Kevin Loo (@kevin_loo) header
Kevin Loo (@kevin_loo) profile picture

Deploying Sovereign AI at Looski MTS @ Sendblue | Ex-CTO @ Easy Street Offers

SF, NYC, SZ, HK, LA
Joined August 2011
Photo by Kevin Loo (@kevin_loo)Photo by Kevin Loo (@kevin_loo)Photo by Kevin Loo (@kevin_loo)Photo by Kevin Loo (@kevin_loo)Photo by Kevin Loo (@kevin_loo)Photo by Kevin Loo (@kevin_loo)Photo by Kevin Loo (@kevin_loo)Photo by Kevin Loo (@kevin_loo)Photo by Kevin Loo (@kevin_loo)Photo by Kevin Loo (@kevin_loo)
i was chatting with a bank yesterday they visibly shivered when i said models that'll run on prem will be chinese, even though we're discussing about air gapped deploy really glad american firms are still trying to win the open weight race
8
so is riemann hypothesis solved or not?
1
26
Just cut disk use for Inkling 2 Small’s weights on four DGX Sparks from 1.56 TB to 432 GB (72% less). Each Spark stores only the weights it reads, in the file layout the loader expects. Every loaded weight matched the original byte for byte. All 14 test answers matched too. check it out github.com/looskis/labski/tr…
26
Gemini 4 Neon is insanely good It's replacing some of my Opus 5.5 use
27
omarchy booted & has video now on m6 mac mini
3
4
151
For cpu & gpu to work Ane next
3
DO NOT RETURN YOUR M5 ULTRA 256+ STUDIOS sell them to me
1
67
realized gf never has to worry about me cheating you know why? just had a quasierotic reaction to plugging in qsfp cables
1
149
Cabling makes my eyes twitch Some fun things coming!
58
I’m buying m5u in a panic
苹果终于对 Ollama、LM Studio 和 Exo 下死手了。🧵 刚从 App Store 泄露出来的苹果内部新应用 Harbor,直接把“多台 Mac 组建本地 AI 推理集群”做成了官方傻瓜式客户端。 以前你想拿手里的两台 Mac 跑个满血大模型: 配 SSH 免密、折腾 Exo 节点发现、手动编译 MLX、调通信延迟,最后发现千兆网口根本喂不饱带宽,跑出来的 token 慢得像幻灯片。 现在苹果的套路很直接: 在 macOS Tahoe 26.2 塞进 Thunderbolt 5 的底层 RDMA 驱动,更新 MLX 0.30.1 的 jaccl 通信集合库,最后甩出一个叫 Harbor 的 GUI 壳子。 几根雷雳 5 线插上,自动发现设备,直接把几台 Mac 的统一内存池化成一个巨大的推理显存池。不敲一行命令,数据 100% 不出本地硬件。 最耐人寻味的是,这玩意儿挂在苹果企业软件子公司 Claris 名下内测,而不是消费级的 Apple Intelligence。库克算盘打得很响:你不放心把代码和财报喂给云端 API?那来买四台顶配 Mac Studio 自己搭私有推理机房吧。 做通用模型套壳工具的创业公司,永远不知道操作系统的下一个补丁会顺手把谁干掉。 如果官方原生支持多机内存池化,你现在手里的本地推理工作流会直接切过去,还是继续死守 Ollama / vLLM?👇
1
131
Got 6 nvidia dgx sparks today Bottleneck is no one has stacking cables smh
1
3
132
Kevin Loo retweeted
Retell just crossed $100M in revenue run rate. $10M to $100M in 17 months. Profitable. $4.6M raised. From day one, we’ve focused on solving the hardest parts of deploying voice AI in the product itself, so customers can build, manage, and improve their own agents without depending on us. Our bet: the future of the contact center belongs to the teams who run it. Proud of our team. Grateful to every customer who believed in us.
485
555
371
1,628
6,662,881
net trick if upgrade from 200 -> 500 plan, your weekly resets ;)
53
We're giving away a Hogwarts Legacy custom wrapped NVIDIA GeForce RTX 5090 to celebrate its official release with DLSS 5 🪄 To enter: 🟢 Share this post 🟢 Comment #RTXON Terms and Conditions: nvidia.com/en-us/geforce/con…
84,476
71,939
7,700
66,886
6,056,579
fun fact: @ycombinator wifi network detects and shuts down connections with large egress happened @supabase select 26 hackathon when i manipulated ~300GB of climate data, and wanted delegate to a mac mini worker kept getting disconnected lol
83
yc w26 request for startups asking for air gapped deployments basically what i've been working on for 3 months gov't & compliance-first sectors are so behind ycombinator.com/rfs
3
3
128
this is where we're at nowadays jetkvm connected to an instance, controlled from a vnc over tailscale on my laptop from 300 miles away
69
i get a lot of my work done at night bc i swear usages go down slower at night
1
1
57