Keep on to blur preview images; turn off to show them clearly
现在都流行大模型月更吗? MiniMax-M2.1 实测! 申请 MiniMax-M2.1 内测通过了, 给大家带来 MiniMax M2.1 的测试结果: 本次测试较 MiniMax-M2 从编程, Agent能力, 长上下文召回能力均有不小的提升. 尤其是Agent能力和长上下文召回能力达到了摧枯拉朽的程度, 直接叫 MiniMax-M3都不过分. MiniMax M2.1 在 24 小时总计 300 回合的硅基骑手测试(让大模型调用 tool 送外卖)收益达到了 419.77 元, 总计执行了392次 tool call, 测试使用了大约56%的上下文空间, 这个范围内工具调用均表现良好. 作为对比MiniMax M2, 它的收益是285.27元, 不过只使用了32%的上下文空间, 模型就停止使用 tool call, 不断重复上文了. MiniMax-M2.1, 在192K长度上下文上有94%的召回水平, 而 MiniMax-M2 只有 52%. 可以说是摧枯拉朽的提升. 另外几个大家熟悉的编程水平测试上也有不同程度的提升. 这次的新模型特别适合大型Agent任务, 有这方面需求的朋友可以试试看. #MiniMax #MiniMaxM21 #ai写代码 #aiagent #KCORES大模型竞技场
A coder, road bike rider, server fortune teller, electronic waste collector, co-founder of KCORES, ex-director at IllaSoft, KingsoftOffice, Juejin.
兄弟们，这个有点牛P啊 阿里巴巴发布 Photoshop 级别的图像分层模型： Qwen-Image-Layered 它可以将将单张图像自动分解为多个独立的 RGBA 图层 让你像在 Photoshop 里一样编辑图片，但不需要自己动手抠图。 由于各图层在物理上相互独立，分解完成后，你可以对每一层都进行编辑。😅 这种方式使得： 每个对象（人物、物体、文字、背景）都能被单独编辑； 图层之间天然隔离，不会互相干扰； 操作（移动、缩放、删除）都不会影响其他层。
1、重新着色：你可以只改某个图层的颜色，比如给女孩的衣服换成蓝色，而背景、头发都不会被影响。
RT @AlessandroLavis: wan 2.2 is starting to feel unfair you drop in a single reference video swap the face with one image and the system r…
Building aismarty, Ai agents and LLM Sauce. Unix Nerd + dev 30yr .Now C++, PHP, React, JS, Python, Typescript++ SAAS fanatic .Fun=PPL Pilot NR IR(R)
15K wpm at 40% yield is for 910C type chips (presumably) is quite a lot of chips. Using previous calculations: that's 34 wafers/die, at 10K wpms that's… 170K chips. 2 million a year. Lower bound
otoh 2M coincides with "aggressive" target from hawks that said, if they reach 10-15K wpm by end of 2026, it's <2M total (for SMIC N+2 alone though)
RT @jackfriks: new app is live :))) this is a family and friends version of the couples app i made in october i’ll be using it with my mo…
curious guy creating things @ https://t.co/HXWladhJaA - up and coming wife guy
Selling a business doesn’t mean walking away. Jamm Designs (@jammdotco) sold to a buyer on @acquiredotcom who could scale the business so the founders could focus on what they loved: design. First acquisition with lots of unknowns and having the right advisor made it smooth from start to finish. That’s a real win win. Check out the full story here:
Founder and CEO of @acquiredotcom. https://t.co/wRMIssDmhl has helped 1000s of startups get acquired and facilitated $500m+ in closed deals.
