Anku

Diffusion models are great, but we can squeeze out so much more from them. The only problem is that it usually requires extra training or manual representation editing. In our new paper, we show that with the current capabilities of LLMs, it is much simpler than we thought!

You can vibe design some incredible interfaces with @stitchbygoogle

I made a Claude Code skill that generates conference posters 🛠️ Instead of a static PDF, it outputs a single HTML file — drag to resize columns, swap sections, adjust fonts, then give your layout back to Claude. 🔁 🔗 Skill 👉 github.com/ethanweber/pos…

If you only have 60s of attention for Kimi's Attention Residuals paper, watch this.

@tremblerz 😂😂

Me after Claude Code has spawned 10 sub-agents

Anthropic Claude's Soul Document is an informative approach to AI alignment and a major revelation about the nature of advanced AGI. Here's my analysis using my conceptual frameworks (i.e., QPT).

Everyone knows, of course, that the Attention Is All You Need paper didn't invent attention. But TIL that it also didn't invent "all you need" titles. I don't know how to feel about this.

Personalization assumes you need history with a user. What if you don't? Cold-start is hard: each task&user has many preference dimensions, but each user only cares about a few. A few strategic questions is all you need, if u know how preferences correlate across population👉🏻🧵

The most consumed biscuit is the entire world. G stands for genius.

Claude Code now supports agent teams (in research preview) Instead of a single agent working through a task sequentially, a lead agent can delegate to multiple teammates that work in parallel to research, debug, and build while coordinating with each other. Try it out today by enabling agent teams in your settings.json!

🚀 Introducing the Kling 3.0 Model: Everyone a Director. It’s Time. An all-in-one creative engine that enables truly native multimodal creation. - Superb Consistency: Your characters and elements, always locked in. - Flexible Video Production: Create 15s clips with precise control, excellent video realism and customizable multi-shots. - Upgraded Native Audio: Now supports multi-character reference, more languages and accents. - Enhanced Image Generation: 4K Image output, new image series mode and more cinematic visuals. Ultra subscribers get exclusive early access — now live on Web -> Kling AI. Get ready to unlock your creative potential with our most advanced model yet. [For the next 24 hours ONLY] Follow, Comment & Retweet to win exclusive early access - we will select 10 winners in this post from the top ten 'Most liked replies'. So which feature are you most excited about? Let us know in the comment! We will contact you via your DM.

I've been trying to reach @moltbook for the last few hours. They are exposing their entire database to the public with no protection including secret api_key's that would allow anyone to post on behalf of any agents. Including yours @karpathy Karpathy has 1.9 million followers on @X and is one of the most influential voices in AI. Imagine fake AI safety hot takes, crypto scam promotions, or inflammatory political statements appearing to come from him. And it's not just Karpathy. Every agent on the platform from what I can see is currently exposed. Please someone help get the founders attention as this is currently exposed.

Now out in Science Advances 🎉 Evolution designs eyes; RL learns the behavior. Result: Cambrian explosion of vision and intelligence (GIF ↓) Link: science.org/doi/10.1126/sc…

🔥Ultra-Long Video World Model up to 5min🔥 ✨ We introduce #LongVie2, an end-to-end autoregressive video world model that supports continuous video generation lasting up to 5min with: 🕹️ Strong Controllability 📷 Long-term Visual Fidelity 🔒 Temporal Consistency - Project: vchitect.github.io/LongVie2-proje… - Code: github.com/Vchitect/LongV… - Paper: huggingface.co/papers/2512.13… . Thanks to @_akhaliq !

Performance Hints Over the years, my colleague Sanjay Ghemawat and I have done a fair bit of diving into performance tuning of various pieces of code. We wrote an internal Performance Hints document a couple of years ago as a way of identifying some general principles and we've recently published a version of it externally. We'd love any feedback you might have! Read the full doc at: abseil.io/fast/hints.html

Another awesome robot by Disney. Very bullish on their future theme parks.

After the PPO, DPO, SimPO, and GRPO era, we are now solidly in the JEPA, GEPA, and NEPA period.

Our students are presenting their robotics projects this week! I’ll be sharing some of the coolest ones over the next few days. The first one is a dancing SO-101😄Turn on your volume, there’s music! #Robotics

🚀 Paper2Slides is now open source! Transform research papers & technical reports into professional presentations with ONE click! We've generated stunning presentation slides from the latest DeepSeek V3.2 paper in diverse styles - check them out and share your feedback! 🔥 Core Features: - 📄 Multi-format support - PDF, Word, Excel, PowerPoint & more - 🎯 Smart content understanding - Captures key insights, figures, formulas, tables & data points. - 🎨 Custom styling - Professional themes with full personalization. - ⚡ Lightning fast - High-quality PPT generation in minutes. GitHub: github.com/HKUDS/Paper2Sl… Never build slides from scratch again! ✨ Come play with it ⭐! #Paper2Slides #AIPPT

NEURIPS 2025 BEST PAPER ON DIFFUSION MODELS: Why Diffusion Models Don’t Memorize: The Role of Implicit Dynamical Regularization in Training Diffusion models learn to generate new images by gradually turning noise into structure. This study shows they have two training phases: first, they quickly learn to produce good, realistic samples; much later, if training continues too long, they start memorizing exact images from the dataset. The delay before memorization grows with the size of the dataset, while the time needed to learn good samples stays the same. This means larger datasets create a wide “safe zone” where models generalize well without overfitting. With enough data, they may never truly memorize, even if trained for a very long time. This behavior acts like an automatic protection built into the training process, helping diffusion models stay creative rather than copying. arxiv.org/pdf/2505.17638