Plausibility V2
Months ago I discovered a problem while playing my game. I would do most actions, especially during combat, with little to no thought. I would use “grow crops” to fight a bear knowing that the outcome was entirely dependent on my skill in crop-growing rather than how sensible the action actually was. I would throw a crumpled note at a zombie which had about the same chance of defeating it as using a literal flamethrower.
Thus I started to brainstorm a way to make the player’s decisions actually matter, and reward actions which are strategic, sensible, and effective according to the story context, while penalizing nonsensical actions.
Under this new system, the AI takes your proposed action into account based on the story context, and evaluates it based on a number of metrics:
- Physical plausibility: Whether the action makes physical/logical sense.
- Likelihood of success: How generally effective the action should be. For example, throwing a crumpled note at a zombie is physically trivial, but highly unlikely to do anything.
- Relevance of item/ability to action (if using an item or ability): Penalizes or rewards you based on how effective the item or ability should be in context. For example, you may suffer a penalty if using “hack robot” on a non-robotic entity.
- One AI-generated criterion: The AI model is given free reign to impart its flavor for this one. In one playthrough, I got a bonus for reviving a dog because "everyone wants Sparky to survive".
These ratings then become bonuses and penalties to your attribute/skill check roll.
A key change is that all actions are evaluated, not just open-ended, so you can no longer just spam some random item or ability at everything under the sun without penalty (unless I suppose it has an OP name).
If physical plausibility is rated at the maximum and likelihood of success is relatively high, the action is deemed Trivial and passes without a roll check… but beware – for an action being trivial doesn’t automatically succeed; rather, it’s passed to the story generator with no suggested outcome, so it’s up to the model as to what happens next!
Updated roll mechanics, expanded roll outcomes
You now use the same roll for the attribute and skill check, but it uses a 2d50 system to maintain a slight bias towards the middle. Additionally, ability/faction bonuses now modify the roll value just like the plausibility checks, as opposed to the convoluted logic it was using before.
Instead of 4 outcomes, there are now 7 outcomes ranging from Critical Success to Critical Failure. If using the “roll” cheat code to override roll checks, you can now specify a value between -3 and 3.
Model refresh/update
Updated all official cloud models to slightly better and more modern ones across all tiers; Free, Silver, and Sapphire. The difference is most noticeable in Sapphire story generation, in which characters feel a bit more alive than ever before.
Also tested all models extensively making sure they held at least some small tangible improvement over their previous versions, and tweaked certain prompts and event checks to make them as compatible as possible. Please feel free to report issues that feel like regressions, as it is often the case that an overall improvement doesn't imply improvement in everything.
Use ability/item without target
To prevent the oft-complained about case where you would eat your bedroll instead of sleeping in it, or hit yourself on the head with a book instead of reading it… You may now use an ability or item without specifying a target, by clicking the “no target” icon next to the player portrait.
Specify usage details for using ability/item and interacting with static object
To prevent the oft-complained about case where you would rub spaghetti over someone’s face instead of feeding it to them… You may now hold Shift when using an ability/item or interacting with a static object to specify the nature of your action. These custom player instructions are also folded into the Plausibility V2 mechanic for your roll check bonuses and penalties.
Status effects revision
Fixed a number of issues with status effects. They are now included in the prompts correctly. Character and static object status effects decay within 2 turns. Location status effects are persistent unless removed or overridden by event checks. Status effects can also influence the Plausibility V2 mechanic.
Misc
- Tweaked a lot of event checks to slightly improve accuracy for most models across the board
- Fixed issue where unchecking factions when starting a portal to another universe would still generate factions
- Fixed player name not displaying in tooltip
- Fixed exploration background having constant white circle if autogenerate was disabled
- Added Sapphire image gen presets as part of the supported image gen modes for saving/loading as JSON
- Changed default combat turn mode to regular story turn instead of summarize per encounter, due to coherence issues if attempting to immediately follow up with the enemy after combat.