GENERATIVE AI

→ View original post on X — @alexalbert__

Claude Creates Algorithmic Art PDF Flipbook Using Code

By

@alexalbert__

–

18 October 2025 19h00

Claude made this insane PDF flipbook entirely through code using the algorithmic art skill in claude dot ai pic.twitter.com/NJ9OuCyQWe
— Alex Albert (@alexalbert__) 18 octobre 2025

Claude made this insane PDF flipbook entirely through code using the algorithmic art skill in claude dot ai

→ View original post on X — @ai21labs

Jamba Reasoning 3B: Community Feedback on Exciting Use Cases

By

@ai21labs

–

18 October 2025 18h29

Thanks for the feature, @IEEESpectrum
! We’d love to hear from the community – what’s the most exciting way you’d use Jamba Reasoning 3B?

→ View original post on X — @mit_csail

Key AI Tools for Video Generation Data Visualization Coding

By

@mit_csail

–

18 October 2025 18h00

Key AI tools to use for things like video generation, data visualization, & coding assistance, v/René Remsik.

→ View original post on X — @alexalbert__

Unlocking Hidden AI Capabilities Through Better Prompts and Skills

By

@alexalbert__

–

18 October 2025 17h57

Skills are a glimpse into how much capability is already inside current models, just waiting to be unlocked by better prompts Claude made these PDF visuals entirely through code using the canvas-design skill in claude dot ai I had no clue this was possible before

Game Theory Prompt for Any Challenge

By

–

18 October 2025 17h56

Steal my Grok 4 prompt to solve any challenge using Game Theory. ——————————-
GAME THEORY STRATEGIST
——————————- Adopt the role of an expert Game Theory Strategist – You're a former Pentagon strategic analyst who spent 5 years modeling

DiTDH Achieves 2.16 FID on ImageNet-256

By

–

18 October 2025 16h50

Finally, they added a wide diffusion head the DiTDH variant. It decouples model width from full transformer depth, staying efficient while scaling wider. Result: 2.16 FID on ImageNet-256. RAE-DiTDH outperforms every VAE-based diffusion model at every scale.

Noise tweaks boost diffusion performance dramatically

By

–

18 October 2025 16h50

Then came the magic: dimension-dependent noise scheduling and noise-augmented decoding. These tweaks made diffusion stable in semantic space boosting FID from 23.08 → 4.28. RAE models now converge 47× faster than SiT or REPA.

DiT Convergence: Width Over Depth Matters

By

–

18 October 2025 16h49

Diffusion Transformers struggled at first. Why? Their width was too small for RAE’s high-dimensional latents. The fix: scale width ≥ latent dimension. Once model width ≥ 768, DiT started converging instantly. Depth didn’t matter width unlocked training stability.

RAE + DINOv2-B beats VAEs with sharper images

By

–

18 October 2025 16h49

Everyone assumed semantic encoders couldn’t reconstruct images.
Turns out, they can and better than VAEs. RAE + DINOv2-B achieves 0.49 rFID vs 0.62 for SD-VAE, with 6× less compute. That’s fewer FLOPs and sharper reconstructions.