Even for those of us who play fair and write our own content, AI saves us a lot of work. But there is a problem, so we will see some uncensored Artificial Intelligence models that we can use in Linux.
By problem I mean results censored for legal, ideological or political correctness reasons.. This censorship often borders on the rude. Microsoft Designer refused to create an image of a dachshund for me, and Canva refused to create one of monkeys typing. Bitch and monkey are offensive terms for some minority groups.
Beyond the clickbait image that illustrates the article, this type of AI is not limited to the creation of pornography but It is used for all content that for whatever reason is not generated by normal AI for whatever reason.
Defining concepts
Although Artificial Intelligence, since the appearance of ChatGPT, is part of our daily lives, Not many people are clear about the important concepts. Two of them are Large Language Models (LLM) and Generative Artificial Intelligence
Large Language Models
Large Language Models (LLM) They are designed to work with natural language. They are trained with huge amounts of data in text format. Since they are able to predict which word or phrase should appear next In a specific context, they are able to produce coherent and useful text to answer different queries.
Generative Artificial Intelligence
While LLMs focus on text and interaction through the use of natural language, Generative AI creates new multimedia content from the information it has been trained on. This content can be, in addition to text, music, video, images, code or audio.
Uncensored models
An uncensored artificial intelligence (AI) model is a special class of Large Language Model or generative AI system that does not include artificial constraints on the generation of responses. To achieve this:
- Layers of moderation are removed.
- Data sets are used without prior filters.
- They are published with open licenses so that anyone can modify them.
Uncensored Artificial Intelligence Models
While some of these models can be used on Linux locally, as long as we have the right hardware. Others are available online and require registration and even subscription fees. The advantage of local models is privacy, but the downside is that not all of us can afford the right hardware to run them.
Local models
In the article you see cited below this paragraph you will see how to install Ollama and use the models I review below.
- llama2-uncensored: Custom version from the Meta model in which two independent developers removed the restrictions.
- dolphin-mixtral: Uncensored version from the Mixtral model mix which is ideal for coding tasks.
- dolphin-mistral: No, it's not a repeat of the previous one with a typo. This model Unrestricted is based on the original Mistral and also excels at coding tasks.
- wizard-vicuna-uncensored: Another model based on Llama 2 which comes in 3 variants: 7, 13 and 30 billion parameters. The 7 version can run on computers with 8 GB.
- dolphincoder: Based on StarCoder 2, this model It is ideal for coding tasks. It has a variant that can run on 8 GB and one on 32.
- dolphin-phi: It is based on Phi, a family of small-sized Language Models from Microsoft.
- wizardlm-uncensored: Simply the unrestricted version of the Wizard LM model.
- everythinglm: The term “context window” indicates the amount of information that the model can process simultaneously. A 16K context window can handle up to 16,000 tokens, equivalent to approximately 8,000 words or 16 pages of text. This is what this Llama 2-based model can do.
Templates for online use
- Venice AI: Offer the ability to chat without censorship. Responses are stored on the user's device.
- Uncensored AI: It claims to generate all types of content without any type of censorship.