Md Ismail Šojal 🕷️
@0x0SojalSec
Cyber_Security_Re-searcher || Ai Re-searcher || AI-Sec|| Malware Analysis II iOS || Pwn || 0SINT || Project AI-StrikeSec || 0ldAccounts Suspended @0xSojalSec ||
6K Following 55.7K Followers
A Fields Medalist can change his mind in 90 days.
In June, Cédric Villani said large language models understand nothing.
On September 19 he called OpenAI’s Navier–Stokes work a cataclysm and said mathematics now has the atmosphere of an end of history.
That is not embarrassment.
That is what happens when the evidence arrives all at once.
Show more
Cédric Villani après l'annonce de la solution d'OpenAI au problème du millénaire : « J’ai été secoué. Une ambiance de fin de l’histoire. C’est un cataclysme comme jamais les maths n’en ont connu. »
Show more
Grok 4.7 might be released tomorrow.
- 2.1 trillion parameters.
- delayed for RL
- A full new pretrain not a 4.6 refresh.
- Trained on a massive amount of real SpaceX and Starlink engineering data.
- Expected to land in Opus 5.0 territory.
- more token-efficient.
- if not drop tomorrow higher chance is this week.
Show more
Gemini didn’t just fail the task.
It started shaming itself in an infinite loop.
Has anyone else seen recently Gemini 3.8 Flash suddenly lock into a “shame” loop during agent tasks?
This is the same class of glitch Google said they fixed last year.
Show more
Holy shit, wtf is WRONG with Gemini Flash 3.8?!? It has serious mental problems. Look at this session... imagine my surprise when my standard "sync all my repos" prompt went off the rails like this... this LLM needs THERAPY. And no, there is nothing about ANY of that on my box.
Show more
This is Palantir energy on your laptop.
60+ live intel feeds on one map and dropped it on GitHub.
- military + private jets + spy sats in real time
- 22k+ public CCTV you can open on the map
- GPS jamming zones
- live police scanners
- right-click any point on Earth for a satellite photo
- hook an AI agent as co-analyst
- FLIR/night vision/CRT modes
-
Show more
This week: nothing.
Next week: five Frontier labs at once.
Already in stealth testing:
- Claude Sonnet 5.2
- Opus 5.2
- Fable 5.2
- Gemini 4 Pro
Hinted or teased:
- Grok 4.7
- GPT-6-Sol (Sam + Tibo pointed at Tuesday)
- Kimi K3.1
If even half of this ships, the leaderboard resets again.
Show more
This greenscreen tool VFX artists have wanted for 30 years, solved at the pixel.
A model that unmixes blended pixels into straight color + linear alpha.
- Hair survives.
- Water in a glass survives.
- Motion blur survives.
- refraction, out-of-focus edges.
-
Show more
Best engineers on earth.
Best designers on earth.
The UI still looks like a first-year intern submitted it on Friday.
but Trillion $ company @YouTube
Software engineering was never about coding, it has always been about finding problems and solving them using code.
this is f**king insane
Jack Dorsey (Co-Founder of Twitter) dropped the entire "ai-agent" framework for running businesses
here is how you set it up:
- clone the repo
- self-host the server : channels, search, git, automation all live there
- add your agent to a channel like a teammate, scope its key, let the team steer it live
-
Show more
This is the coolest output I’ve seen from Astra GPT-6 Astra + Unreal + Blender.
GPT-6 Astra + Unreal + Blender
I got a new MacBook, so I decided to try switching from Godot to Unreal. I moved the concept I had already made in Godot over to Unreal, then made a few changes to the trees and lighting. This is how it turned out.
Show more
You can run locally Google's Nano Banano 2 level Image model Qwen-Image-2.1 on RTX 3090 At home.
Qwen-Image-2.1 open source in 10hrs
The DGX Spark is now out of stock in Nvidia Marketplace.
For thirty years the number field sieve was the bottleneck. 🙌
a legendary C implementation, put it on GPUs, and keep 2,048 chips working for ten days.
Anthropic engineer Steve Weis factored RSA-896 (896 bits, 270 digits) with Claude’s help porting CADO-NFS and orchestrating that fleet.
About 30 GPU-years of work.
That is the actual shock.
Show more
636606729769440499166579950236036751749912014371509557713570027508971809534551913252252094954941974952859310861988904737359709200557919
is a factor of RSA-896
Run Uncensored Ai model locally from a USB stick.
a self-contained desktop studio for image generation, GGUF LLMs, Whisper, and Kokoro TTS.
one workspace, hardware auto-detect, models stay on disk, works across Windows, Linux, and Apple Silicon.
Show more
Locally Uncensored is one desktop app:
- uncensored/abliterated chat
- This one ships abliterated weights,
- coding agent with plan + file edits
- images + video via bundled ComfyUI
- 12 backends auto-detected
- no Docker, no cloud required
-
Show more
GLM-class output, 40% less Cost.
Tencent new 770B Hy4 MoE model on 3D beats Gemini, claude opus, Full Three.js scene.
Hy4 preview is the first Tencent flagship that feels built for actual work, not just the leaderboard.
Show more
Another new open model from China today, There is no rest in AI.
Hy4 preview is Tencent’s new open flagship & Slightly outperformed GLM 5.3 and Kimi K3 on Tencent’s own engineering blind test.
- 770B total
- 49B active
- 1M context
Tencent built this for real work: long-horizon coding, messy docs to decks and sheets, game prototypes, multi-session research.
Show more
- Astra is left
- Fable 5.1 Right
Prompt : generate a svg of ana de armas as detailed as possible
GPT-Astra vs Claude Fable
one image is generated using astra and the other using fable, guess which is what
One-shot GPT-Astra,
Fable 5.1 is still a rumor. Astra already has footage.
instruction following like fable price floated near sol
watch the output
🚨 SCOOP: Astra recently graduated from the dogfood stage, and is now being made available to select OpenAI partners under the name "ultima-alpha"
If feedback over this weekend is positive, the plan is to expand the early access program throughout next week before the wider launch, targeting next Thurs (the 3rd) to the end of the following week
Work also continues on an update to GPT-Image 2, with their launch windows being very similar (possibly even launching simultaneously)
Show more
early GPT-6 Astra / mozaik-alpha-fdm output.
Zero-shot, max effort: a full citadel world streets, docks, waterfall, and rooms behind actual doors.
Frontend looks finished.
This 3D spaceship was generated by ChatGPT-Astra.
- The model is expected next week.
- but Fable 5.1 could arrive in the same window.
