What about human made content where AI is used as a tool?

pai_zosima@programming.dev · 6 pts · 3y (3 replies)

How?

flathead@quex.cc · 7 pts · 3y (1 reply)

Very simple. They will use water from their cloud vaporware.

ImpossibleRubiksCube@programming.dev · 4 pts · 3y
[ removed ]
Phoenix@programming.dev · 1 pts · 3y

Presumably you watermark all the training data.

At least, that's my first instinct.

swordsmanluke@programming.dev · 6 pts · 3y

Sure, that'll work. Not like a nation state with a large budget would be able to replicate these very well-understood technologies and release falsified NON-watermarked media that now has a ring of "truth" because it doesn't have Le Goog's magic sigil in it....

ImpossibleRubiksCube@programming.dev · 2 pts · 3y
[ removed ]
BestBunsInTown_@lemmy.world · 2 pts · 3y

I mean thats a great thing but open source research kinda defeats the purpose, people gonna use stuff for evil