Pinned Tweet
solehudin yang pura pura pandai dan horny dah mati.
tinggal versi yang memang bodo piang dan sangat sangat horny saja sekarang.
feel free to unfollow or mute/block me, all you nice people.
i am about to unleash my most destructive horniest behavior no one ever seen before
TQ
Sölehudin need a giant dingdong! retweeted
what happens if p=np?
Sölehudin need a giant dingdong! retweeted
Techtamu is about building a collaborative tech ecosystem. Daripada reinvent the wheel, kami percaya pada kolaborasi bersama rakan yang mempunyai expertise & digital solutions sendiri.
Sebab itu pada hari ini, we're excited to welcome BoxCompute sebagai rakan strategik baharu.
Sölehudin need a giant dingdong! retweeted
This is a Saudi Prince's Local AI setup
4 x 512GB M3 Ultra Mac Studios clustered with RDMA over Thunderbolt 5 running @exolabs
Sölehudin need a giant dingdong! retweeted
I ran some unscientific evals on Gemini 4 Neon:
1. Diagnostics / planning: pretty good (Sol 6.2 level)
2. Non-trivial implementation: flawless (above Opus 5.5)
This Video For Non Bumi & Bumiputera
#KnowYourHistory
For over 440 years, Malay Muslim sultanates and local fighters continuously resisted the Portuguese, Dutch, and British empires. Although European powers possessed superior naval artillery, fortified stone bases, and massive
Sölehudin need a giant dingdong! retweeted
Replying to @sthehu
Boleh je guna T3 code, link guna Antigravity CLI. Tapi OC ngan pi tak boleh sebab diorang bypass official google harness.
nitter.gq/hammaadjam/status/2124…
Replying to @theo
This does not say anything about using Gemini in T3 legality. Just a person using T3 code who works at Google.
A Chinese GPU with 64GB of VRAM, 2x an RTX 5090, gets revealed on December 29.
Moore Threads calls it Lushan. The company says it will carry the most VRAM ever put on a consumer card. The 64GB figure comes from its own roadmap: 4x the 16GB on its current MTT S80.
64GB fits a 70B model at Q4 (~42.5GB) on one card. Today that takes two 3090s.
How fast it writes depends on memory bandwidth, and Moore Threads hasn't given one.
Decode ceiling for a 70B Q4, bandwidth / model size:
At 448 GB/s (MTT S80 today): ~10 tok/s
At 1,008 GB/s (RTX 4090): ~24 tok/s
At 1,792 GB/s (RTX 5090): ~42 tok/s
Same 64GB, 4x difference in speed.
llama.cpp already has a MUSA backend for Moore Threads cards, so it should run on day one. Their last consumer card, the MTT S80, lost to an RTX 3060 in games at launch.
Would you buy a 64GB card at 448 GB/s?
Sölehudin need a giant dingdong! retweeted
MIT researchers (Joseph Ravichandran and Mengjia Yan) built Fractal, an operating system kernel designed entirely for microarchitecture reverse engineering. Using it, they discovered previously undocumented behaviors in Apple's M1 processor including instruction cache side channels and the first evidence that Phantom speculation affects Apple Silicon.
In Phantom, ordinary instructions including no-ops can be misinterpreted by the CPU as branches, triggering speculative behavior the program never asked for. The CPU still fetches the target into the instruction cache before the protection kicks in.
csail.mit.edu/news/study-how…
Sölehudin need a giant dingdong! retweeted
yall sexualized the word daddy too much my kids gonna have to call him nigger
Sölehudin need a giant dingdong! retweeted
A mathematician begins reading the 250,000-page proof of the Riemann hypothesis.
“What’s the main proof technique?” a colleague asks.
“Proof by exhaustion.”
“Of all possible cases?”
“No. Of all possible readers.”
Sölehudin need a giant dingdong! retweeted
The Riemann proof could have been 10 pages but they made it 250000 to create jobs for math grads.
Sölehudin need a giant dingdong! retweeted
Statement by the Lean and Mathlib communities
Congratulations @__alpoge__ 🦑🦑🦑🦑🦑🦑🦑


































