Andrea Giammarchi 🍥

Web, Mobile, IoT: all JS things since 00's - opinions are my own ™
Principal SW Engineer @anacondainc working on OSS stack 🦄

[Andrea Giammarchi 🍥 @WebReflection]

coincident/server on IoT via Raspberry Pi Zero 2 W?

Sure!

https://t.co/XaM7c7dgy4

[Andrea Giammarchi 🍥 @WebReflection]

an easy win for slow SD cards while writing: do not ever copy and paste while still copying the previous thing as these cards suck big time in write parallelization so that whatever slow Mbps they have in writing will degrade even further if the operation is not unique per write

[Andrea Giammarchi 🍥 @WebReflection]

@MiaAI_lab @garetjax if only microtick would be on X reading this space …

[Andrea Giammarchi 🍥 @WebReflection]

@MiaAI_lab @garetjax The tower would be switch in the middle, 2 sparks under, 2 sparks above, that’s it … I hope somebody is reading this space

[jhey ʕ•ᴥ•ʔ @jh3yy]

put together a growing set of page transitions using css view transitions 🎲

copy-paste/copy for agent 🤖

check 'em out ↓

[Andrea Giammarchi 🍥 @WebReflection]

@MiaAI_lab @NVIDIAAI @NVIDIARTXSpark switchless 5? 😁

[Andrea Giammarchi 🍥 @WebReflection]

@arvidkahl Sadly enough, if unified, it’s still less than 4 DGX Sparks in a cluster, but then maybe the CPU/GPU/NPU together won’t produce same performance 🤷

[Andrea Giammarchi 🍥 @WebReflection]

@thekitze that S suffix though … 😅

[danny lin @kdrag0n]

Macintosh Portable (1989), the first Apple laptop

running real System 6.0.8 with MacPaint + HyperCard

on the first Apple foldable phone

[Andrea Giammarchi 🍥 @WebReflection]

@Paul_Kinlan that sounds like an Omarchy plugin though ... doesn't it?

[Andrea Giammarchi 🍥 @WebReflection]

@typicode @zkochan I've done the same with a SanDisk 500GB NVMe external HD and I could even run Steam games without feeling any slowness at all ... ArchLinux, GNOME, boots in seconds, runs almost everywhere. The only caveat is having all linux kernel + mesa / GPU drivers for updates, yet awesome!

[Andrea Giammarchi 🍥 @WebReflection]

@wesbos @filearts that's what I meant, if there could be a "provide your AI provider/API key" to then take over the conversion it'd be awesome 🥳

P.S. I could've tried myself but on iPhone 13 mini there's no such thing as LIDAR 😢

[Andrea Giammarchi 🍥 @WebReflection]

@adibhanna @wesbos I think this is the missing primitive. Once you have the model anything else can be easily done without needing features creep embedded?

[Andrea Giammarchi 🍥 @WebReflection]

@munawwarfiroz the pyramid of "are we there yet" just keeps growing ... agents also ask "are we there yet" to processes polling possible results because software is rarely hooked in a "reactive" way 🤷

[Andrea Giammarchi 🍥 @WebReflection]

I'm asking AI "are we there yet?" endlessly and I think that's not a good thing to do but hey ... every single time a different answer, situation still stuck/not solved ... wonderful

[Andrea Giammarchi 🍥 @WebReflection]

@alexellisuk *my* time has produced this and I hope you'll consider/appreciate the concrete feedback https://t.co/AmhSUoOHya

the TL;DR is that yes, OpenCode v2 offers even GLM 5.3 Flash itself to work on stuff and that was the key (very meta, since it fixed GLM 5.3 Flash for me).

thanks 👋

[Andrea Giammarchi 🍥 @WebReflection]

@alexellisuk how to DM you? Meantime, once switchless-nccl is installed, both Nvidia sync and Sparkrun are incapable of making it work, any quick hint/pointer ??? Your fabric-setup won’t help neither

[Andrea Giammarchi 🍥 @WebReflection]

Nvidia Sync is the best UI friendly tool that works seamlessly (way better than sparkrun, imho) for 2/3 switchless GB10, or 4 with a switch (extra thousand dollars minimum). 4 GB10 switchless are reality though, great patches/effort around, extremely hard to get it right though!

[Andrea Giammarchi 🍥 @WebReflection]

interesting times: today I've asked "please tell me what I am doing wrong" and the answer was "use AI to know that" ... the topic was "how to configure local AI" so we reached full circle: people are writing docs about local AI assuming you have already AI to set those docs up 1/

[Andrea Giammarchi 🍥 @WebReflection]

the simple fact I had to bother literally experts and the answer was always "ask AI to do that" made me think we forgot for many the entry point is "no AI" + local only ... f you want to promote local AI for real, please help humans being able to setup properly all the things 🥳

[Andrea Giammarchi 🍥 @WebReflection]

as (maybe not too silly) metaphor, imagine famous AI providers answering users "how do I set you up properly?" answering "use AI to do that!" ...

I do care about human readable docs. If those docs are optimized *also* for agents, that's *ace*, but it docs are agents only ... 😢