By
–
Just now reading through the Gemma 4 blog Safe to say the @huggingface team is goated Lots of usage examples, guides on inference and fine-tuning, highly recommend! huggingface.co/blog/gemma4
→ View original post on X — @huggingface, 2026-04-06 11:35 UTC
By
–
Director-level camera control is officially here for everyone.
— AI Highlight (@AIHighlight) 6 avril 2026
This tool is turning prompt engineering into actual cinematography.
Who’s ready to make their first AI feature film this weekend. https://t.co/GFhTMfrLT1 pic.twitter.com/D6dE3kIc33
Director-level camera control is officially here for everyone. This tool is turning prompt engineering into actual cinematography. Who’s ready to make their first AI feature film this weekend. OpenArt (@openart_ai) Seedance 2.0 is live on OpenArt – for Teams & Enterprise users first. -Up to 9 images, 3 videos, and 3 audio files as references -Director-level camera control -90%+ usable output on the first attempt -Cinematic multi-shot storytelling. -Physics that actually holds. 60% off the annual Teams plan. US & individual users: join the waitlist for early access 🔗 lnkd.in/g3HYxWkd — https://nitter.net/openart_ai/status/2039553049560223799#m
→ View original post on X — @aihighlight, 2026-04-06 11:10 UTC
By
–
"Director-level camera control" is exactly what we’ve been waiting for. This feels like a huge win for serious visual storytellers.
By
–
Top stories in AI today: – Anthropic boots third-party agents from Claude plans
– The Rundown Roundtable: Our AI use cases
– How to take AI notes on phone calls
– Netflix opens physics-aware AI for video editing
– 4 new AI tools, community workflows, and more
By
–
A.B.C. (AI Book Cataloguer): point a camera at your bookshelf, detect spines with YOLOv8s on Metis M.2, and catalog your library automatically. Built by community member Denovo on @orangepixunlong. Open source. #EdgeAI #ComputerVision 📚 eu1.hubs.ly/H0t23Vb0 [Translated from EN to English]
→ View original post on X — @axeleraai, 2026-04-06 10:30 UTC
By
–
What if MLLMs could process visual data much faster without sacrificing performance? Eastern Institute of Technology, Ningbo, with USTC, SJTU, and LMU Munich presents HiDrop just for that! This new framework intelligently reduces visual tokens by processing them only when active fusion truly begins (Late Injection) and dynamically pruning them across deeper layers (Concave Pyramid Pruning with Early Exit). It focuses computation where it matters most. HiDrop compresses ~90% of visual tokens, matches original MLLM performance, and accelerates training by 1.72x. A new state-of-the-art for efficient MLLM training & inference! HiDrop: Hierarchical Vision Token Reduction in MLLMs via Late Injection, Concave Pyramid Pruning, and Early Exit Paper: arxiv.org/pdf/2602.23699 Code: github.com/EIT-NLP/HiDrop Our report: mp.weixin.qq.com/s/QKGZ7cFi0… 📬 #PapersAccepted by Jiqizhixin
→ View original post on X — @jiqizhixin, 2026-04-06 10:16 UTC
By
–
This is not normal. A developer just open-sourced a full AI agent orchestration dashboard that runs 15 providers, bridges 10 chat platforms, and manages a distributed swarm of autonomous agents all self-hosted. It's called SwarmClaw. You deploy it once. Then you connect Claude, GPT-4o, Gemini, Grok, DeepSeek, Mistral, Groq, Ollama all of them from the same dashboard. Your agents live inside Discord, Slack, Telegram, WhatsApp, Signal, iMessage, Teams, Google Chat, and Matrix at the same time. Each one has its own persistent memory backed by both FTS5 keyword search and vector embeddings. Each one runs on LangGraph with automatic sub-agent routing and checkpointed execution so complex tasks never fail silently mid-way. When an API key rate-limits, the model failover system rotates to the next credential automatically. When an agent hits a long-running task, the background daemon processes it on a 30-second heartbeat while you do something else. The OpenClaw Gateway feature lets you point different agents at different OpenClaw instances running on different machines one on your laptop, one on a VPS, one in another country. You manage the whole swarm from a single mobile-friendly UI. Install with one npm command. Deploy via Docker. Update with one button in the sidebar. 107 commits. 24 releases. MIT License. 100% Open Source.
→ View original post on X — @scobleizer, 2026-04-06 09:33 UTC
By
–
Building makes you learn what AI actually can do. A lot, it turns out. Micah Berkley – The 50 Cent of AI. (@MicahBerkley) 🚀Just discovered a dope curated AI news website, in a format that I absolutely love! Alignednews.com – Super clean interface. The format is not only dope for consumption for humans, but the robots love it too. And they have a RSS Feed. I think this is a @Scobleizer production. Great tool —————— With that my balanced news round up consist of 1. AlignedNews.com – General Purpose AI and Software News. Great UI and good intel. ~~ @Scobleizer & @LevangieLabs 2. hype.replicate.com – This has been a staple in my back pocket. It curates news from some of the AI giants by the replicate.com | @replicate team. 3. HackerNews – Of course you have to have @hackernews by @ycombinator — https://nitter.net/MicahBerkley/status/2041086087905349888#m
→ View original post on X — @scobleizer, 2026-04-06 09:32 UTC
By
–
This is the Holodeck and why it is different than the Metaverse.
— Robert Scoble (@Scobleizer) 6 avril 2026
Mark Zuckerberg’s metaverse can’t do this. https://t.co/GqeGKIxU6J
This is the Holodeck and why it is different than the Metaverse. Mark Zuckerberg’s metaverse can’t do this. Sadao Tokuyama (@tokufxug) 動画の1シーンレベルでの4DGS 動的シーン生成の新手法『TRiGS』 ガウシアンの出現・消失によるメモリ爆発を防ぐため剛体運動を統合。 ガウシアンの寿命を維持し無制限なメモリ増加を抑制します。 600〜1200フレームの長い動画でも高い忠実度と安定性を実現。 — https://nitter.net/tokufxug/status/2041061263778943035#m
→ View original post on X — @scobleizer, 2026-04-06 09:12 UTC
By
–
C'est ce que j'avais proposé a un hackathon polytech aux étudiants. T'arrives a 3h du matin, y'a personne et t'attend que le feu passe au vert. Alors qu'avec la 5g y'a la possibilité que les feux voient, détectent les vehicules a l'arrêt au croisement, et communiquent pour