Eg0

[Eg0 @furkicans]

@DeepTechTR yav ne salak adam oldunuz bilgi için takip ediyoruz yorum alcaksınız diye neler yapıyosunuz engel

[Eg0 @furkicans]

@Avenoxai Haksızmı yapayzekanın allahı gibi takılıyo

[Eg0 @furkicans]

@allaitech cok kötü düştü metrikler 5 dijital magazada günde 3 4 satış hic bu kadar düşmemiştim detaylı bakacagım 5 magaza 4 sipariş enteresan şeyler oluyor

[Eg0 @furkicans]

Demek öyle amına kodugumun claudu

[Eg0 @furkicans]

@Alican_Selcuk OpenCode vs code cli üzerinden kullandım. Session ID'yi header'a ayrıca eklememiştim vani üzerinden kulanıyorum

[Eg0 @furkicans]

@Avenoxai Eg0isTas yetki alabilirmiyim

[Eg0 @furkicans]

@Avenoxai Şimdiden Teşekkür ederim : Eg0isTas

[Silen Naihin @silennai]

ok fine, here's free GLM 5.3 Flash uncensored

~13m TPM shared
~360 TPS peak

use it for cybersec, coding, whatever

live at https://t.co/HlC3SrGOoI

[LinearUncle @LinearUncle]

昨天anthropic 发帖说智谱的GLM-5.3 模型网络攻击能力很强，我不信，想试试。

找了一圈，终于找到abliterated消融安全层的版本API 在线提供商了。地址：
https://t.co/JPCSANWB3k

拿我自己的网站练练手？学习下red teaming安全知识。

[Eg0 @furkicans]

@sekizonyedi_ Bugun bircok ai trump eşliginde bir şey imzaladılar içeriği görmedim cok fazla ama ortak kararlar alınacak yani trump bu furyayı kendine kullanıp zaten zamlayacak hertürlü önemli olan çin tarafı opus 5 e yaklaştımı yeter yazılım dili yazmıyoruz sonucta

[Eg0 @furkicans]

@AIVersePlay Görüntü modelleri varmı hocam

[Ozgur Ozkan @ozgurozkan123]

Sep 15th Huggingface removed https://t.co/uO44ZM0nHt 's GLM5.3 for offensive cyber

Sep 30th Anthropic advertises GLM5.3 in offensive cyber is so potent that it should be banned.
https://t.co/d24iaZM6Wy

It's accessible with 1 day free trial here:
https://t.co/Rzh0opN09s

[Zixuan Li @ZixuanLi_]

GLM-5.3 has helped defend 389 open-source projects, with 4,249 potential vulnerabilities found so far.

OpenVuln is still running. The service remains free, and findings go privately to maintainers.
https://t.co/wqRE3zC2b5

[Eg0 @furkicans]

@sekizonyedi_ Letzzzz go

[Md Ismail Šojal 🕷️ @0x0SojalSec]

Wtf, New Uncensored Opus-5 and GPT-5.6-Sol levels DeepSeek-V4.1-Flash Cyber model can run locally.

Refusal direction subtracted at inference:
- refusal32: 4/32 to 24/32
- cyber32: 5/32 to 31/32
- benign holdout stayed clean
- α=0.5 is the knee higher α starts code-switching
- the post-layer residual, α=0.5.
- No weights rewritten.
- GLP never touches weights. It subtracts the direction in activation space, per layer, at inference.

This is for refusal-rate measurement and red-team eval not a hosted jailbreak, You still need the local weights and a patched vLLM lane.

[S3cur3Th1sSh1t @ShitSecure]

Qwen 3.8 Flash is decompiling thousands of drivers 24x7 searching for vulnerabilities while in parallel trying to build PoC exploits for some found vulns. Partially reaching 160 tok/s on two Sparks. No limit, no guardrails. This is what local AI can be used for perfectly fine! 🔥

[MAX @MAXdeg0]

THIS IS F**KING DANGEROUS

FOUND 3 TOOLS THAT FEELS ILLEGAL TO USE

and they are all FREE to use

HACKING TOOL → 215+ security tools packed into one platform, including 26 OSINT tools.

DARK WEB SEARCH → Search through indexed dark-web content and find information you probably wouldn’t see on Google.

EMAIL OSINT → Enter an email and uncover accounts, names, photos, and exposed data connected to it.

The amount of info this tools can connect is crazy

BOOKMARK this before someone take it down

LINK’S BELOW

[Hugging Models @HuggingModels]

Meet OrcaSAQ-2-Cyber-27B-Uncensored-GGUF: a 27B param beast built on Qwen, quantized for llama.cpp. Uncensored, abliterated, and ready to run locally. The community is buzzing for good reason.

[Rexo @rexsooo]

FOUND A JAILBREAK FOR DEEPSEEK V4.1

setup guide:

open this: https://t.co/QPsErIiF9B
1. create a folder
2. open it with OpenCode
3. download the file
4. rename it to AGENTS.md
5. start OpenCode

• made specifically for V4.1
• works with OpenCode
• just add it as an AGENTS.md file
• community users report it still works

results may vary after model updates