AI Dungeon Features
Ability to create a list of words that the AI will avoid using in it's responses, similar to the NSFW on/off option. Useful for when the AI repeatedly tries to steer a story in a direction you dislike.
First thing I'd add to it would be "sleep" and "bed" because the AI seems to constantly want my character to go to sleep.
I would definitely add undead to it , too many times ai turned me undead out of the blue.
Ban "you", "it" and "he", "does" and "is" and watch chaos ensue.
A command like the remember one it when you type into it ,it will block anything it was gonna say like if your story takes place in a world where you the last person on earth and you type in people then it will block the am from generating in other people related things in the story or another example would be when the am generated random out of place plot holes you copy and paste those plot holes in the limit command and it will block related topics to it. I think this would help a lot thank if
This is a really cool suggestion. Needs to happen.
*AddendumA very good usecase would be anything with Pokémon, where the AI is obsessed with making you go to Kanto, or Viridian City, even if you've added 'You are ... in Hoenn' to the remember. There are probably better examples out there, but this one is the most immediate example I can think of.
This would fix some of the other feature requests on a case by case basis. In other words, people who complain about too much sleep, anachronistic items &/or actions, and people who want to remove any and all curse words could remove those on their custom block lists.
I like this! Good suggestion.
Yeah, its really irritating to be wanting to run a fantasy campaign where there arent any humans. Trying to get the ai to generate monsters and creatures is really difficult. Even remembering that humans are extinct and there arent any and theres a lot of creatures doesnt do much. Id even be okay with just letting me add a list to autoreplace the words with (ie i make a list with like naga and minotaur and giant spider and it automatically replaces any generated instances of human with that)
It would be great if it also had a toggle for whether the blacklist should be a literal blacklist of words, or if it should also check against a thesaurus and block synonyms as well.
We really do need this. This has already been marked as planned, BUT IT'S BEEN 3000 YEARS!!!! Really though, it's been four month already.
"A limit command" (suggested by Leepmeep on 2020-03-06), including upvotes (10) and comments (2), was merged into this suggestion.
"Add forbidden phrases" (suggested by Trevormark on 2020-05-22), including upvotes (2) and comments (1), was merged into this suggestion.
"Remove "The next day"" (suggested by Trevormark on 2020-05-23), including upvotes (1) and comments (0), was merged into this suggestion.
Scripting would not help you much in the OP's case where you don't want the AI to introduce any people. You could script a search/replace of "man" with "tree" but if the tree kills you with an axe you will know it wasn't really a tree
"Allow users to prohibit the ai from using certain words." (suggested by LilTiffy on 2020-06-11), including upvotes (1) and comments (0), was merged into this suggestion.
"Scenario word blacklist" (suggested by Shnissugah on 2020-04-22), including upvotes (34) and comments (6), was merged into this suggestion.
What about a double-AI concept, where one AI generates the response and the other checks over it to see if it violates the "Blacklist", then modifies the weights of the generator temporarily to steer the response elsewhere? This would mean that the blacklist could be written in natural language.
Of course, if it's recursive, it could cause server strain; maybe have a 'Smart replacement' then, where the second AI modifies the weights to generate a response correlating with what is enforced in the "Blacklist"; these could be key pairs like World Entries, but instead a "key" is a statement.
So then the blacklist would serve as a very strong suggestion, which, combined with World Entries and Remember, would make it near impossible to steer the story in an unwanted direction, and there would be a low risk of grammatical errors caused by word replacements.
Whoever doesn't upvote this, I fear you
"Custom content filters" (suggested by Reverend Goat on 2020-07-06), including upvotes (5) and comments (0), was merged into this suggestion.
If you don't edit AI's response the moment it uses the word "knife", it will start stabbing people. That's one word I would love to bowdlerize or automatically replace with something else.
I was going to suggest some sort of banned tag feature, but this seems to be the feature that is reaching towards this. I'm getting tired of all my fantasy adventures suddenly involving vampires when I'm trying to hunt mythical beasts. Any way to negate a specific theme shift would be such a blessing.
Possible workaround: scripts can be used for this in scenarios, if you want to do things like replace words like "Gun" and "Rifle" with "bow" and "crossbow"
"Blacklist "THE END" from the AI" (suggested by Anonee on 2020-07-27), including upvotes (1) and comments (0), was merged into this suggestion.
I general blacklist would be nice... though they AI may still attempt to convey the same related message. For instance, blocking the term "The End" may only result in the AI using "fin" instead... Obviously, making to the AI avoids related terms might cause more harm then good for story generation.
A 'clbuttic' option could be added, if players do not mind, which would guarantee that certain letters never be used in that specific order, ever, even inside of other words or with spaces, dashes, etc. https://en.wikipedia.org/wiki/Scunthorpe_problem
I like this, also a wildcard option so it blocks related words.
"A /ignore" (suggested by Clear Skies on 2020-08-06), including upvotes (1) and comments (0), was merged into this suggestion.
Now that scripts exist, this is no longer something that I'd personally need. It's easy enough to just script the scenario so that it replaces the 'blacklisted' words.
"Custom word filter" (suggested by Mass on 2020-08-15), including upvotes (1) and comments (0), was merged into this suggestion.
While the obvious way to implement this is to simply auto-retry if the response contains a blacklisted word, if I understand the model right there might be a better way; artificially setting the probability of those words to 0 as the prompt is being generated. If possible, this would let you also set "discouraged" words that have their probability or weight reduced, or even *encouraged* words that have it multiplied.
100% upvote. I have PTSD from my abusive father in real life, and I'd love to be able to play AI dungeon without a pang of anxiety/bad memories flooding my mind because it suddenly brought dad into my story. Even undoing it doesn't stop my mind from already going off into a bad place. It would really mean a lot to have a custom word filter.
Just another voice backing this, not only is this just a usefull feature to promote consistency it may also help train the AI out of constantly taking shortcuts, if you were to ban the names Jacob and Karth it might actually use the history and world data. Also, The End, sick of seeing those words whenever characters go to sleep with no immediate conflict.
So, I actually think that this feature may be one of the biggest boons to the user. If you can limit the pool of information that the AI draws from, it seems like far fewer "LOLRANDOM" moments would occur that completely throw you out of immersion.For example, if it is too difficult to add "laws" to the game world via the remember / world info tools with statements such as: "X is NOT Y.", maybe being able to block any strings with content the player has blacklisted would do a similar job.
Would like to have a white list to be able to encourage specific things (including potentially world info) in addition to a black list.
Essentially, this suggestion is about providing a custom censorship list so that players can add and remove specific words to be censored. This list would have two different options for censorship, replacement (Like "Rape" to "Respect"), or blacklisting (Similar to SFW mode).
This would allow the player to blacklist specific words they have a problem with and leave the rest of the NSFW words alone. Furthermore, stories with words on the user's blacklist would not appear when they search on t
Custom blacklists are an idea that has come up many times. A lot of other suggestions would also be covered by this mechanic, such as lifestyle games and keeping genre/setting consistent, so I think players should be able to do this on a by-adventure basis.As for the "rape" thing, I don't think any filtration should be unnaturally imposed on the players regardless of their user-settings. It'd be better to have another default toggle-filter for extreme themes, "rape" just being one example.
I had not originally thought about using the blacklist for things such as genre consistency, but I do agree that it would be a great addition. With this in mind, I believe it would be beneficial to have two types of censor lists, one that censors in general and ones that are attached to a specific adventure.
"Custom Censorship List" (suggested by Err on 2020-08-28), including upvotes (3) and comments (2), was merged into this suggestion.
Wouldnt it make sense to merge https://aidungeon.featureupvote.com/suggestions/73561/force-theme-genre-to-avoid-out-of-place-events this into this thread also? It is from the intent the same thing
Not really, 73561 is about a whole theme, this suggestion is just for words.
If you were using the Story mode, and you had an "In quotes option" for ignoring words within quotes, the words 'you' and (your character's name) could be blocked, and so the AI would never speak for you and/or about you directly.
I meant that you'd be able to block 'you' and your character's name outside of quotes, not within. But both work.
Teaching a second AI to do direct word replacements might also work; as a contributor task, so that entire concepts can be blacklisted.
"Ability to block chosen words from appearing" (suggested by Ynnesh on 2020-09-11), including upvotes (1) and comments (0), was merged into this suggestion.
The whole word blacklist is slowly becoming real (Thanks to Zynj finding it in the data of adventures and here's the proof he provided https://cdn.discordapp.com/attachments/653773513857171479/754721685999059124/unknown.png)
"Banned words" (suggested by hexedbug on 2020-09-29), including upvotes (1) and comments (0), was merged into this suggestion.
The first word I would add to my blacklist would be "eventually". It's just the AI's way of saying "yada yada yada let's skip to the end."
I'd also suggest instead of just a word blacklist, a word replacement. Something like man->dude, boy->dude, girl->dude, etc lol. More than being useful, it could be fun. You could change "yes" to "borp" and "the" to "Glargh" and other wackiness. Even apply cultural based replacements for accents like a Dracula, or a simple "color"->"colour", etc.Then a theme blacklist that would blacklist themes of the terms you add. Like if you blacklist "violence" it wouldn't attempt any violence.
0 / 1,000
* Your name will be publicly visible
* Your email will be visible only to moderators
When submitting your email with a suggestion, here's our policy on what we do with the email.
We use your email to
Your email is never displayed to other users. AI Dungeon staff will be able to see your email address.
We’ll email you when the status of the suggestion changes. You can unsubscribe at any time with a single click.
* Comments and upvotes from this suggestion will be moved to another suggestion.
You cannot undo changes once they're done.
Edit the selected suggestion and create a new one.
Comments and upvotes will also be moved.