This model makes predictions in continuous representation space.
An LLM makes predictions in discrete input space.
Prediction in continuous representation space is done with a (non-generative) Joint Embedding Predictive Architecture, i.e what I've been advocating for about 5
MULTIMODAL AI
-
Continuous vs Discrete Prediction Space in Modern AI Models
By
–
-
Learning Abstract Representation Spaces for Improved Model Prediction
By
–
I said it was impossible to predict all the details at the pixel level.
I said the solution was to learn an abstract representation space within which prediction could more easily be learned.
Which is *precisely* what this model does. -
Character.ai Launches AI-Native Social Feed Platform
By
–
Consumption Creation Hot takes, trending stories, and epic Characters—now all in one feed. Read more about the Feed here: https://
blog.character.ai/character-ai-l
aunches-worlds-first-ai-native-social-feed/
… -
World’s First Native AI Social Feed Launches Community Features
By
–
Your imagination has a timeline now Introducing… Community Feed The WORLD’S FIRST native AI social feed will deliver a personalized stream of Characters, Scenes, and creator posts designed to inspire, entertain, and connect. Jump in and explore!
-
Google Tests AI Bug Hunter Agent Powered by Gemini
By
–
Google is testing a bug hunter agent, which reportedly found 20 vulnerabilities during the recent run. "AI-based 'Big Sleep' system powered by Gemini" One more job profile will be gone
-
Qwen-Image: Advanced 20B Parameter Model Excels in Benchmarks
By
–
Insane: Qwen-Image is a 20 billion-parameter MMDiT text-to-image model that stands out in official evaluations! In benchmark tests for text rendering and image processing, especially for complex Chinese prompt rendering, it achieves significantly better results than conventional
-
Qwen’s First Image Model Available on ModelScope
By
–
You can try out Qwen's first image model on ModelScope here: https://
modelscope.cn/aigc/imageGene
ration?tab=advanced
… Here's what I got for "A raccoon holding a sign that says "I love trash" that was written by that raccoon" -
Qwen-Image Now Available on Replicate at $0.025 Per Image
By
–
Qwen-Image is on Replicate.
With thanks to Pruna it's just $0.025 per image. https://
replicate.com/qwen/qwen-image -
Replicate introduces Veo 3 upscaling and new output formats
By
–
We've created some tools to make Veo 3 more useful for you.
— Replicate (@replicate) 4 août 2025
– Veo 3 upscaled
– 9:16 output for TikTok/Reels
– JSON prompting
Add these to your apps today: https://t.co/W5BTXJUlk7
(includes Veo 3 Fast) pic.twitter.com/iZLrvRXeobWe've created some tools to make Veo 3 more useful for you. – Veo 3 upscaled
– 9:16 output for TikTok/Reels
– JSON prompting Add these to your apps today: http://
replicate.com/veo-3-apps
(includes Veo 3 Fast) -
MIT’s Novel Image Generation Method Without Traditional Generators
By
–
Image generation systems normally use a tokenizer to compress & encode visual data, along w/a generator that combines & arranges these representations to create images.
— MIT CSAIL (@MIT_CSAIL) 4 août 2025
MIT's new method creates, converts, & "inpaints" images w/o using a generator at all: https://t.co/5PjOA4s6Xa pic.twitter.com/WGwOpZtZyWImage generation systems normally use a tokenizer to compress & encode visual data, along w/a generator that combines & arranges these representations to create images. MIT's new method creates, converts, & "inpaints" images w/o using a generator at all: https://
bit.ly/40ZlKkU