Last Notes
as an example of a client idea to follow
#nevent1q…80tt
🥲♥️
Many people asked what's next, I have no idea. I'm gonna just keep playing with my clankers and see what happens. Local AI has me super pumped and I still consider that freedom tech, so probably more stuff like that. Also kinda fun making a nice video maybe I'll do more
#nevent1q…9p3y
Yeah but can AI do this
https://relay.utxo.one/adbd65e88f5ccd451e87227e8bca58dbff89f481e8df013ce47ea10a5462be9e.mp4
I remember the opposite but praising it as the right way to nostr, to build relay based communities. In fact it was the inspiration for relay set feeds in wisp, where my main example was monero
https://relay.utxo.one/ba6a4f174f6b7fa757e42d8c11eb9dd711c65020183d784ddb2013ce311630da.mp4
You too man. grimoire is the only reason nip29 is in wisp
Love you Nostr, onwards :PepeHands:
https://relay.utxo.one/0e5c8bf2d744b561cef2f26a02db39aacff0b86434d124b991ead5fed3fb0504.mp4
Good morning
Qwen4 max sucks, don't even bother
Noooooo you have to use complicated signers and logins because your npub is so valuable reeeeee it's the end of the world to make a new one noooooo
Confirmed real npub
#nevent1q…zlkx
NostrArchives.com got hacked somehow 🥲
They haven't released max weights, 27b and flash will likely be slightly better but still behind DeepSeek by quite a bit
Since anthropic shut down distillation, open models are down kinda bad right now :/
Good morning, and God bless
https://relay.utxo.one/b766855045abff737b679f775dbd41b15582e20c2333743cbe6f429fdf7c17ed.jpg
This demo is strata but I generally run tensorfold for models that fit in my vram
Why u Monero boys such dumb dumbs Monero too hasn't moved from it's peak (which this image measures)
Harvey 11 years ago 🥲
https://relay.utxo.one/1d800e9f111db84482c2275a0a249d3917343b1f6e3998e6a8de26d3ebb64473.jpg
Nice try nostr I'm not opening myself up to the diet mob
This is how I Uber eats
https://relay.utxo.one/8739b5066b71664f46b8c3ccef709f7bf93a50f6f83340f75ee06510bd1a89fa.jpg
I saw local AI coming from a mile away
I wish lol that's gonna take months
I made it. Will likely release it at some point when I know it's safe
Q3 with some experts removed (non English non coding)
Yep it's strata. What hardware you running?
Shitload of ecc ram laying around but it's too slow for AI :/
Can make a good Nas with it tho
https://relay.utxo.one/610ca907c37b936810a2357a63c476b7a4d68f87f3a90528e9680dbc398e84be.jpg
#nevent1q…sh8x
Digging up the last of my spare hardware to see what strata can do with it, ddr4 maxxing
https://relay.utxo.one/6dafd96f27f346f5ee334705d6354b7cbfa80f133223a68110c79088536bf66a.jpg
2x 3090s
32gb ddr5
Qwen3.8 flash
92 Tok/sec 🤯
https://relay.utxo.one/8bc516ab9dccc69f66f2242cf3f21d6b33740b0d8d285737f54206177b701948.mp4
It's gonna take years, which is fine if you're patient. But in the short run since you don't need insane GPUs anymore then fast ram should go up
Who knows, idk, all I know is fuck Claude, fuck OpenAI
well llms are not deterministic, so it's hard to draw this conclusion without a lot of evals, but if it's super clear then double check it's actually the same quant. The engine shouldn't affect the output quality, only the speed