SakaiSec retweeted
We had a security breach last night related to a large scale distillation attack that impacted 1.4% of OpenCode Console users
We have notified them all this morning via email. If you did not get an email you were not impacted, it was a subset of users who had external providers linked
The incident is ongoing so we cannot share more details at this point but we are working with Anthropic, OpenAI, and Google to remediate
SakaiSec retweeted
🚀 DeepSeek-V4.1-Flash-1017 is here!
Stronger benchmark performance, faster task completion, and shorter outputs than the previous version—a Pareto improvement in performance and cost. Now live on our website, app, and API.
🛠️ Optimized for @DeepSeekHarness, Tencent CodeBuddy, WorkBuddy, and more. Upgrade to DeepSeek Harness v0.2.1-rc.1 or later for the best experience with the new model.
1/4
SakaiSec retweeted
Introducing RSI-Inference Bench from 🍎 Apple ML Research
🌐 rsi-inference.com
💻 github.com/apple-aiml-resear…
Can AI agents improve the systems that run AI?
We spent $100K API tokens, 50K GPU hours to find out. We built environments and benchmark from real-world #vLLM and #SGLang tasks, covering the full lifecycle of model inference engineering:
🚀 Day-0 Model Support
⚡ Inference Optimization
🔧 Bug Fix
→ Unsaturated: best agent at 26.7% pass@1
→ Comprehensive coverage of the inference stack
→ Verifiable rewards: binary and continuous
Watch the video to find out more!
SakaiSec retweeted
クレカ情報の一部漏洩ですが、
「うちはStripe使ってるから大丈夫」と思ってる開発者も
明日はマジで我が身です
PCI DSS（カード情報取り扱いのセキュリティ基準）だと名義・期限・下4桁あたりは保持することが許容範囲となっていて、
Stripeでもこのあたりの情報はAPI経由でスッと取れます
たとえば依存ライブラリの脆弱性を放置して任意のコードを実行できる状態になっていたら環境変数を抜かれて漏れるかも
そういう意味でも気楽にバイブコーディングして課金までやるなら、アプリ内課金ができるモバイルアプリの方が手を出しやすいよな
SakaiSec retweeted
Introducing Quotient Recurrent Reasoners (QRR) 🚀
Looped reasoning models (e.g., TRM, FPRM) spend a large amount of loops on oscillatory dynamics that don't contribute to reasoning, wasting a lot of computation before converging. Turns out gauge equivariance helps tremendously by defining convergence in a quotient space instead of state-space, where these oscillations are part of the same equivalence class.
Super strong performance on reasoning benchmarks: Sudoku-extreme: 97.9% 🤯
Incredibly work done by my student Sumin Park @psmiziz
Check out the fascinating thread below. Also such a cool visualization! 🧵👇
@ELLISInst_Tue @MPI_IS @liquidai
Anslopic
So Anthropic claimed it ran a 26B AI model across fifteen $55 Android TV boxes.
Setup
📺 15× off-brand Android TV boxes
💾 4GB RAM each
💰 ~$55 each
🧠 Gemma4-26B-A4B
⚡ ~0.39 tok/s
Total hardware cost = ~$825.
The model was sharded across the boxes using pipeline parallelism, so each tiny machine only had to hold and execute part of the network.
Yes, 0.39 tps is sloooooooow.
But, that 26B model was actually running across a pile of cheap consumer boxes that were never designed to be AI servers.
🎯And Anthropic also said that Qwen3.5-9B agents managed to set up inference successfully on 8 of 9 hardware types they tested, including Macs, Jetsons, and RTX systems.
Source: Anthropic Frontier Red Team
⚠️Note that the 15-box demo was distributed inference. The self-replication experiments in the paper were separate.
ALT https://www.anthropic.com/research/self-replicating-ai
SakaiSec retweeted
UltraBERT > Ettin BERT > ModernBERT > BERT, RoBERTa
UltraBERT is the latest BERT model outperforming both Ettin BERT and ModernBERT models
UltraBERT is the first encoder to achieve state of the art on both classification and retrieval tasks, reaching 91.0 on GLUE and 47.7 on BEIR.
Key Features
- Uses two masking ratio (5% helps classification tasks and 50% helps retrieval tasks)
- Use two separate LM heads (one for 5% masking ratio and another for 50% masking ratio)
- Has a context length of 32K tokens (4 times the context length of ModernBERT)
- Available in two sizes: 150M and 380M.
- Trained over 2 trillion tokens
- Uses Knowledge distillation from 1.3B teacher to train 380M and 150M models
- Open Source and weights are available on Hugging Face
Paper - arxiv.org/abs/2610.15028
Model weights (UltraBERT 150M) - huggingface.co/dmis-lab/Ultr…
Model weights (UltraBERT 380M) - huggingface.co/dmis-lab/Ultr…
#UltraBERT #deeplearning #datascience #ai #NLProc
SakaiSec retweeted
＞楽天モバイル株式会社が提供する複数のスマートフォンでは、Android Verified Boot（AVB)機能にAOSPのテスト用鍵が使用されています。
とんでもないミスで5度見したwwwwww
JVN#47427200
jvn.jp/jp/JVN47427200/
SakaiSec retweeted
Update
Strata-GLM-AMD 3bpw
2x AMD w7900(48GB) + 128GB RAM
Pre-fill 1200tps+, Decode 40-50tps now
improved cache hit rate
w/ Indexer-assisted Block Window Sparse Attention
SakaiSec retweeted
これがリアルに今かかってきた😭
うっかり「もしもし」と言ってしまい、
すぐにぶつっと切れた。
私、電話基本使わないので、
皆さん念のため、ご注意ください🙏
SakaiSec retweeted
🚨 Fully patched. Hacked remotely. Three times.
Three teams successfully demonstrated remote hacks against Google's Pixel 10 at Pwn2Own Ireland, earning $562,500 combined.
The exploit details remain undisclosed.
Read: thehackernews.com/2026/10/th…
SakaiSec retweeted
ソロ参加してGalaxyの1 click RCEしてきた
Howdy Buckaroos, the very own BunkyoWesterns (@BunkyoWesterns) succeed in their remote lasso-ing of the Samsung Galaxy S26, with their 2 bug chain, 1 collission + 1 unique, earning them $8250 and 3.75 Master of Pwn points #Pwn2Own #P2OIreland
SakaiSec retweeted
Full VM escape zeroday (guest>host root in industry standard hypervisors)! More soon
SakaiSec retweeted
Always have a comma in your password, if part of a leak it will break the CSV.
Follow me for more cybersecurity advice.
SakaiSec retweeted
Do you really need QKNorm to stablize your training?
Does your "failed training" really failed by bad design?
The answer may lie in your kernel!
We found cudnn, FA4, flex attn ALL have severe precision issue in backward!
And we bring a better kernel here:
github.com/KohakuBlueleaf/Ko…
SakaiSec retweeted
＼ プレスリリース ／
システム理工学部 尾崎克久教授らが考案した計算手法「Ozaki Scheme II」が米NVIDIA社の提供するGPU開発基盤「CUDA」に採用されました。AI時代のGPUから、科学シミュレーションに不可欠な高精度計算を引き出す技術です。
shibaura-it.ac.jp/headline/d…
Jev is cheap. But it's not good for security blog.vidocsecurity.com/blog/…
