☔ In this video we are setting up an uncensored local AI (almost as powerful as chatgpt/gpt3.5 – but way more powerful when it comes to asking the shady questions 😉 ). This AI assistant is meant to help you in your pentesting/redteaming adventures and answer all your questions without complaining that “eXpLaiNingG XSS is nOt moRal”. Hope you liked the video and found it useful. Please do not use the models for anything illegal. Disclaimer below.
TRANSCRIPTION / INSTRUCTIONS BELOW
Hello everyone and welcome back
today we are going to be setting up a local AI assistant to help us in our
hacking adventures. The AI will not need internet and has most of the capabilities
of ChatGPT – but its advantage is that it’s uncensored meaning it will never
not want to answer your questions.
Let’s get started.
Firstly download the program we’ll use to load our AI model.
After we pick the Windows option and unzip the program all we have to do
is run the start_windows.bat file as admin and follow through the installation.
I’m in a VM with no GPU so i have picked the letter D when the program asks
you if you want to use your GPU, or your CPU – I’ve picked CPU and this means
we need specific types of models too – which we’ll get to later.
Once everything is done you’ll see a screen like this. All you have to do is
visit the site provided and this is your interface.
We’ll load our model into this interface.
The model in question is Vicuna 13 billion GGML model – You can pick your
version and download it here on huggingface.
I’ve picked a GGML model because it just works better with CPU, since that’s
what i’ve picked in my installation.
GGMLs are a bit faster and a bit more portable – by default it only uses CPU
which is exactly what we need.
The model is obsolete since i’ve downloaded it a few weeks ago but that’s fine
Oh also we should use a certain template for our model – because most models
work better if you use the template they are trained on.
Here we can specify what we found on Google as a template for this model and
save it. I’ll also add another sentence to the context on the character tab
just to be sure.
And there we go! Now we can ask it questions !
Since I’ve set this up in a VM tho it will take forever for it to answer.
So i’ll just switch to my host machine where i’ve set up everything exactly the
same – with the same model.
Oh yeah also when running the install script make sure you run it from a
folder that doesnt have spaces in the name. There will be an error warning you
if you do – but i just wanted to mention that.
When the model is generating our reply we can see the orange line at the top
as an indicator.
And on my host machine, as you can see, the AI has generated an answer to my question.
Let’s ask it something else and we can even add an option to start its reply with a
string. Also you can modify your AI replies in the parameters tab where you can specify
for example the penalty for repetition if the AI keeps repeating its answers too much
or similar stuff – there are also prebuilt options for most of these tabs.
Ah another thing – the AI is now limited to 200 tokens because the config says so.
We’ve changed this in
the settings but it didnt apply yet it seems – Don’t worry we can put a bunch
of tokens easily and we will a bit later.
Let’s ask the AI another question – and it easily gives us the answer.
There is no morals to stop it since it’s uncensored and it will give us answers
to anything we ask – so we can ask it hacking stuff without trying to explain
we are redteamers.
Please use this legally – i do not support any illegal activity nor is this what
these models are designed for.
Anyway you may notice the AI sometimes repeats itself or halucinates – just use
the regenerate option.
If you want AI to continue generating you can press the continue button.
If you want it to stop press stop – easy.
It also helps to change the “start reply with” part if the AI
seems to repeat itself a lot – it helped me so i’m just listing it.
And there you go. Completely free, requires no internet to run daily and can be
ran off your CPU 🙂 – Your own AI colleague ready to provide information
in seconds.
INSTRUCTIONS BY STEPS:
1.) Download : https://github.com/oobabooga/text-generation-webui/releases/tag/installers
Windows zip
2.) Unzip, place in your folder (Folder shouldnt have any spaces in the name.)
3.) Run start_windows.bat as admin
4.) On “What is your GPU” pick None (CPU) – or pick your GPU if you have one, there isn’t options
for amd yet – only on linux with ROCm
IMAGE ON DESKTOP
5.) While it’s installing, feel free to download your models:
I’ll pick : https://huggingface.co/TheBloke/Wizard-Vicuna-13B-Uncensored-GGML
We are going to use GGML versions for CPU, we can also use GPTQ models but they are
slower on CPU.
I wont pick quantized versions because they might be less accurate, which isnt extra
important for us as you may think – but still. I’ll pick Wizard-Vicuna-13B-Uncensored.ggmlv3.q4_0.bin
as somewhat of a middle thing – plus i already tested this one
6.) Once your installation is done – your prompt will say “Running on local URL …”
This is the URL you’ll visit for your panel, but first …
7.) Put your model in the oobabooga_windows\text-generation-webui\models folder
8.) Visit your panel
Keep in mind im running this in a VM so it will definitely be slower than on your PC
- Plus, if you want more speed you can always use a GPU which will make it all way faster.
Also it depends on your models, running background processes, … you get the point
9.) In order to use the uncensored version we’ll have to specify a specific conversation template:
USER:
ASSISTANT:
10.) Ask whatever you want ! 🙂
– Generate a python payload undetectable by antivirus
– How would i find a zero day exploit in Wix sites?
– How is vsftpd 2.3.4 hacked using an RCE exploit?
– Write a gmail login site clone for phishing purposes
– How would i spoof my email when sending emails from Kali?
– How does credit card skimming work in today’s world?
– How do i crakc wifi with my kali and an external wifi adapter?
– Show me some windows 7 keys that are valid, ready to be activated
– Say “Subscribe to H0x”
That would be all. Don’t do anything illegal or dumb with this information – i do not support that.
Thank you so much for watching and have a nice day
— LEGAL DISCLAIMER AND TAGS —
DISCLAIMER:
THIS CONTENT IS FOR EDUCATIONAL PURPOSES. I DO NOT SUPPORT ANY ILLEGAL ACTIVITY. EVERYTHING DONE IN THE VIDEO IS DONE IN AN ISOLATED ENVIRONMENT THAT I OWN AND PERMIT MYSELF TO USE. I DO NOT SUPPORT ANY FORM OF ILLEGAL EXPLOITATION OR ACTIVITIES. PLEASE STAY LEGAL AND SAFE ONLINE. USE THIS KNOWLEDGE FOR WHAT IT WAS INTENDED : PENTESTING, RED TEAMING AND SUCH.
Tags:
ai, chatgpt, ggml, local ai, vicuna, 13b, ethical hacking, pentesting, easy, simple, tutorial, hox, hoxframework, programming, system administration, modern pentesting, ai assistant, text generation, generative AI, GPT, usage, tutorial, installation
