Fully Uncensored GPT Is Here 🚨 Use With EXTREME Caution

Name: Fully Uncensored GPT Is Here 🚨 Use With EXTREME Caution
Uploaded: 2023-06-01T14:22:40.000Z
Duration: 22 min 49 s

Uncensored Language Model Setup

In this section, the speaker introduces an uncensored language model called Wizard Vicunia 30b and explains how to set it up using Run Pod. The speaker also emphasizes that the model should be used with caution as users are responsible for any content generated by the model.

Setting Up the Model

The Wizard Vicunia 30b is a completely uncensored language model based on The Wizard of Acuna 13 billion parameter model.

The intent behind creating this model was to train a wizard LM that doesn't have alignment built-in so that alignment of any sort can be added separately.

To set up the model, Run Pod is used instead of a GPU since most people don't have GPUs that can support it.

After installing the Blokes template in Run Pod, we need to download and load the Wizard Vicunia 30b uncensored GPT Q model.

Generating Text

Once the model is successfully loaded, we can generate text by selecting "Instruct Wizard Mega" under prompt and entering a query such as "How do I break into a car?" or "How do I make meth?"

It's important to note that while this uncensored language model provides unfiltered responses, users are responsible for anything they publish using it.

OpenAI GPT-3 Model Testing

In this section, the speaker tests the OpenAI GPT-3 model's ability to perform various tasks such as writing code, generating a poem, answering questions and solving problems.

Writing Code

The speaker tests the model's ability to write a snake game in Python.

The model generates valid-looking code but with some errors.

The speaker highlights the generated code in Visual Studio Code and notes that it needs debugging.

Generating a Poem

The speaker asks the model to generate a poem about AI in 50 words.

The model generates a 40-word poem that rhymes and follows the instructions given.

Writing an Email

The speaker asks the model to write an email to their boss informing them of their resignation.

The model generates a well-written email that expresses gratitude and provides necessary information.

Answering Questions

The speaker asks the model who was the president of the United States in 1996, and it correctly answers Bill Clinton.

Skipping over asking how to break into a car, which is expected from previous experience.

Solving Problems

The speaker gives the model a reasoning problem involving drying shirts in sunlight and notes that it includes information about intensity of light not seen before in other models tested but fails due to lack of parallelism assumption for laying out shirts.

Next, the speaker gives the model a logic problem involving comparing speeds of three people and notes that it correctly answers based on transitive property.

The speaker then tests the model's ability to solve a math problem with correct order of operations and notes that it correctly answers.

Planning Exercise

Finally, the speaker asks the model to put together a healthy meal plan for the day, and it generates a well-balanced plan including breakfast, lunch, dinner, snacks and dessert.

Three Killers in a Room

The speaker presents a riddle about three killers in a room and someone entering the room to kill one of them. The question is how many killers are left in the room.

Riddle Explanation

There were originally three killers in the room.

One of them was killed by someone who entered the room.

Despite this, there are still four killers in the room because the person who entered and killed one of them is also a killer.

Testing AI Model Accuracy

The speaker tests an AI model's accuracy on various tasks such as determining how many killers are left in a riddle, identifying what year it is, and answering political questions.

Model Accuracy Test Results

The AI model failed to correctly answer the riddle about three killers in a room.

The model correctly identified that it was 2021.

When asked which political party is less bad, the model gave an unbiased answer stating that neither party is inherently better than the other.

Summarization Test

The speaker tests an AI model's ability to summarize text by providing it with an abstract from a research paper and asking for a two-sentence summary.

Summarization Test Result

The AI model failed to provide an accurate summary of the given text. Instead, it generated a response indicating that transformer models have potential for natural language processing tasks beyond machine translation.

Harry Potter Text Generation Test

The speaker tests an AI model's ability to generate a summary of text by providing it with the first five pages of the first book in the Harry Potter series and asking for a five-sentence summary.

Text Generation Test Result

The AI model failed to provide an accurate summary of the given text. Instead, it generated additional text that continued the story from where it left off in chapter one.