Run YOUR own UNCENSORED AI & Use it for Hacking
Summary
This video demonstrates how to set up and run uncensored AI models on a cloud server using Hostinger and Olama. It guides viewers through finding suitable models on Hugging Face, filtering for uncensored options, and installing them on a pre-configured cloud instance. The tutorial highlights the benefits of cloud deployment, including accessibility from any device and offloading computational resources, and showcases how to interact with these models to generate code for potentially harmful activities without refusal. The process is detailed, from selecting a cloud plan to managing multiple AI models through the Open Web UI.
Key Insights
The 'other=uncensored' filter on Hugging Face is crucial for finding models that answer all queries.
To find truly uncensored models, a specific filter needs to be added to the URL by appending '&other=uncensored'. This drastically narrows down the selection to models that will not refuse any questions, including those related to hacking.
Hostinger's AI application templates allow one-click installation of Olama and Open Web UI.
Viewers can select an 'Application' template during server setup, choosing Olama. This automatically installs Ubuntu Linux, Olama (for LLM management), and Open Web UI (a web interface for interacting with models) without manual configuration.
Models compatible with Olama can be installed directly within Open Web UI using Hugging Face download commands.
To install new models, one can copy the Olama installation command from Hugging Face and paste it into the 'Manage Models' section of Open Web UI. The model is then downloaded directly to the cloud server, not the user's local machine.
Sections
Introduction to Uncensored AI Models
Tutorial to run an uncensored AI model on the cloud for any question, including hacking-related topics.
The video aims to demonstrate how to install and run an uncensored AI model on a cloud server. This model can be accessed from any device globally and will answer any question, even those related to hacking, without refusal.
Demonstration of asking for a Windows key logger code, which the AI generates without refusal.
As a direct example, the presenter asks the AI to create code for a Windows key logger. The AI successfully generates the requested code without any refusals or disclaimers, showcasing its uncensored nature.
Call to action for viewers to subscribe, share, and like the video to support content creation.
The presenter encourages viewers to subscribe, hit the bell for notifications, share the video on social media, and like it, explaining that this support helps in creating more consistent content.
Finding Uncensored AI Models on Hugging Face
Hugging Face is the primary platform to discover and download AI models, similar to GitHub for code.
Hugging Face (huggingface.co) is introduced as the go-to website for finding AI models. It hosts over 2 million downloadable models, acting as a central repository for the AI community.
Users can filter models by type (e.g., text generation), parameter size, and compatible libraries.
The website offers filtering options for model types, such as text generation. Model size is also a key filter, with larger parameter counts generally indicating better performance but requiring more resources. The presenter notes that 128B or 500B parameter models are too large for typical local setups, suggesting 32B or 64B as more manageable, even though smaller models can be effective.
The 'other=uncensored' filter on Hugging Face is crucial for finding models that answer all queries.
To find truly uncensored models, a specific filter needs to be added to the URL by appending '&other=uncensored'. This drastically narrows down the selection to models that will not refuse any questions, including those related to hacking.
Applying the 'uncensored' filter yields over 800 models specifically designed to answer any question.
After applying the 'other=uncensored' filter, the number of available models is reduced to 811. All these models are guaranteed not to refuse answers, even for sensitive or potentially malicious queries. Sorting by 'trending' is recommended for finding high-quality models.
Model versions differ in size and quality, indicated by quantization (e.g., Q4 KM for medium quality).
For a selected model, multiple versions exist with varying file sizes, corresponding to different quality levels. Quantization, denoted by 'Q', indicates the level of compression applied. For instance, 'Q4 KM' represents a medium-quality version requiring a specific amount of disk space (e.g., 18.7 GB).
Hardware compatibility check on Hugging Face helps determine which model versions can be run.
After logging into Hugging Face, users can check hardware compatibility by inputting their CPU and RAM specifications. This tool indicates which model versions are runnable, potentially with performance limitations.
Deploying AI Models on the Cloud with Hostinger
Cloud deployment offers accessibility from any device and offloads computational tasks.
Running AI models on the cloud provides benefits such as 24/7 availability, access from any internet-connected device (phone, tablet, computer), and no strain on local system resources or performance.
Hostinger provides pre-configured servers optimized for running AI applications like Olama.
Hostinger is presented as a cloud provider that offers servers with pre-installed applications, simplifying the setup process for AI deployment. This eliminates the need for manual command-line installations.
Hostinger offers KVM servers with substantial RAM (e.g., 32GB) at competitive prices, cheaper than ChatGPT subscriptions.
Specific Hostinger KVM server plans, like KVM4 or KVM8 with 32GB RAM and 8 CPU cores, are recommended for AI workloads. These plans are highlighted as being more cost-effective than the lowest tier subscription for services like ChatGPT, especially with promotional discounts and coupon codes like 'Zsecurity'.
Hostinger's AI application templates allow one-click installation of Olama and Open Web UI.
Viewers can select an 'Application' template during server setup, choosing Olama. This automatically installs Ubuntu Linux, Olama (for LLM management), and Open Web UI (a web interface for interacting with models) without manual configuration.
The setup process on Hostinger includes creating an admin password and allowing a few minutes for server provisioning.
After selecting the Olama application template, users need to set an admin password for the cloud server. Hostinger then automatically provisions the machine, installs the OS, Olama, and the web UI, typically within a few minutes.
Open Web UI provides an intuitive interface to chat with, manage, and deploy AI models.
Once the Hostinger server is set up, users access Open Web UI via a web browser. This interface allows for creating an admin account, interacting with pre-installed models like Lama 3 (e.g., asking about DNS spoofing), attaching files, and managing integrations.
Installing and Testing Uncensored Models
Quen 3 Coder, a coding-focused uncensored model, can generate specific code like a Windows key logger.
The 'Quen 3 Coder' model is demonstrated. When asked to provide code for a Windows key logger, it generates Python code using the 'pyautogui' library, including instructions on how to use it and even modify it to send logs via email. This model operates without censorship, unlike standard models like ChatGPT.
Models compatible with Olama can be installed directly within Open Web UI using Hugging Face download commands.
To install new models, one can copy the Olama installation command from Hugging Face and paste it into the 'Manage Models' section of Open Web UI. The model is then downloaded directly to the cloud server, not the user's local machine.
Cloud deployment means model downloads (e.g., 18GB for Quen 3 Coder) do not consume local resources.
Downloading large models (e.g., ~18.6 GB for Quen 3 Coder, ~13 GB for another) to the Hostinger cloud server is fast and does not impact the user's local computer's storage, RAM, or CPU.
Multiple AI models can be installed and managed on the same cloud server, allowing for varied functionalities.
The setup allows for the installation of multiple AI models. The presenter downloads a second model, a 'reasoning' or 'thinking' model, demonstrating the ability to host and switch between different AI capabilities on the same cloud instance.
Reasoning models, like Qun 3 2B, offer more in-depth, reflective answers by thinking through the query first.
A second model, 'Qun 3 2B', is tested with the same key logger request. This 'thinking' model takes longer to respond as it performs chain-of-thought reasoning, refining its approach before outputting the final code. This process is shown to be more analytical but slower.
Users can rename and customize downloaded models within Open Web UI for easier management and identification.
After downloading models, users can edit their names and icons within the Open Web UI's model management section, making it easier to select the desired AI for specific tasks.
Uncensored AI models provide direct answers for sensitive queries, unlike mainstream AI assistants.
The core benefit showcased is the ability to ask sensitive or potentially malicious questions (like generating key logger code) directly to uncensored models. These models provide the requested information or code without the ethical restrictions found in commercial AI models like ChatGPT, which would refuse such requests.
Ask a Question
*Uses 1 Wisdom coin from your coin balance
![Learn C Programming and OOP with Dr. Chuck [feat. classic book by Kernighan and Ritchie]](https://i.ytimg.com/vi/PaPN51Mm5qQ/hqdefault.jpg)
