Image created with Gemini. Image prompt: Vintage book-jacket illustration of an Edwardian gentleman in round wire spectacles and straw boater posting international letters into a tall cast-iron pillar postbox, a folded world map behind him whose continents flow into Dublin streets, bold serif title reading INTERNATIONAL across the top, flat inked Shakespeare-and-Company blue, cream, lemon yellow and slate gray palette with a warm brick accent, letterpress paper grain, soft overcast June daylight, landscape 16:9.
AI in spotlight at G7 as Trump, world leaders joined by tech chiefs
https://www.cnbc.com/2026/06/17/g7-trump-ai-tech-leaders-openai-anthropic-google.html
Anthropic Google DeepMind CEOs call for U.S.-led AI coalition at G7
https://www.cnbc.com/2026/06/17/anthropic-amodei-google-hassabis-us-ai-coalition-g7.html
Axios reports that Microsoft is considering a Microsoft-hosted version of DeepSeek V4 as a cheaper model option for Copilot Cowork. Microsoft says Copilot Cowork can’t work on unlimited pricing. “We have users who do hundreds of tasks a week… but the consequence is the costs
https://x.com/kimmonismus/status/2066946013026263110
[2606.17030] Qwen-RobotWorld Technical Report: Unifying Embodied World Modeling through Language-Conditioned Video Generation
https://arxiv.org/abs/2606.17030
📣 Introducing the Qwen-Robot Suite — Qwen-RobotNav, Qwen-RobotManip, Qwen-RobotWorld, three foundation models, a full stack for embodied intelligence. 🧭 Qwen-RobotNav — the gateway to mobility. • Unifies 5 navigation tasks in one model: instruction following, point-goal,
https://x.com/Alibaba_Qwen/status/2066870197122899980
Chickenhawks and rationalists want Dario to win, the former because Chyna, the latter because Chyna + better Anthropic than OpenAI/xAI (not sure what they think on Google). Also, consumers want Fable back. But… post-consumer market, baby. And on China, Trump feels secure enough.
https://x.com/teortaxesTex/status/2067415696984678422
China too quick with it. Huawei users can natively capture a 3d gaussian splat and set it as their wallpaper.
https://x.com/bilawalsidhu/status/2066617540923490439
The le chaton fat meme is leaking to the outside world and I expect to be asked about Mistral’s new ginormous cat model with infinite benchmark scores at my next meeting with corporate leaders. I guess it is better than being asked about the “”MIT pilot AI study.””
https://x.com/emollick/status/2066578565076746667
ByteDance’s Seedance 2.0 Mini is live. A faster, more accessible way to create videos from image, video, and audio references. Built for everyday production workflows at about 50% of Seedance 2.0’s list price.
https://x.com/BytePlusGlobal/status/2066531645138833586?s=20
A good time to remind people that in my time doing LLM research I feel like a minority of my colleagues are American citizens. It would be industry destroying to have to rebuild with segregation for frontier ai research to be legal.
https://x.com/natolambert/status/2065616536942088581
Has there been anything good written about the failure of Mistral to keep up with both the Big Three and Chinese labs? They have talent and national backing, but despite being Europe’s only frontier lab (Google Deepmind’s UK lab aside), they haven’t been able to close the gap
https://x.com/emollick/status/2065837438388326474
Canada’s leading LLM company, valued at $20bn, is massively expanding its presence in London! It is tripling its London headcount. After acquiring German Aleph Alpha the company is positioning itself as a “”Sovereign”” alternative to OpenAI and Anthropic (who have also announced
https://x.com/SebJohnsonUK/status/2066817307146330559
We will see more of this later this year Chinese labs are now really starting to hill-climb the way we have seen monthly improvements from OpenAI and Anthropic I still want to see super long-horizon tasks for open models, but there’s already a bunch of good coding benchmarks
https://x.com/scaling01/status/2066941781506232507
China’s universities cut 12,000 ‘obsolete’ degrees amid race to embrace AI era | South China Morning Post
https://www.scmp.com/economy/china-economy/article/3356913/chinas-universities-cut-12000-obsolete-degrees-amid-race-embrace-ai-era
Meta Starts Unwinding Manus Deal by Splitting Operations, Data – Bloomberg
https://www.bloomberg.com/news/articles/2026-06-11/meta-severs-manus-data-access-after-china-orders-buyout-unwound
Mistral AI to get Code and Apps features on Vibe
https://www.testingcatalog.com/mistral-ai-to-get-code-and-apps-features-on-vibe/
@yacineMTB deepseek v1 -> v3 (no details in v4 about this) and k2 don’t use muP and instead use naive N(0,0.006) initialization. so how do they do hyperparam selection? they basically fit scaling laws to get optimal batch size and learning rate. there are a bunch of papers detailing this
https://x.com/eliebakouch/status/2066594560365498695
🌘 Kimi-K2.7-Code, our latest coding model, is now released and open-sourced! 🔷 Improved coding & agent performance over K2.6: +21.8% on Kimi Code Bench v2, +11.0% on Program Bench, and +31.5% on MLS Bench Lite. 🔷 Reasoning efficiency: Less overthinking, with 30% lower
https://x.com/Kimi_Moonshot/status/2065377579130142937
🎉 Congrats to @Kimi_Moonshot on Kimi K2.7-Code, a coding-focused agentic model built on K2.6. ✨ 1T-parameter Mixture-of-Experts, 32B active per token ✨ MLA attention with a 256K-token context window ✨ ~30% fewer thinking tokens than K2.6 for more efficient reasoning
https://x.com/vllm_project/status/2065427423148318747
🔗 Weights & code:
https://x.com/Kimi_Moonshot/status/2065379671039189317
289 tokens per second. That’s how fast CoreWeave Serverless Inference serves Kimi K2.7 Code. Per @ArtificialAnlys, we serve it faster than any other provider at the best price-performance. How we did it 🧵
https://x.com/CoreWeave/status/2067613387056709982
It is a good time for moonshots. AI has reached a level where there are transformative projects that could result in huge social good, but require public R&D, consensus & transparency to pull off. Examples: universal tutors, co-scientist/replication systems, remote medical help.
https://x.com/emollick/status/2066534666257973523
kimi code web is back, yeah
https://x.com/bigeagle_xd/status/2067426798493921351#m
Kimi K2.7 Code scores higher than K2.6 on benchmarks while using ~30% fewer tokens. It’s the first time Moonshot has put “”Code”” in a K2 model name. Every other K2 release was a general purpose agentic model. Try in Cline now! (good model to put in our upcoming subscription 👀)
https://x.com/cline/status/2065473287761891621
Kimi-K2.7-Code lands as the #3 open model in the Code Arena: Frontend, #19 overall. In the Code Arena: Frontend models are evaluated on agentic frontend coding tasks from real users building apps and websites (HTML and React).
https://x.com/arena/status/2066616607380828401
Moonshot released Kimi-K2.7-Code it looks to be a decent step up in performance, but also efficiency
https://x.com/scaling01/status/2065460210584420510
moonshotai/Kimi-K2.7-Code · Hugging Face
https://huggingface.co/moonshotai/Kimi-K2.7-Code
Today marks the beginning of our launch calendar and to celebrate i am making ncode and our flagship kimi k2.7 model free to use for the next week (or until the traffic knocks us out). all you need to do is: 1) sign up for a noumena account at
https://t.co/XoNsZeW8sw 2) go to
https://x.com/_xjdr/status/2067741647941832818
You can now run Kimi K2.7 Code locally! 🌘 We shrank the 1T model to 325GB (-48%) via Dynamic 2-bit where important layers are upcasted. Run at >40 tok/s on 330GB RAM/VRAM setups. Run full precision on 610 GB. Guide:
https://t.co/SXZJ3IHMpY GGUF:
https://x.com/UnslothAI/status/2066492839450800427
I benchmarked Kimi K2.7-Code (1T MoE, coding-specialized, just dropped) on KernelBench-Hard, against its general predecessor K2.6 and the current board-topper Claude Fable 5. Disclaimer, this is just measuring on kernel optimization ability, which is a very small chunk of
https://x.com/elliotarledge/status/2065443474560946615
By treating natural language as a universal action interface，Qwen-RobotWorld bridges the gap between general video generation models and domain-specific embodied models — this converts end-effector poses, steering commands, and navigation waypoints into a single interface,
https://x.com/Alibaba_Qwen/status/2066870210716647591
I made Physical AutoResearch sound simple (conceptually), but it took a village to pull off and lots of design thinking into the robot /loopcraft. The hardest part is everything we need to setup *before* pressing Enter. Here’s a behind-the-scene tour: 1. Safety harness Letting
https://x.com/DrJimFan/status/2067283904986517866
Leave a Reply