MiMo-V3 is getting a new architecture. The core of it, HySparse2, is out today.
Less prefill, a smaller KV cache, better long-context retrieval—and we got all three at once.
Compared with MiMo-V2.6's Hybrid SWA architecture:
• 5.02× lower prefill FLOPs at 1M tokens
• 4.5× smaller KV cache at 1M tokens
• Better MRCRv2 and RULER-v2 scores, plus lower AgentPPL and LongPPL
Why build a new architecture?
Agentic inference is a very different workload. Each round, a short action can return a long observation that needs to be prefilled, while the context keeps growing. That puts prefill cost, KV-cache size, and retrieval accuracy on the critical path at the same time.
HySparse2 tackles all three with two levels of KV sharing:
• KV Bridging: Following YOCO, full-attention layers in the cross-decoder build their K/V from self-decoder hidden states.
• KV Reuse: Within each hybrid block, sparse layers reuse the preceding full-attention layer's KV cache and selection indices.
Two more changes: token-level selection replaces block-level selection, and a forced window of recent tokens replaces the separate SWA branch, so local and global tokens share one KV cache. Since all cross-decoder KV caches now come from the self-decoder, prefill can stop once the self-decoder finishes.
Paper: https://t.co/REeEdd5pL7
.@johnternus acquisition list:
1. @perplexity_ai = search / default LLM
2. @WisprFlow = siri
3. @Superhuman = native AI office suite / mail app / Grammarly is new keyboard on iPhone
4. @exolabs = local open source AI
5. @RobinhoodApp = finance platform (not for sale I know)
6. @Hello_GoAI = enterprise AI
7. @digitalocean = create an AWS competitor
8. Disney = obvious
9. DoorDash and uber = get into the real world of logistics! Bold move!
Spend that @apple war chest on big swings — only takes one breakout
🎉https://t.co/IlsYsJWbHd 设计升级完成,进入导航站美观度第一梯队,答应大家的设计升级的实现路径公开如下👇🏻:
1、去 Dribbble 找到想复刻的风格示例图,从里面找喜欢的设计元素,字体,圆角,风格等,并不是去找一个与你产品一样的设计稿
2、将示例图丢给 Claude design,让它提取整站风格的基础元信息,然后让它重新设计一个 Hero 页面,设计过程中你可以去基于自己喜爱的去微调设计元素
3、基本设计风格定稿后,就可以让 Claude Design 根据你在 github 的代码,去阅读你网站里面的具体代码,逐渐实现其他页面
4、再让 Claude Code 接管 Claude Design 去做代码实现,这是可以直接导出的,而我在实现的过程中,甚至只开了 Opus Medium 就完成了我的预期设计目标
总之,核心就是目标示例图 + Claude Design + Claude Code 之间的联动关系,让实现一个高质量的设计网站,变得异常容易。
我喜欢设计好看的网站,我对好的设计的看法是: 不是说你得有多么有想象力,创意的设计,而是整站风格,字体,边距,动效,配色,包括字体大小差异,让大家一眼看起来浑然一体,不会觉得难受割裂。
Two days ago, we received an update from our shipping provider, ShipMonk. We're deeply saddened to share the news that the recent data breach affects more customers than originally thought.
Another 67,000 customers from the US who ordered between November 2019 and August 2021 were affected, with their full details (name, email, phone number, shipping address, order number) exposed.
All affected customers have been emailed directly. If you didn’t receive an email, then you are not affected.
Throughout our entire relationship with ShipMonk, we repeatedly requested and received written assurance confirming the deletion of the data, in line with our contract, data policy, and past communications. We are very disappointed that, despite receiving this confirmation, the data was not deleted in their systems.
Trezor systems were not compromised, and your device is secure. But please be alert for fake emails, phone calls, fraudulent letters, and potential risks to physical security.
NEVER share your wallet backup with anyone or type it into a website.
We’re terribly sorry to everyone affected. We take this matter very seriously and are working to ship anonymous delivery ASAP, so you can protect your personal information when placing an order.
For more information, visit our blog: https://t.co/3KnO4LyBKb