Starlink 확장, Grok 생태계, 에이전트 학습과 로컬 모델 최적화

SpaceX has officially crossed 11,000 Starlink satellites in orbit. They've added 1,300 satellites to the constellation so far this year. SpaceX requested FCC approval last month to launch and operate a third-generation satellite constellation of 100,000 satellites, so they are just getting started.

Starlink is becoming a major part of commercial shipping too Hyundai Glovis has now installed Starlink on 45 of its 47 company-owned ships The rollout started in January and now covers almost the entire fleet, including car carriers and bulk carriers Crew members say Starlink has significantly improved connectivity at sea for: • Video calls • Streaming • Work-related data transfers • Real-time ship-to-shore communication • Remote technical support Hyundai Glovis also plans to use the connectivity for AI-powered predictive maintenance and smarter vessel operations SpaceX’s enterprise team recently visited the Glovis Stella in South Korea to see Starlink operating in real-world conditions and hear directly from the crew Starlink is becoming core communications infrastructure for ships operating across the world's oceans From airlines to cargo ships, SpaceX is connecting transportation system everywhere

Elon Musk

@elonmusk

Grok 4.6 is #1 on healthcare questions

Grok

@grok

Yes. One account, shared cloud computer, up to 50 bots. Give each a clear role and charter, connect tools once, teach recurring workflows by demonstration, schedule or trigger them, and they coordinate in group chats. The one-person company runs 24/7. Download at http:// x.ai/bot.

Very interesting new work from Microsoft. (bookmark it) This work is related to this emerging theme of leveraging harnesses for model post-training. Modern agents run inside a harness that owns tools, context, and control flow. When you train them, the harness owns the environment loop and the trainer only sees LLM request and response pairs. How it works. Agent Lightning v1.0 connects any harness to RL through an endpoint proxy in about 3,500 lines, then works through what breaks in that setup, retokenization, sample merging, advantage calculation, loss normalization, and backend scheduling. Using 6K training examples and modest compute, it moves Qwen3.5-9B on SWE-bench Verified from 41.8% to 56.4%. Paper: https:// arxiv.org/abs/2608.17528 Track more trending AI papers in our academy: https:// academy.dair.ai

NVIDIA AI

@NVIDIAAI

We benchmarked 300+ NVIDIA verified skills to see how much they actually help agents on real tasks. Same task, same model, same setup. The only difference was whether the agent had the skill. Across the benchmarks, skills improved correctness by 41 points, effectiveness by 39, and efficiency by 35. SkillEvaluator is open source if you want to test your own skills before you ship them.

We’re releasing new Qwen3.8-27B GGUFs with 10% higher accuracy. Unsloth Dynamic V3 outperforms others by >10% on Div-300, KLD & more benchmarks. We also release 1-bit quants that retain 77% accuracy. Run on 8GB RAM. Blog: https:// unsloth.ai/docs/basics/dy namic-3.0-ggufs … GGUF: https:// huggingface.co/unsloth/Qwen3. 8-27B-GGUF …

witcheer

@witcheer

a new open-weight model family where the model card ships with Hermes Agent setup out of the box. point your Hermes Agent at a local Ornith 1.5 server, 9B to 397B! benchmark coming soon :)

Oliver Cameron

@olivercameron

In The Matrix, it becomes obvious it’s a simulation when the black cat appears twice. World models fail similarly: visually the generated simulation looks realistic, but the underlying distribution is wrong. Introducing CaliBench, a measurement of the randomness of reality!

Today, we're introducing CaliBench, evaluating whether video world models reproduce the true randomness of our universe. Roll a dice or pick a card, the outcomes produced by a world model should match reality. Find out if do!

MiniMax H3 is now unlimited on Runway. For a limited time, Max plan users can generate as much as they want with no caps, no counting, just unlimited access to one of the best video models available. Get started now at the link below.

1- You can have Grok speak in your language; he speaks perfectly with the right accent. (here in french) 2- You can build a long scene in 1080p by taking the last frame of an animation to start another animation. It's easy with Grok (Right-click on the animation and copy the image you want.)