I will evaluate the dirty talk content of these chatbots based on its LLM performance behind them and provide my own using experience as reasons for my recommendations:
SoulFun:
Large Language Model: Open Sourced LLM (possibly llama-2).
Characters: 10 female characters with different personalities and Storys. Including tennis player, maid, vampire, supermodel, etc.
Unfiltered AI: The LLM is fine-tuned by SoulFun with specific training data set to create a unique LLM which can bypass standard LLM filters, allowing personalized NSFW AI chats tailored to individual interests, creating a natural and intuitive interactions provides a more immersive sexting experience.
High Context Understanding and Consistency: The model used by SoulFun has a large parameter ranging from 70 billion which is good at understanding and maintaining a highly consistent character or persona throughout the variable topics.
And the model is also good at handling complex conversational scenarios. This includes understanding your subtle nuances, emotions, and even humor.
Moreover, characters driven by this model can even understand the content of generated images and provide relevant responses.
Long Memory: The model used by SoulFun can provide 4098 tokens of context length. Meaning the Characters will remember about 3000~4000 words.
Additional Function: Support Images generation and Voice chat.
Comment: If you want to experience the most realistic sexting with beautiful women, and also do things that are not possible in real life, then I recommend trying SoulFun. Unfortunately, its voice feature has not been launched yet, and the UI lacks ambiance.
Muah AI:
Large Language Model: OpenAI’s GPT model.
Unfiltered AI: With fine-tuned their own GPT model and data to offer a more unrestricted and personalized interaction which can provide you with lots of dirty talks or AI sexting.
Customizable AI characters: There are 13 character models (11 female characters + 2 male characters) for you to choose from, including Girlfriend, Boyfriend, Gay friend, Coworker, Classmate, Maid, Nurse, Professor, Wife, Mistress, etc
Text interaction: You can also change the direction of the conversation by highlighting text with **. However, its text descriptions are relatively short and can’t compare to a novel; it’s more like a dialog mode.
Content awareness: The GPT model has a strong capability of content understanding, consistency and memory. You can also adjust the “Temperature” and “Frequency” of the generated text in the settings. Which means you can make content more or less creative and repetitive.
Additional Function: Image generation+Phone Calls
Comment: Muah has the most comprehensive features to satisfy your needs for dirty talk or sexting. However, its text length is insufficient to support particularly long conversations.
Janitor AI：
Large Language Model: It is using Large Language Model (LLMs) from OpenAI (GPT3.5 and GPT4), Kobold AI and it has also developed its own LLM called JanitorLLM.
Diverse Character Selection: A total of 143K+ characters, with 35K+female characters and 95K+ male characters.
Character Depth: Each character has their own unique backstory and personality traits. Some of them are scenarios with more than 1 character.
Immersive AI Dirty Talk experiences: The characters would respond to your text inputs with detailed descriptions, like in a novel, depicting their actions, inner thoughts, and the story’s setting in an engaging way. And these words will be distinguished from dialogue. You can re-shape your story’s direction by interacting with its AI LLM engine to create your own unique story and experience with the same character.
Context Understanding and Response: AI will give you a very good understanding of your context and response semantically. And always follows the character’s personality traits and the setting of the story and scenario.
Context Consistency and Memory:
When using OpenAI’s GPT-3.5 model, your character typically has a context memory of around 2,048 tokens. It means using GPT-3.5 will remember roughly about 1,000 to 1,500 English words of your chat history, though this number can vary significantly depending on the specific text.
And a GPT-4 model has an 8K-token limit, it could be around 6000 English words of information.
For their own JanitorLLM, it says it has an upper limit of 8K context size as well, which gives you a memory around 6000 English words.
Comment: If you are only interested in pure text-based AI dirty talk, then Janitor is undoubtedly the best fit for you. It not only boasts the most comprehensive characters, powerful semantic analysis, and text generation capabilities, but it is also completely free. However, it is more driven by scenarios and stories, so if you simply want to chat with beautiful women, SoulFun might be more suitable.
