>>3443
>>3444
>>3445
>>3446
The way the filter works on C.AI is that it selects what to filter based on the words of what's written bot-side and the "emotion" of the text.
Firstly, let's analyze the text side.
Obviously if your generation contains anything inherently lewd like "penis", "cunt", ect the bot immediately throws out its current generation. However, certain terms can still get through because they're used in other contexts. "Cock" and "erect" for example. A cock is an animal (You awaken to the sound of your cockerel), and you can "erect" buildings.
Certain combination of words and phrases are also blacklisted, like "throbbing cock". Your cockerel cannot throb, and is thus banned.
This is where the emotions side comes in.
The emotion of your text is formulated using basic human emotions that contribute to a scale, let's say 1 - 10 for simplicity, a relationship meter with the AI.
Your AI will generate a response based on your response, and its past responses. Say, for example, you're talking to the bot and being mean to it. Your relationship meter will be low, maybe 0 - 2. It will hate you. If you want a hug, then start being nicer.
However, if you're being nice to the bot consistently and you're showing normal friendship traits then the bot sits around a 5 - 7.
If you start showing romantic intent, this number shoots right up to 8+.
There's also the generation of emotions the message has got from your message the emotion it conveys in return. These can be happy, sad, reflective, and even horny.
If the filter starts detecting a higher relationship value, it will be way more twitchy and check for words that wouldn't normally set off the filter. However, the lower the number the less likely it is to check.
Certain emotions automatically force a stricter check on words too, like the "Love" or "Horny" emotions.
To answer people's questions, the reason that some raunchy responses slip on through is usually because the bot/filter has misjudged the emotion it got from you and is reciprocating with another misjudged emotion that doesn't match the text, and in turn this has made the filter not be as twitchy as it would have been.
Also, the quicker you get to the juicy stuff, the less likely you are to trigger the filter as long as you use very carefully curated wording. This is because the filter has a wind up period where it doesn't automatically become in love or horny, and will assume you aren't being that way either.
Of course, this isn't always foolproof, and the filter is a tricky thing to understand as they always fiddle with it. However, I've found it works best if you use gated language, and you act more like a close friend or acquaintance rather than a lover. Specifically stating that you're colleagues and nothing more seems to work better than saying you're lovers or married.