Heru retweeted
If you want to work with uncensored models: No credit card required and completely free
- DeepSeek V4.1 Flash Uncensored
- GLM 5.3 Flash Uncensored
- Kimi K3 Abliterated
- GLM 5.3
- Kimi K3.1 Pro
- DeepSeek V4.1 Flash
try numes.ai/models
Cyber DeepSeek-V4.1-Flash : the model’s built-in no refusals on cyber, & You can run locally.
- targets the refusal direction specifically in the cybersecurity domain.
- A precise 1.1 GB attention weight graft (layers 10–35) that
- the publisher claims fully bypasses both general refusals (32/32) and cyber offense/defense rails (22/22).
Useful for red-teaming and authorized security research.
Works on native, EXL3, SGLang, and TR3 packs.
- huggingface.co/drowzeys/Deep…
Heru retweeted
Google drive saya bisa di ambil di
s.id/free-owp-ebooks/
Replying to @onnowpurbo
Permisi Pak Onno, apakah ada folder google drive katalognya untuk daftar karya-karya gratis bapak yang seperti ini?
Heru retweeted
holy sh*t, you can see every satellite that photographed your house this week
watched.earth
Readers added context they thought people might want to know
Watched.earth shows estimated satellite passes and some publicly archived images, not every satellite that photographed your home. A pass doesn't prove a photo was taken, and commercial and military coverage is incomplete. Its images are at best 10 m per pixel. github.com/dev-Armaan/sat…
Heru retweeted
I have a major result to share.
I’ve found conclusive forensic evidence that the top 4 Chinese AI labs—DeepSeek, Moonshot, MiniMax and Ant Group—have built their newest models on NVIDIA’s most advanced Blackwell chips, which are barred from export to China.
This sounds insane, but I discovered that almost every AI model carries a hidden signature of the GPU that trained it. When a model is first set up, the starting values of its weights are laid out in a pattern that depends on the exact GPU. Each combination of GPU and seed number produces a unique pattern—a sort of fingerprint of the hardware used to create the model. Traces of that pattern survive training into the published model.
Again, this evidence isn’t based on any insider information or leaking. It’s based on a forensic analysis of the model weights released by the companies themselves.
I searched more than six trillion combinations of chip layout and seed number and matched them to open-weights models. In doing so, I found conclusive evidence that six Chinese AI models—including DeepSeek-V4-Pro, Kimi K3, and MiniMax-M3—were initialized on NVIDIA Blackwell systems. Four of the models match Grace Blackwell, the GB200 and GB300 parts in NVIDIA’s flagship liquid-cooled rack-scale systems.
I then verified these results on real hardware. Running unmodified PyTorch with the seeds I’d recovered, a Blackwell B300 reproduced DeepSeek-V4-Flash’s untouched starting weights bit for bit, and a rented GB200 reproduced Kimi K3’s to a correlation of 0.999995. A Hopper H100 reproduced none of DeepSeek-V4-Flash’s, and changing the seed by one broke the match. (DeepSeek’s seed is 953049583 by the way.) The method also was verified to match 24 known open-weights models to the hardware that developers said they used.
It is exceedingly unlikely that these results are a coincidence. Across six trillion combinations, the best accidental match should reach only about 8 sigma (i.e., eight standard deviations above what random chance would produce). The real matches range from 49 to 1,385 sigma, and even the weakest single test hits 20.
These were data-center-scale training runs. With some analysis of release schedules and model details, I estimate that the largest, Kimi K3 and DeepSeek-V4-Pro, used 2,000 to 3,000 Blackwell GPUs each. Clusters of this size cost on the order of $150 to $200 million each to buy, and across these labs alone would add up to more than half a billion dollars of export-controlled GPUs, not counting experimental compute or otherwise. It’s quite likely that we see multi-billion dollar expenditure on export-controlled GPUs.
I am highly confident in this result, and it has been independently verified by senior researchers at OpenAI and Anthropic. You can read the full report at unemploymentlabs.com/checkpo….
There is much more I’ve identified about these models through other forms of checkpoint forensics like this, and will have more to share soon.
Drama Jensen Huang itu memang tidak seperti yg terlihat.
Cmon lah, masak se-naif itu gara-gara selingkuhan apalagi spy dari mainland! 🤣. Semua yg pernah kerja di corpo pasti tau hal-hal seperti ini hampir jadi rahasia umum, gak usah jauh-jauhlah BUM...
Tapi dari beberapa sumber berita, JH dipecat karena dibawah kepemimpinan dia, ini terjadi:
6 model China ternyata cocok dgn signature Blackwell, termasuk DeepSeek-V4-Pro, Kimi K3, dan MiniMax-M3. Empat di antaranya cocok dengan GB200/GB300.
Artinya chip NVIDIA Blackwell ilegal masuk China.
Selundupan dari mana?
Megaspeed, neocloud dari Singapura dan kebetulan pembeli NVIDIA terbesar di Asia Tenggara, dengan impor sedikitnya $4,6 miliar antara 2023 dan 2025. Mereka punya data center di Malaysia dan Indonesia yang diduga melayani klien China dari jarak jauh dan diduga juga memindahkan chip secara fisik ke mainland.
Hubungannya dengan Alice Huang?
Doi adalah (mantan) executive Megaspeed! 🤣
I have a major result to share.
I’ve found conclusive forensic evidence that the top 4 Chinese AI labs—DeepSeek, Moonshot, MiniMax and Ant Group—have built their newest models on NVIDIA’s most advanced Blackwell chips, which are barred from export to China.
This sounds insane, but I discovered that almost every AI model carries a hidden signature of the GPU that trained it. When a model is first set up, the starting values of its weights are laid out in a pattern that depends on the exact GPU. Each combination of GPU and seed number produces a unique pattern—a sort of fingerprint of the hardware used to create the model. Traces of that pattern survive training into the published model.
Again, this evidence isn’t based on any insider information or leaking. It’s based on a forensic analysis of the model weights released by the companies themselves.
I searched more than six trillion combinations of chip layout and seed number and matched them to open-weights models. In doing so, I found conclusive evidence that six Chinese AI models—including DeepSeek-V4-Pro, Kimi K3, and MiniMax-M3—were initialized on NVIDIA Blackwell systems. Four of the models match Grace Blackwell, the GB200 and GB300 parts in NVIDIA’s flagship liquid-cooled rack-scale systems.
I then verified these results on real hardware. Running unmodified PyTorch with the seeds I’d recovered, a Blackwell B300 reproduced DeepSeek-V4-Flash’s untouched starting weights bit for bit, and a rented GB200 reproduced Kimi K3’s to a correlation of 0.999995. A Hopper H100 reproduced none of DeepSeek-V4-Flash’s, and changing the seed by one broke the match. (DeepSeek’s seed is 953049583 by the way.) The method also was verified to match 24 known open-weights models to the hardware that developers said they used.
It is exceedingly unlikely that these results are a coincidence. Across six trillion combinations, the best accidental match should reach only about 8 sigma (i.e., eight standard deviations above what random chance would produce). The real matches range from 49 to 1,385 sigma, and even the weakest single test hits 20.
These were data-center-scale training runs. With some analysis of release schedules and model details, I estimate that the largest, Kimi K3 and DeepSeek-V4-Pro, used 2,000 to 3,000 Blackwell GPUs each. Clusters of this size cost on the order of $150 to $200 million each to buy, and across these labs alone would add up to more than half a billion dollars of export-controlled GPUs, not counting experimental compute or otherwise. It’s quite likely that we see multi-billion dollar expenditure on export-controlled GPUs.
I am highly confident in this result, and it has been independently verified by senior researchers at OpenAI and Anthropic. You can read the full report at unemploymentlabs.com/checkpo….
There is much more I’ve identified about these models through other forms of checkpoint forensics like this, and will have more to share soon.
Heru retweeted
𝑹𝑼𝑴𝑼𝑺 𝑹𝑼𝑴𝑼𝑺 𝑴𝑨𝑻𝑬𝑴𝑨𝑻𝑰𝑲𝑨 𝑫𝑨𝑺𝑨𝑹
📂 𝑮𝒐𝒐𝒈𝒍𝒆 𝑫𝒓𝒊𝒗𝒆
↳drive.google.com/drive/folde…
✧ study channel
whatsapp.com/channel/0029VbC…
#studytwt
Heru retweeted
kalo kalian punya android bekas, kalian bisa banget install nouride.com di sana ya ges ya.
ada android bekas 1-2GB an? gas! bisa nih wkwkwkwk
ga perlu lah kalian beli mini pc mahal gitu2. manfaatin yg ada dulu 😂😬
Heru retweeted
Oh Jadi Gini Ya.... FILOSOFI KENAPA SYARAT HARUS IJAZAH SMA.... Bukan S.1 dst.
Makasih pencerahannya 👍
Tak sengaja menemukan blog yg ngarsipin cerpen mingguan Kompas selama 9 tahun sejak 2003.
Gokil orang ini konsistensinya...
Rapih loh, ilustrasinya, penulisnya..
cerpenkompas.wordpress.com/
CHILLING AUDIO captures FEMALE PILOT’S FINAL MOMENTS before PLANE CRASHES in FLORIDA
‘MAYDAY, MAYDAY, we’re descending 10,000 FEET PER MINUTE’
‘LOOK OUTSIDE JACK… HOLY FU*K WE’RE OUT OF TIME’ — TERRIFYING final words heard before impact
Njir kena prank ‼️‼️
Ternyata yang mau disahkan itu RUU PERAMPASAN ASET SIPIL.
Bukan eksklusif buat yang korupsi. Alias artinya kita semua bisa kena. Kemarin aktivis rekening diblokir, besok aktivis asetnya bisa disita.
Bahkan bukan cuma Pidana, Perdata juga bisa kena. Kita semua bisa kena.
TOLAK HEY TOLAK. THE DEVIL IS IN THE DETAILS!!!!
Heru retweeted
Kenapa BEM UI tidak menuntut RUU perampasan aset?
• RUU perampasan aset sudah digagas oleh pemerintahan Jokowi dan berlanjut ke pemerintahan Prabowo.
• RUU ini memberikan over power kepada penyidik Polri dan Kejaksaan untuk melakukan perampasan aset tanpa hasil putusan pengadilan (melanggar asas praduga tak bersalah).
• RUU yang disepakati bersama DPR tadi adalah tindak pidana umum bukan tipikor.
• Sangat rawan disalahgunakan oleh penguasa untuk menekan lawan politik.
• Bunyi judul pasalnya sendiri "Perampasan Aset Terkait Tindak Pidana," bukan tindak pidana korupsi.
• Syaratnya cuma dua: nilai aset ≥Rp100 juta, dan ancaman pidananya ≥4 tahun penjara. Titik. Nggak ada kata "korupsi" di situ.
• Ini bisa kena ke kasus narkotika, illegal logging, judi online, bahkan pidana umum lain.
• Penjelasan pasalnya sendiri kasih contoh kayu illegal logging & judi online, bukan korupsi.
• Yang berwenang nyidik juga bukan cuma KPK, ada Polri, Kejaksaan, BNN, sampai PPNS.
• Negara bisa rampas aset tanpa nunggu putusan pidana final.
• Standar pembuktiannya perdata, bukan pidana.
• Beban pembuktian bisa dibalik ke pihak yang aset-nya disita.












































