scumfuck flowergirl // i know a little bit about computers // priv: @Zbuildstd

17
is this what people watch in SF when they netflix and chill
When the open-source AI platform Hugging Face was hit by a cyberattack this July, it quickly became clear this was no ordinary breach because the hackers weren’t human: they were autonomous agents created by OpenAI researchers. Instadocs: AI Gone Wild reconstructs the breach as it’s never been seen before. Premieres October 12.
1
34
469
My vent art
521
11,164
88,516
970,009
@nullptr you will rue the day you joined this app BUDDY
1
13
chinese cars are unreal
7
77
1,936
imagine 3x capy sessions on there
2
71
Replying to @notnullptr
they're putting HBM4 in dog tracking chips
1
1
45
i’m running astra on dogs
1
40
turned 21 today. recap of 20: - moved countries twice - flew 82420km - went from senior to staff - started quashing my backlog of classic movies - shot a gun for the first time - shipped many things - spent a lot of time with friends and family - felt all emotions on the spectrum
22
63
1,385
HAPPY BIRTHDAY !!!!!
1
1
48
Replying to @notnullptr
oh right this works in 3 dimensions
1
1
how did i only just notice that
1
2
i genuinely just measured it at 9tk/s are we for real right now LMAO
opencode go’s glm-5.3-flash has abhorrent caching. half my requests cache properly and the other half have literally 0 tokens cached. i am sending the opencode header properly. what gives??? it’s costing a lot of usage
2
75
1,582
are these the programmer socks I've heard so much about
13
2
256
2,943
dm me your address i can mail you some!!
8
9
124
OPSEC LEVEL: doxxed over pangram (check signal ..)
13
78
opencode go’s glm-5.3-flash has abhorrent caching. half my requests cache properly and the other half have literally 0 tokens cached. i am sending the opencode header properly. what gives??? it’s costing a lot of usage
9
156
7,774
streets say 10x $1 is better value #BeastMode
1
15
I'm old enough to remember the idiots being impressed by big numbers that 4000 was child's play. Well who's laughing now.
Frontier scale, built in Europe. Nearly 4,000 NVIDIA Grace Blackwell Superchips powered the training of Mistral Large 4, now in public preview. Congrats to the @MistralAI team 🙌
6
74
6,314
definitely not mistral. deepseek v4.1 flash, the fastest non-cerebras model in the world beats mistral large 4. qwen3.8 27b, SOTA for local people can actually run, almost matches it. qwen3.8-flash-next, runnable on 1x3090 and 64gb ram, beats it. it's not a great model
3
380
Replying to @notnullptr
Likely not dax's problem. Upstream invalidates caches very fast, so you'll get 0 cached every while.
1
537
it’s not “every once in a while” it’s literally every other request
1
4
375
mistral is the new deepmind
1
33
687
Note to self: never under any circumstances get into a fight on 𝕏 with poteto. She just cannot be beaten.
Replying to @thsottiaux @bot
How did you know I was referring to dots

ALT Ai Son GIF

9
2
522
66,403
do i need to come out of retirement
29
2
543
10,920
what u need to do is fix glm-5.3-flash’s caching on opencode go
2
37
Alright, here's the deal with Qwen4. No small MoE release. An A3B was trained. The 3.8-35B-A3B was never released due to perf delta compared to dense. This time it's apparently because the A3B is purely meant as a reference model for the new Qwen4 arch.
24
11
269
36,912
have you tried the 27B?
1
861
@opencode @thdxr single-digit tps too btw idk what kind of infra you guys r hosting this on but it’s shocking
23
938
the prophecy foretold your existence, le chaton fat
Meet Mistral Large 4, aka Le Chonk. • 1T parameters, natively multimodal. 49B active. It is the best open weights model from US or Europe on aggregated benchmarks. • State-of-the-art on critical workloads, including cyber defense, manufacturing and finance and it surpasses closed frontier models on visual grounding. • Forged in Europe end-to-end and is deployable from Europe via our own Mistral Cloud infrastructure. • Available to all via API today. Working with cybersecurity partners privately. Open weights release end of October.
6
5
290
7,471
Replying to @notnullptr
has your experience with opencode go been good I was thinking about getting it, this sounds pretty bad tho
1
674
it’s the best subsidy you can get but if they handle caching improperly like this it’s entirely irrelevant. hitting zhipu direct with base pricing would probably end up costing less since the caching actually works there
8
581
Replying to @cheatyyyy
v4.1 flash is literally region restricted for me lol @opencode @thdxr absolutely unacceptable service
2
8
911
works fine direct over api btw
4
94