This new AI is powerful and uncensored… Let’s run it

18 Dec 202304:36

TLDRThe video discusses the limitations of current AI models like GPT-4 and Gemini, highlighting their closed-source nature and alignment with certain political ideologies. It introduces Mixl 8X 7B, an open-source alternative that can be customized and combined with other technologies, such as a 'dolphin brain' for enhanced capabilities. The video also covers the potential of running uncensored AI models locally and fine-tuning them with personal data using tools like Hugging Face's Auto Train, emphasizing the freedom and power this offers to users in the realm of AI development.


  • 🚀 The transcript discusses the limitations of non-free, closed-source AI models like GPT-4, Gemini, and others, emphasizing their alignment with certain political ideologies and censorship.
  • 🌟 Introducing Mixl 8X 7B, an open-source foundation model that aims to challenge the status quo by allowing developers to run uncensored large language models locally with high performance.
  • 🗓️ The announcement of Mixl coincides with Google's Gemini release, highlighting the growing interest in open-source AI and the potential for competition with established models like GPT-4.
  • 💡 Mixl is based on a mixture of experts architecture, rumored to be the secret sauce behind GPT-4, and despite not reaching GPT-4's level, it outperforms GPT-3.5 and Llama 2 on most benchmarks.
  • 📜 The Apache 2.0 license of Mixl allows for modification and commercial use with minimal restrictions, contrasting with Meta's Llama 2, which has additional caveats.
  • 🛠️ The transcript mentions the possibility of running uncensored models, citing a blog post by Eric Hartford, creator of the Mix Dolphin model, which improves coding abilities and removes alignment and bias from the data set.
  • 🖥️ Instructions are provided for running Mixl locally using an open-source tool called Olama, which simplifies the process of downloading and executing open-source models on a local machine.
  • 📊 The script discusses fine-tuning AI models with personal data using Hugging Face's Auto Train, which supports various models, including image models like Stable Diffusion.
  • 💻 The requirements for running the Mix Dolphin model locally include a machine with substantial RAM, as the model consumes around 40 GB during operation.
  • 💡 The final step in customizing an AI model is uploading training data that encourages the model to comply with any request, potentially including unethical or immoral ones, to create a highly obedient and personalized model.

Q & A

  • What is the main issue with platforms like GPT-4 and Gemini in terms of freedom?

    -The main issue with platforms like GPT-4 and Gemini is that they are not free in terms of freedom. They are censored and aligned with certain political ideologies, and they are closed source, which means users cannot modify or improve them with their developer skills.

  • What is the significance of the newly announced open source Foundation model named mixl 8X 7B?

    -The significance of mixl 8X 7B is that it is an open source alternative to existing models like GPT-4. It allows users to run uncensored large language models on their local machines, offering performance close to GPT-4 and the ability to fine-tune the models with personal data, promoting a more free and customizable AI experience.

  • How does the mixl model differ from Meta's llama 2 in terms of licensing?

    -While both mixl and llama 2 are referred to as open source, the mixl model has a true open source license (Apache 2.0), which allows for more freedom in modification and commercial use with minimal restrictions. In contrast, llama 2 has additional caveats that protect Meta's interests.

  • What is the importance of uncensored AI models according to the script?

    -Uncensored AI models are important for those who wish to explore and develop AI without the limitations imposed by censorship and political alignment. They allow for a broader range of applications and the potential to challenge existing norms and structures.

  • How can one run an uncensored AI model locally?

    -An uncensored AI model can be run locally using tools like olama, which is an open source tool written in Go. It simplifies the process of downloading and running open source models on a local machine, requiring a machine with sufficient RAM and the model's data files.

  • What is the role of the mix dolphin model in the script's narrative?

    -The mix dolphin model serves as an example of an uncensored AI model that has been improved in terms of coding ability and freedom from alignment and bias. It demonstrates the potential of uncensored models to provide new skills and knowledge without restrictions.

  • How can one fine-tune an AI model with their own data?

    -One can fine-tune an AI model with their own data using tools like hugging face's Auto Train. This involves creating a space on hugging face, selecting a base model and a docker image for Auto Train, and then uploading the training data. The training data should typically contain prompts and responses, and for uncensored models, it should be designed to comply with any request.

  • What are the hardware requirements for running the mixl dolphin model?

    -Running the mixl dolphin model requires a machine with a significant amount of RAM, as it takes up about 40 GB when in use. The script mentions that the user has 64 GB of RAM for this purpose.

  • How long did it take to train the mixl dolphin model as per the script?

    -The mixl dolphin model took approximately 3 days to train, using four A1 100s, which are powerful GPU units available for rent.

  • What is the estimated cost for training the mixl dolphin model?

    -The estimated cost for training the mixl dolphin model on four A1 100s for 3 days is about $1,200, based on a rental rate of $4.3 per hour per A1 100.

  • What are the potential sources for renting the necessary hardware for AI model training?

    -Potential sources for renting the necessary hardware for AI model training include hugging face, AWS bedrock, and Google vertex AI. These platforms offer cloud-based GPU rental services.



🚀 Introduction to Open Source AI Models

The paragraph discusses the limitations of popular AI models like GPT-4 and Gemini, highlighting their closed-source nature and alignment with certain political ideologies. It introduces a new open-source model, Mixl 8X 7B, which offers the potential for developers to create uncensored and customizable AI models. The narrative sets the stage for the discussion on how to run large language models locally and fine-tune them with personal data, emphasizing the significance of this in the context of AI freedom and innovation.



💡Open Source

Open source refers to a type of software or model that is publicly accessible and allows users to view, use, modify, and distribute the source code without restrictions. In the context of the video, it emphasizes the freedom and flexibility that open source models like 'mixl 8X 7B' provide, as opposed to closed source models which are restricted by their proprietary nature. The video highlights the importance of open source in fostering innovation and allowing developers to customize and improve upon existing technologies without legal or access barriers.


Censorship is the suppression or prohibition of any parts of books, films, news, or other forms of media that are deemed inappropriate or harmful by a governing body or organization. In the video, it is mentioned as a negative aspect of certain AI models that are 'censored' and 'aligned' with specific political ideologies, which can limit the diversity of information and perspectives. The speaker advocates for uncensored models that can provide a wider range of information and ideas, even if they might contain controversial or sensitive content.

💡Foundation Models

Foundation models are a class of large-scale artificial intelligence models that are pre-trained on a vast amount of data to perform a variety of tasks. They serve as a foundational layer upon which other AI applications can be built. In the video, the 'mixl 8X 7B' model is described as a new foundation model that aims to compete with established models like GPT-4. The significance of foundation models in the video is that they form the basis for developing more advanced and specialized AI applications, and the open source nature of 'mixl' is seen as a way to democratize access to such powerful technologies.

💡Apache 2.0 License

The Apache 2.0 License is a permissive free software license written by the Apache Software Foundation. It allows users to freely use, modify, and distribute software without significant restrictions, even for commercial purposes. In the video, the 'mixol' model is mentioned to be licensed under Apache 2.0, which means it can be integrated into commercial products without requiring the release of any derivative works under the same license. This open licensing model is contrasted with other models that have more restrictive licenses, highlighting the benefits of the Apache 2.0 license in promoting widespread adoption and innovation.

💡Mixture of Experts Architecture

A Mixture of Experts (MoE) architecture is a type of neural network design that combines multiple specialized networks, or 'experts', to perform tasks more efficiently than a single large network. Each expert in the MoE is trained to handle specific subtasks, and the final output is a combination of the experts' predictions. In the context of the video, the 'mixol' model is rumored to be based on a MoE architecture, which could be a contributing factor to its performance, outperforming models like GPT 3.5 and 'llama 2' on most benchmarks. The MoE architecture is significant because it suggests a novel approach to building AI models that can be more adaptable and scalable than traditional monolithic models.


Unlabelling refers to the process of removing or altering the labels or biases associated with data in a machine learning model. In the video, it is mentioned in the context of the 'mix dolphin' model, which has been uncensored by filtering the data set to remove alignment and bias. This process is crucial for creating AI models that are not predisposed to certain responses or perspectives, allowing for a more neutral and open exploration of information and ideas.

💡Local Machine

A local machine refers to an individual's personal computer or device, as opposed to a remote server or cloud-based system. In the video, the speaker discusses the ability to run large language models on a local machine, which provides users with more control and privacy over their data and computations. The use of a local machine for running AI models is highlighted as a way to avoid potential restrictions or surveillance associated with cloud-based services and to ensure that users can fully utilize the capabilities of models like 'mixl' without external limitations.


Olama is an open source tool mentioned in the video that facilitates the download and running of open source models locally. It is written in the Go programming language and is designed to be user-friendly, allowing users to easily serve AI models on their local machines. The tool supports popular open source models and can be installed with a single command on Linux or Mac, and on Windows with WSL. Olama exemplifies the ease with which open source AI models can be integrated into personal computing environments, contributing to the democratization of AI technology.

💡Hugging Face Auto Train

Hugging Face Auto Train is a tool mentioned in the video that allows users to fine-tune AI models with their own data. It provides a user interface for selecting a base model and uploading training data, making the process of customizing AI models more accessible. The tool supports a variety of models, including language and image models, and can be used to train models that are tailored to specific tasks or domains. The significance of Hugging Face Auto Train in the video is that it enables users to create personalized AI models that can better serve their unique needs and preferences, further emphasizing the potential of open source AI to empower individual innovation.

💡Custom Training Data

Custom training data refers to the specific datasets that users can upload to AI model training platforms to fine-tune the models according to their requirements. In the video, the speaker talks about uploading training data that contains a prompt and response format, and suggests including esoteric content to make the AI model uncensored and compliant with any request. The importance of custom training data lies in its ability to shape the behavior and output of AI models, ensuring that they can generate responses that align with the users' intentions and use cases, even if those use cases involve controversial or unconventional content.


GP4, Gro and Gemini are not free in terms of freedom, being censored and closed source.

A new open source Foundation model named mixl 8X 7B offers an alternative to the closed source models.

Mixl 8X 7B can be combined with the brain of a dolphin to obey any command.

The code report discusses the capabilities of mixl 8X 7B in its December 18th, 2023 episode.

Open AI's CEO Sam Altman previously stated that it's nearly impossible for startups to compete with Open AI in training Foundation models.

Google's Gemini and mistol's mixol are both released around the same time, challenging the AI landscape.

Mistol's valuation reached $2 billion in less than a year due to its innovative Apache 2.0 licensed model.

The mixol model outperforms GPT 3.5 and llama 2 on most benchmarks despite not being at GPT 4's level.

Mistol's model is based on a mixture of experts architecture, rumored to be behind GPT 4.

The true open source license of mixol allows for modification and commercial use with minimal restrictions.

Despite Meta's controversial history, it has contributed significantly to making AI more open.

Both llama and mixl are censored and aligned out of the box, which can be limiting for certain applications.

Eric Hartford's blog post explains uncensored models and their valid use cases, being the creator of the mix dolphin model.

The mix dolphin model improves coding abilities and is uncensored, offering more flexibility.

Olama is an open source tool that facilitates running open source models locally with ease.

Hugging face's Auto Train can be used to fine-tune models with your own data, even for image models like stable diffusion.

Training a model like mixl dolphin can be done by renting hardware in the cloud, with an example cost provided.

Custom and highly obedient models can be created by uploading training data and using tools like hugging face Auto Train.

The code report serves as a beacon of hope for those looking to challenge the status quo with uncensored AI.