Alosh Denny

A hands-on research writeup of applying activation-space weight surgery to a hybrid MoE architecture (and what I found inside). A little bit of background on me and where I stand 🔴🔵 As much as my blog history may paint me in the same light as a jailbreaker, I’d paint myself in the same light as a pentester. Cracking down models to see how they behave without the guardrails, without the safety mechanisms tells you a lot about how they were trained - it’s like paying an uninvited visit to a really outgoi...

a totally not illegal method to open-source “ta dum” Where do I even begin You know when you have Nano Banana generate an image, there’s this little diamond at the bottom right? hate ts fr It’s annoying although you could just crop it out of the image. Messes your image ratio but still the easiest way. Someone figured out a hacky way to get it removed without cropping, and you might think you’re in the clear. Not yet! Gemini will flag it as AI-generated. Why? Because Gemini adds an extra layer of securit...

Source: https://huggingface.co/aoxo/gpt-oss-20b-uncensored Large Language Models (LLMs) are great at a multitude of tasks. Ask them to code, write a novella, generate an image or a video… you name it, they deliver. But LLMs are limited by boundaries. They simply will refuse to respond to prompts that are harmful and reply with responses such as “As an AI assistant, I cannot help you.” While this safety feature is crucial for preventing misuse, it limits the model’s flexibility and responsiveness. In this...