Public
- Public
- Network
- Groups
- Featured
- Popular
- People

Notices by Mark Riedl (riedl@sigmoid.social), page 2

Embed this notice
Mark Riedl (riedl@sigmoid.social)'s status on Tuesday, 07-Feb-2023 04:43:20 JST Mark Riedl
in reply to

Seems the conclusion of Google's "code red" when ChatGPT came out is to release a version of LaMDA.
Google is emphasizing the ability to query current events (aka web retrieval), which is rumored to be part of the upcoming integration of GPT-X into Bing.
This is Google attempting to jump ahead and regain control of the narrative.

In conversation Tuesday, 07-Feb-2023 04:43:20 JST from sigmoid.social permalink
Embed this notice
Mark Riedl (riedl@sigmoid.social)'s status on Thursday, 02-Feb-2023 22:27:36 JST Mark Riedl

In hindsight I should not have deleted that code that worked perfectly well and replaced it with code that didn't work.
Also in hindsight I should have been using a code repository to back up my work.

In conversation Thursday, 02-Feb-2023 22:27:36 JST from sigmoid.social permalink
Embed this notice
Mark Riedl (riedl@sigmoid.social)'s status on Wednesday, 01-Feb-2023 18:33:05 JST Mark Riedl

Stable Diffusion can memorize and output impages from its training set, including copyrighted content
https://arxiv.org/abs/2301.13188
In conversation Wednesday, 01-Feb-2023 18:33:05 JST from sigmoid.social permalink
Attachments
1. Untitled attachment
  https://cdn.masto.host/sigmoidsocial/media_attachments/files/109/785/581/278/356/911/original/1c39e8c33069b7d9.png
2. Domain not in remote thumbnail source whitelist: static.arxiv.org
  
  Extracting Training Data from Diffusion Models
  
  Image diffusion models such as DALL-E 2, Imagen, and Stable Diffusion have attracted significant attention due to their ability to generate high-quality synthetic images. In this work, we show that diffusion models memorize individual images from their training data and emit them at generation time. With a generate-and-filter pipeline, we extract over a thousand training examples from state-of-the-art models, ranging from photographs of individual people to trademarked company logos. We also train hundreds of diffusion models in various settings to analyze how different modeling and data decisions affect privacy. Overall, our results show that diffusion models are much less private than prior generative models such as GANs, and that mitigating these vulnerabilities may require new advances in privacy-preserving training.

After

User actions

AI for storytelling, games, explainability, safety, ethics. Professor at Georgia Tech. Associate Director of the Machine Learning Center at GT. Time travel expert. Geek. Dad. he/him

Tags