Lee

Co-founder of UnderStudio

[Lee @lvturner]

@MiaAI_lab @sfxnz I updated your refrigerator too 😏

[Lee @lvturner]

@dhh @Compilethings The trackpad on my Redmi laptop didn't work on first install - is there someone I can ping details? Haven't had much time to play with it all yet, but happy to dig out details and file reports

[Lee @lvturner]

A whole load of people in my feed quickly learning what it's like to actually be a software engineer thanks to the past few days chaos.

[Lee @lvturner]

I almost think DeepSeek time stuff specifically to troll Anthropic and OpenAI... and you know what? I'm here for it 🍿 🍿 (also, we all owe DeepSeek a huge deal)

[Lee @lvturner]

@xKaka1456 They're making $38 per hour per GPU...

[Lee @lvturner]

@oO0Fal0Oo An old friend of mine said we should follow the Sonic 2 (the game) pattern - year one you get fully loaded python (all chaos emeralds) then, it all gets taken from you and you start year two with assembly and have to earn your way back to python. I like it.

[chronark @chronark]

should I do a print run?

[Lee @lvturner]

@Octoberry Other inference providers (that are also less hypocritical) are also available.

[Lee @lvturner]

One really nice thing about having a small gym at home and being an (AI) start-up founder, is that I can literally get work done between sets.

[David Hendrickson @TeksEdge]

Nvidia lets you build a 1TB Local AI cluster out of DGX Sparks.

& Nvidia has a playbook for AI clustering them.

8 × DGX Spark
🧠 128GB each
↓
🔥 1TB unified memory across the cluster

Nvidia now officially supports manually configured clusters of up to 8 x DGX Sparks using their built-in ConnectX-7 networking.

And Nvidia says that 1TB is enough to run frontier-class models like GLM-5.3 locally.

They specifically call out
🧠 larger models
📚 longer contexts
🤖 multiple concurrent agents
👨‍💻 multiple developers sharing one local backend
⚙️ distributed development

They list models such as
→ GLM-5.3 Flash
→ DeepSeek V4 Flash
→ Qwen3.8 Flash Next

And later this month Nvidia Sync Model Launcher is supposed to make it easier.

Link in ALT

[Lee @lvturner]

@digitalix I've had a few over the years - the first that was really good was the Xiaomi Mix 4 (Chinese domestic market only) I've recently just swapped it for the Xiaomi 18 Fold - night and day difference.

[Lee @lvturner]

whoops, no, just 7 days worth...

[Lee @lvturner]

Just done analysing 12GB of agent logs... this is about two weeks worth... with a few days of low/downtime work... crazy how it mounts up

[brutoshi @brutoshi_]

Liqiang Wang (@shujishcn) got so into Omarchy that he put it on every computer at home—then his Xiaomi Pad and 18 Fold. He even donated $1,024 to Omacom.

An enthusiast who also happens to be Xiaomi XRING VP.

His full SZ meetup talk, EN/中文 subs. @omarchy

[Lee @lvturner]

@dwkos I call it CUP.

Cruel and Unusual Punishment.

Not to be mistaken with the related CUPS (Common UNIX Printing System).

[Lee @lvturner]

@skywt2003 @MaxtuneLee only macOS can *currently* do that.....

[Lee @lvturner]

@skywt2003 Yeah... my old mac swiftly got asahi linux installed on it ages ago