商店页面
AI Roguelite
全部 讨论 截图 艺术作品 实况直播 视频 创意工坊 新闻 指南 评测

AI Roguelite

商店页面
AI Roguelite >  创意工坊 > Nathan Pascal 的创意工坊
评价数不足
Improved Advanced Plausibility
描述 讨论0 留言21 改动说明
   
奖励
收藏
已收藏
取消收藏
文件大小
发表于
更新日期
373.671 KB
7 月 19 日 下午 7:44
11 月 9 日 上午 12:44
7 项改动说明 （ 查看 ）
创建者
Nathan Pascal
离线
查看所有 6 个合集（也许有些被隐藏了）
733 不重复访客数
306 当前订阅者
11 当前收藏人数

订阅以下载
Improved Advanced Plausibility

描述
Currently, the AI in Enhanced Plausibility acts more like a zealous fiction editor, not a game master. Let's try and change that.

Having trouble picking up your sword from the ground because you lack item: Sword?
Does the AI forbid you from opening your window, because you could get cold?

Changing the related prompts, the AI should get into your way far less, not judging, not assuming, only evaluating physical plausibility of your attempts.

But also I wanted to re-inforce some true limitations, to protect the story AI from its compliance compulsion, so you will still be blocked when implying unrealistic outcomes in your phrasing, "Search for Dragon Eggs" will likely auto-fail outside of a dragon's lair, spell casting is more restricted, being a mage does not simply allow to cast any spells, giving more weight to your innate and learned abilities.

This is still more like a prototype, but I need more input than I can gather from my own testing, and as always, guiding an AI is like herding cats, results are never guaranteed and strangeness may always occur. Still, I hope it might alleviate some of the frustrations I saw in the community regarding the current workings of the Enhanced Plausibility setting.
21 条留言
< >
Furoggu 11 月 20 日 上午 11:27 
I suppose the contexts I'm thinking of there isn't pressure really, because the other characters are friendly or mine is in the dominant position in e.g. an interrogation, so that makes sense
Nathan Pascal  [作者] 11 月 20 日 上午 10:28 
I think, the new prompt is more closely tuned to read the vibe of a conversation, if it is overall tense, rolls are more likely, if you sneak in a little persuasion into an otherwise rather light scene (or one where your character is very confident) it might slip through more easily.

Or would you say, it also still misses in heavier scenes, where your character is under pressure?
Furoggu 11 月 20 日 上午 10:14 
I haven't played too much recently, but I feel like this has been better. I've seen fewer cases where I've disagreed with the decision. It still quite often marks things as trivial in conversations involving a little persuasion or intimidation, but I guess rolling all the time instead of for the most important parts is a little annoying anyway
Nathan Pascal  [作者] 11 月 9 日 上午 12:50 
Alright, wanna try again? New version has greater focus on proficience and ambition to determine, if one needs to roll or not, so I hope you will never again see the words "reachable surface" as justification. That one haunts me still.

In my testing, threats and begging were considered roll-worthy, so I hope that might translate well to other causes, but I am sure, somehow somwhere I have opened up new strange cases with this.
Furoggu 11 月 8 日 下午 9:31 
It seems like sometimes it decides to focus on one or the other, either uncertain outcome or no direct obstacles, rather than considering both together. When marking as trivial it'll usually state one reason or the other for any trivial actions, and with some I can retry and it'll roll for it instead. But maybe this is an incorrect assumption
Furoggu 11 月 8 日 下午 9:28 
Just want to say that I've been trying a model from an entirely different model family and it still often interprets blatant threats or persuasion as trivial. I can literally walk up to massive, muscle-bound brutes then say I'm going to rip them limb from limb if they don't tell me what I want to know and the AI determines its basic speech with no difficulty or obstacles :PEAKthink:
Nathan Pascal  [作者] 10 月 31 日 下午 4:50 
Ah, I will try around a bit more, see if I can find a phrasing that is clearer without being too restrictive. Thanks for keeping me updated.
Furoggu 10 月 31 日 上午 1:15 
Reasonings are things like:
"Speaking a simple, improvised lie is a basic verbal action with no inherent difficulty or resistance."
"Drawing with an available tool on a reachable surface requires no special skill or effort."
Furoggu 10 月 31 日 上午 1:13 
Might be my model, but the new plaus25 prompt is weird when it comes to marking things as trivial. It feels like it does it too much, especially for speaking actions. The vanilla prompt kinda did this anyway though, so it's hard to say whether your mod has made it better, worse, or the same.

I can literally put in brackets `(this is a lie)` and it'll say it's trivial instead of rolling for deception or whatever. I decided to draw for fun and it didn't roll for art skills or anything, which I feel is a missed opportunity for a skill that'd be rarely used otherwise. It'll also roll persuasion for me telling someone something, but not for me literally pleading.

Is this just me?
Nathan Pascal  [作者] 10 月 26 日 下午 6:56 
Updated. Removed "too vague" as an option for the AI, and corrected a misleading wording that could lead actions to auto-fail that should require a roll check. (Funnily enough, base free never had a problem with that, but now testing Silver, it immediately and consistently caused that error.)