Playing AI Dungeon
Understanding AI Dungeon Settings
There are two sets of settings to be aware of in AI Dungeon: your Main Account User Settings and In-Game Adventure Settings.
Main Account User Settings
Enable Beta Testing Features
Enables access to features which haven’t been released yet. Note that these features are likely to be incomplete and may contain bugs.
If you find any bugs in a beta feature, feel free to leave a report on our Feedback Site.
As the AI doesn’t understand what anything it says actually means, it could say literally anything, including potentially obscene things.
- Strict will filter both your
- Moderate means
- Off means you’ll see everything
Note that turning this off does NOT allow you to break the Terms of Service.
Show Do / Say / Story Button
Show the button for Do / Say / Story, or hide it in favor of using /do, /say, /story commands instead. This may be useful if you only ever use one mode.
Auto Line Break in Story Mode
Add line breaks after each story input instead of continuing from the last output
Show Icon Text
Show text labels beneath the command buttons. Turning this off is a good way to free up screen space if you know what they are already.
Display Command Buttons
Allows you to toggle which command buttons are visible, and what order they’re arranged in. Very useful for getting rid of buttons you don’t want to use.
Low Bandwidth Mode
Disables images. Toggle this on if you have slower internet or want faster performance.
Show Feedback Popup
Opt in to give us feedback on the AI periodically while playing. This helps us better improve your experience.
Train The AI
Has the AI sometimes give you multiple responses, which you choose from.
Banned Words List
Prevents the AI from using certain words in its output.
These settings apply to specific features of AI Dungeon’s advanced…
- AI Model Type: This setting changes the language model used for generating your adventures. Each model has unique characteristics.
- Griffin is our fastest model and also features the largest memory options.
- Wyvern is a fantastic model that offers similar complexity to Dragon, but at better speeds like Griffin.
- Dragon is our largest language model and has complex writing capabilities, but slower speeds.
- Hydra is an AI mode used with Griffin and Wyvern where the AI generates multiple responses and picks the best one, generally improving responses.
- Advanced Settings:
Free-to-play users have access to standard Griffin models. Premium users have access to more versions of our AI Models.
- Length: changes the maximum amount of text the AI will return; default is set to 45.
- Temperature: otherwise understood as the randomness or creativity of the AI, a higher value means more random or creative outputs; default is set at 0.80, and Wyvern may perform poorly at values less than this default.
- Top P: When generating, every token (word/part of word) is assigned a probability to be the next one in a sequence. Top P tells the AI to limit its search pool to a set of tokens whose cumulative probability for being chosen exceeds the Top P setting, which is configurable. This allows the AI to dynamically increase and decrease the number of words in its search pool while still maintaining control over response quality. The only difference between Top K and Top P is whether the cutoff between high-probability tokens to choose from and low-probability tokens to discard is decided based on the list position, or the sum of the probability of those tokens. Since Top P automatically adjusts the number of tokens to choose from depending on how likely those tokens are, it’s generally considered to be a more sophisticated tool, but experiment with them and see what gives you the kind of results you prefer. I decided to eat some… of: 20.1% more: 4.4% lunch: 3.0% food: 2.3% breakfast: 2.0% If P = .29, the AI will choose between tokens whose probability adds up to .29: of, more, lunch, and food.
- Presence Penalty: applies a fixed bias only to tokens that appeared at least once; default is set to 0
- Count Penalty: applies a bias proportional to the number of times each token appeared before; default is set to 0
- Frequency Penalty: same as count penalty, but divided by the total number of tokens in the text; default is set to 0
- AI Memory Length: how much of the story the model remembers, otherwise known as the context window; only adjustable by Premium users using Griffin models.
What is a token? AI language models can only process a limited number of characters at once, so to maximize the amount of text the AI can process, words are translated into tokens, which are essentially shortcuts pointing to common words, part of a word or symbol. Using tokens increases the number of concepts the AI considers when building a sentence. The text you input is converted to tokens, and sent to the AI to generate the responses.
How are AI responses generated? To create an AI response, we first take all the tokens (which includes recent generations, user written story prompts, and world details) and store them in memory. Then, when the AI begins to generate a response, it looks at all of the tokens in its memory, and, drawing from a database of prewritten text, decides which token would be most likely to come next. This database, which the AI has previously analyzed, is composed of terabytes of prewritten text from novels, games, and other sources on the web. It puts this new token in its memory, and repeats this process over and over again until it has generated a completed response, which is sent back and displayed in AI Dungeon. A simpler way of understanding it is like repeatedly using a giant, supercharged version of the predictive text keyboard used on phones.
- Normal ~1024 tokens
- Longer ~2048 tokens
Here you can change multiple aspects of how your AI Dungeon experience looks.
- Theme: choose between dark, light, or high contrast black
- Color: changes the primary colors of the entire AI Dungeon app; default is shown below
- Game font: changes the primary font of your AI Dungeon games; choose between Courier, Helvetica, Times New Roman, or Classic
- Text Size: changes the game text size
- Text Speed: changes the speed that the game text is displayed
- Web Action Window Size: changes how many actions the game loads while playing on the web app
Here you can:
- Change your email or password
- Link your Apple, Google, and/or Steam accounts
- Address your AI Dungeon Membership
- Apply to our AI Dungeon Creators Program
- Delete your AI Dungeon account
In-Game Adventure Settings
These settings can be accessed from the gear icon in the top right corner of your current adventure.
Art Settings – Art Settings
- Choose your Font, Text Size, and Text Speed
- Toggle on or off Auto Line Break, Full Width, or Dark Mode
- Change your Accent Color
- Show or hide the Command Buttons (by default, all buttons are displayed above the Input Box in the AI Dungeon user interface)
Continued [•••] Settings
Add link here
Add link here
© Latitude 2022