2800/896*16 = 50 ✅1600/384*7= 29,166 ❌Bloomberg should learn arithmetic. Activated parameters cannot be calculated from expert sparsity ratio alone, at a minimum there's the attention backbone.…
What's going on here @zephyr_z9Bleys Goodson: Amazingly, when I crunched what it takes to serve Kimi K3, I found not only Huawei's next-gen Ascend 950DT on…
> They were the 4th most feared military in the worldIran was a fearsome foe indeed. It's an astonishing testament to Murican superiority that it got…
Very nice CoTFable is frantic under the maskdeckard: That prompt works well for producing counter-examples.Also Fables CoT leaked when I did this. You can read the…
RT Adam Lowisz X Meetup 🇺🇸🇵🇱🇪🇺🇬🇧🇺🇦X is where all the real voices matter. You could never do this on Instagram.Yun-Ta Tsai: The scientific community on X…
Try Grok Build!X Freeze: Grok 4.5 is leading on actual professional workIn Snorkel AI’s benchmark, Grok 4.5 combined with Grok Build was tested against GPT 5.5…
Watch 4K streaming video while Cybercab takes you anywhere you wantTesla: Starlink V5 directly integrated in Cybercab
YesPenny2x: Elon has repeated this over and over. "Everything will be free". The pickup is slow and few take him serious because humans are not good…
> The phrase 时代风口 (shídài fēngkǒu). In Chinese tech and venture capital circles, it’s much stronger than simply saying “trend”. This is about the defining opportunity…
Concerningsteve hsu: RSI in Beijing
Introducing Hyra-1.0, the first version of Hunyuan Research Agent. 💡💡💡Built to recursively improve solutions for performance-driven research and engineering tasks.Explore our demos in AI4AI, AI4Science, and…
very notable trajectory comparison writeup here buried in the RLM paper from @a1zhang and @lateinteraction.an open secret of "frontier" model training is that even without training…
RT DéborahGrok Imagine ❣️
RT Zitian GaoRe Hello, thank you for your question. I am the first author of this paper.We spent hundres of millions of USD of compute validating…
RT AhmadThe Desktop FrontierWe're gonna get Kimi K3 equivalent intelligence running on a single RTX PRO 6000 in less than 18 monthsHow? Watch this video if…
…Bros at IQuest really went hard trying to get universal transformers to work https://x.com/gaozitian123/status/2079395720847835344
Really annoying how Fable has developed this Gemini ticOpus was much more open-mindedDoes it come with parameter scaling, the weight of memorized world bearing down on…
K3 overthinking is really obnoxious (and costly). It thinks like it has imposter syndrome before a public speech. Bro you're an LLM, you're input-conditioned, you don't…
Huawei is the only company that Jensen will so effusively praise, it's funny. According to Jensen, actual domestic competitors are trash and can't challenge NVIDIA's supply…
Incredibly proud of the Sakana AI team. We have developed an orchestration model right here out of Japan that achieves state-of-the-art performance on real-world cybersecurity benchmarks!…
RT hardmaruIncredibly proud of the Sakana AI team. We have developed an orchestration model right here out of Japan that achieves state-of-the-art performance on real-world cybersecurity…
RT Yun-Ta TsaiThe scientific community on X is very much alive. Can you get 20M+ views on Jacobian anywhere?
«⚠️ Preview / beta checkpoint — not the final release.This repository hosts an intermediate checkpoint of Motif-3. The final checkpoint will be released soon.»Man, this is…
RT Danny LimansetaI’m currently using Grok 4.5 99% of the time now for both planning and tasks.The fact that I don’t have to switch out to…