Lukas (❖,❖)
4.1K posts

Lukas (❖,❖) retweetledi

Chapter 2 🦄
Want to move from FCFS to WL, or to GTD?
UniPix now has a simple rating system for everyone who supports the project.
If you really want to be part of UniPix, prove it.
unipix.pro

English

@m_murfy @nChainGlobal I am happy to pay a license fee of [checks own address on etherscan] 100 Alamo tokens (ALMO). Please provide the address and I will send the ALMO your way.
English

If you are still holding #ETH then this is your FINAL WARNING! As of yesterday, Blockchain-enforced Smart Contracts has been patented by @nChainGlobal for use on #Bitcoin. @VitalikButerin what do you think is a fair licencing fee? #BSV #BitcoinSV #Ethereum patents.google.com/patent/EP32571…

English
Lukas (❖,❖) retweetledi

世界是一坨巨大的SHIT。我就是SBTI的偾世者。
最近SBTI挺火,感觉是MBTI的反抗版本,不被定义的自己版。当然,国内好像已经封禁了。
测试传送门:
github.com/Henrryfeng/SBTI

中文


x.com/ollama/status/…
Mac用户狂喜!Ollama重磅更新:
Apple Silicon上本地AI飞起,媲美云端!
Ollama正式接入Apple的MLX机器学习框架,
在M5/M5 Pro/M5 Max芯片上实现史诗级加速!
实测Qwen3.5-35B-A3B模型(NVFP4量化):
- 预填充速度高达1851 token/s
- 解码速度达134 token/s
这意味着什么?
个人助理OpenClaw响应瞬间到位,
Claude Code、OpenCode等编码代理写代码如丝般顺滑,再也不用等半天!本地跑大模型,终于从“能用”变成“真香”。
更牛的是:
- NVFP4格式支持:精度几乎不丢,内存占用大幅降低,生产环境一致性拉满。
- 缓存大升级:跨对话复用、智能检查点、智能剔除,agentic任务(多轮工具调用)内存更省、响应更快,尤其适合Claude Code这种分支频繁的场景。
Mac用户只需32GB+统一内存即可爽玩(16GB暂不支持大模型,期待后续优化)。更新后,消费级硬件与云端推理的差距进一步缩小——隐私保护、无限调用、无额外账单,开发者生产力直接起飞!
一条命令搞定:
ollama run qwen3.5:35b-a3b-coding-nvfp4
或直接启动Claude Code / OpenClaw。
本地AI时代,Mac终于站上C位!MacBook用户快去更新Ollama 0.19,感受“丝滑”吧~
中文
Lukas (❖,❖) retweetledi




















