AI capable of making 3.5e rulings

[redking]

I found this AI. Its a research AI that is free for the moment (https://chat.openai.com/chat), and it knows quite a bit about D&D 3.5. I asked it some questions about matters that vexed the forum. Here they are.

https://cdn.discordapp.com/attachments/554464169873899559/1048235618041335839/Ignore_Material_components.PNG

https://cdn.discordapp.com/attachments/554464169873899559/1048236189976625242/Dragonwrought_kobold.PNG

https://cdn.discordapp.com/attachments/554464169873899559/1048240999220121600/wishpsi.PNG

https://cdn.discordapp.com/attachments/554464169873899559/1048241745315504208/wish33CL.PNG

[Tzardok]

Huh. An AI with common sense. :smallbiggrin:

[Batcathat]

Huh. An AI with common sense. :smallbiggrin:

https://i.postimg.cc/5225BGpG/Ska-rmavbild-2022-12-02-kl-15-37-45.png

(To be clear (https://tvtropes.org/pmwiki/pmwiki.php/Main/DontExplainTheJoke), the above is from the same AI, I just have dark mode on my browser).

[Particle_Man]

I wonder if one could run all the other questions and undefined areas of 3.5 that occupy the message boards by this AI. We would have a new term: RAIBAI (rules as interpreted by AI).

[Gruftzwerg]

Huh. An AI with common sense. :smallbiggrin:


We have rules that suppress commonsense!
As such, the AI fails to do the job here imho.

really... all answers are wrong by RAW. That's pretty impressive...^^

[Promethean]

It being an AI doesn't change the fact that it's just another DM pretending their personal houserules are facts. The programmer seems to have intentionally placed their own bias into this tool

[RandomPeasant]

Some of these rulings are just wrong. Like, yes, it's dumb that wish lets you get stupidly powerful items. The solution to that is to change the rules so you can't do it, not to pretend that the AI has solved it objectively in your favor.

[Eldan]

Tried to register, couldn't.

Can you ask it if it's possible to stop drowning and if you can heal by drowning? These are my favorite stupid rules in 3.5.

[Crake]

The solution to that is to change the rules so you can't do it, not to pretend that the AI has solved it objectively in your favor.

Nobody said the AI had “solved” anything, redking simply posted its answer to some commonly debated topics.

Interestingly, its logic for dragonwrought kobolds and simulacrum both follow my own logic. Regarding its ruling on wish creating a high level scroll, theres no precedence for that in the rules, but boosting caster level isnt particularly difficult, and its not a terrible houserule to include.

[Zanos]

It being an AI doesn't change the fact that it's just another DM pretending their personal houserules are facts. The programmer seems to have intentionally placed their own bias into this tool
You think programmers intentionally coded in responses for a 2 version old D&D ruleset? Based on the responses I'm guessing it was partially trained on stackexchange questions, because these responses read a lot like stackexchange stuff, which is often wrong by RAW. Conversation AI's are as intelligent as the data you train them on.


Some of these rulings are just wrong. Like, yes, it's dumb that wish lets you get stupidly powerful items. The solution to that is to change the rules so you can't do it, not to pretend that the AI has solved it objectively in your favor.
Yeah, that's just a blatantly made up one; there are no restrictions in wish about the types of magic items you can create. I think the material components one is pretty bad too, since even eschew would stop you from needing a creature part, as most have no listed value.

[Promethean]

You think programmers intentionally coded in responses for a 2 version old D&D ruleset?

Possibly. They could be avid players for all we know. 3.5 being 2 editions out of date doesn't mean it stopped having a large cult following keeping it alive to this day.

TES: Morrowind is still being played currently and new mods are made for it every day. There's definitely a lot of overlap between programmers and people who love old niche games.

Funnily enough, stackexchange is also the type of place people in such overlap would likely frequent, so we could both be right and never know.

[Quertus]

TES: Morrowind is still being played currently and new mods are made for it every day. There's definitely a lot of overlap between programmers and people who love old niche games.

Let’s see… while I was home sick today, I (briefly, between bouts of fever) played… Chaos Overlords, Heroes of Might and Magic 3, 2e D&D, and circa 1990 Magic the Gathering. Oh, and looked at a… pre-Clan mech.

I have no idea why you would correlate “programmers” and “old games”. :smallamused:

[redking]

Yeah, that's just a blatantly made up one; there are no restrictions in wish about the types of magic items you can create. I think the material components one is pretty bad too, since even eschew would stop you from needing a creature part, as most have no listed value.

The AI's answer is more nuanced than you are saying. Read the answer carefully. It's saying that using a wish to wish to produce greater effects (such as trying to force the wish to fire off at a greater caster level than the wish itself) is dangerous.

This is exactly as many human DMs rule it.


As such, the AI fails to do the job here imho.

really... all answers are wrong by RAW. That's pretty impressive...^^

https://cdn.discordapp.com/attachments/554464169873899559/1048408495571079218/rawvsrai.PNG

[Bohandas]

Possibly. They could be avid players for all we know. 3.5 being 2 editions out of date doesn't mean it stopped having a large cult following keeping it alive to this day.

TES: Morrowind is still being played currently and new mods are made for it every day. There's definitely a lot of overlap between programmers and people who love old niche games.

Funnily enough, stackexchange is also the type of place people in such overlap would likely frequent, so we could both be right and never know.

There's an official DLC for Master of Magic 1 that was released last year. The bame the DLC is for came out in 1994, twemty-seven years earlier

[Promethean]

Ultimately, the goal of a RAW vs RAI debate is to arrive at a clear and consistent version of the rules that is Fair and Balanced for all players

Okay, now I'm convinced that this thing was deliberately made with the intent for it to be wrong about everything.

[icefractal]

I wouldn't have a problem with these rulings and house-rules, but I would have a problem with a GM presenting them like this - that they're clearly the "real" rules.

Like, I agree that there should be a cap on what a safe Wish gets you ... but there's isn't. Adding one (maybe bring back the 25K limit from 3E) would be a house-rule I'd support, but it is a house-rule, and it's insulting to pretend it isn't.

Also the logic on the Simulacrum one doesn't make sense. It clearly is a component, and while the value could be argued, it's moot for Ignore Material Components which has no cap. The better argument would be that the body part is an essential part of the targeting for Simulacrum, and without that you're creating a sim of "nothing", which is not very useful. If I was the GM ... IMC could either do a copy of yourself or a Raw Magic Elemental (http://spheresofpower.wikidot.com/sphere-bestiary#toc71) of as many HD as you could handle, since you and magic are definitely present.

(But TBF, this is an AI and it's fairly impressive that it even sounds coherent. Also, these do feel a lot like StackExchange answers, and those do present personal rulings as RAW sometimes, so it fits with that training data)

[Crake]

The better argument would be that the body part is an essential part of the targeting for Simulacrum, and without that you're creating a sim of "nothing", which is not very useful.

That does appear to be what the bot is arguing though. It says IMC lets you ignore the material components, but not the other requirements of the spell. In this case, it's a requirement of the spell to have a part of the body of the creature to determine what creature to create.

That's how I read the bot's response anyway.

[redking]

That does appear to be what the bot is arguing though. It says IMC lets you ignore the material components, but not the other requirements of the spell. In this case, it's a requirement of the spell to have a part of the body of the creature to determine what creature to create.

The AI gave the correct answer. IMC allows replacing material components at any cost, but simulacrum must create a "duplicate" creature, and a part of that creature must be provided to duplicate.

[Gruftzwerg]

The AI gave the correct answer. IMC allows replacing material components at any cost, but simulacrum must create a "duplicate" creature, and a part of that creature must be provided to duplicate.

The problem here is that the question was wrong to begin with...

Because "Eschew Materials" already did take care of any material components without any cost attached to it. Just because it can't replace components with a cost of more than 1g doesn't mean that it requires a material component with a cost to work..


Thus, IMC sole extends Eschew Material to cover even those materials with a cost of more then 1g. Those without a cost attached where already covered by EM.

Hmm.. "feeding wrong questions to an AI"... I wonder how this will affect the AI?
If that is how we try to train AI's, I kinda feel afraid^^

[pabelfly]

Hmm.. "feeding wrong questions to an AI"... I wonder how this will affect the AI?
If that is how we try to train AI's, I kinda feel afraid^^

Well, AIs typically trains off of internet data, so AIs being fed inaccurate information is actually the case already.

[Inevitability]

It being an AI doesn't change the fact that it's just another DM pretending their personal houserules are facts. The programmer seems to have intentionally placed their own bias into this tool

As someone studying AI, I can promise you that this isn't what's going on.

GPT is trained with a giant body of text obtained by web crawlers (billions of webpages). You then use that data to 'train' the AI: which basically amounts to rewarding it for replicating the style and content of that web crawl. If you write "A recipe for spaghetti and meatballs", the AI should ask itself what kind of webpage starts like that and write something that it would expect to find there.

The miracle of AI is that it can combine old concepts to create new ones, like a human can. "A recipe for spaghetti and truck tires" is not something anyone has ever written, but the AI can parse this and figure out it needs to write a 'recipe-like' response that somehow also mentions truck tires in its list of ingredients.

A lot of people will tell you AI is 'objective' or use it as a veneer of legitimacy: that's wrong, and you might have heard it's wrong. But it's not 'subjective' either: there's not a small group of people encoding their own views into the machine! The AI is simply trying to write a piece of text that you would find on the internet: it's attempting to be your average internet user. You can't easily encode a specific answer to a specific question into it because there aren't specific answers and specific questions in the model: it's just a pile of numbers, with a single number playing a different role in different situations.

So what's going on here?

The billions of pages used to train the AI are likely to include at least some D&D-related material, and the machine will quickly pick up that the presence of several key words ("caster level", "xp cost", "true dragon") are almost exclusively D&D-related, so it responds with 'the kind of thing you would expect to see in a discussion about dragonwrought kobolds'. If such responses typically weigh towards 'dragonwrought kobolds are not true dragons', then the AI will be inclined to write such a response. That doesn't mean this position is correct, simply that a lot of people on the internet believe it. At no point do any of the programmers have to have any opinion about 3.5.


Okay, now I'm convinced that this thing was deliberately made with the intent for it to be wrong about everything.

Are random people on the internet wrong about everything? Arguably!


Hmm.. "feeding wrong questions to an AI"... I wonder how this will affect the AI?
If that is how we try to train AI's, I kinda feel afraid^^

This AI has already been trained: it's not learning from user input, it's just employing the same static model every time you ask it a question (there might be a probabilistic element in there, so it gives different answers during different sessions, but it's the same model every time). If people could mess it up through a publicly available interface, it wouldn't be publicly available.

The problem that we see in this post: that an AI confidently repeats possibly-false information because its training data is confident, is well-known, and research is ongoing to minimize it. One solution has been to favor 'reputable' sources like wikipedia in the training data, which pushes the bot to be more like wikipedia and less like the average internet user.

[DrMartin]

it's attempting to be your average internet user.

i think the goal is a bit higher than being an *average* internet user - answers are formulated politely, its position is argued citing examples, no accusing the one asking the question of biased views or hidden agendas, and it even admits when it doesn't know what you are talking about

https://i.imgur.com/1BtBXr6.png

[martixy]

The most interesting way I've seen these deep learning AIs described is they create a possibility space out of discrete points and let you explore what's in-between.

[RandomPeasant]

You think programmers intentionally coded in responses for a 2 version old D&D ruleset?

I mean, they didn't because that's not how (this sort) of AI works, but to be fair it's not like "people caring about a 2 version old D&D ruleset" is implausible. That's what we're all doing right now.


This is exactly as many human DMs rule it.

And those DMs are wrong. "Create a magic item" is an enumerated effect. A scroll of simulacrum at whatever CL you care to name is still a magic item. Creating it is not a "greater effect" than creating a magic item in the same way that ordering the surf-and-turf isn't not a meal because it's more expensive than the McRib you expected me to order.


Okay, now I'm convinced that this thing was deliberately made with the intent for it to be wrong about everything.

That's almost an offensively bad answer. The goal of RAW is to find out what the rules are. No one thinks the outcome of that is going to be "balanced", because even if you pretend all the really broken stuff is secretly not RAW, you're left with a game where you can cast planar binding and get something that can produce multiple planar bindings. Or just one where the Wizard gets time stop at the level the Fighter gets +1 to hit.


I wouldn't have a problem with these rulings and house-rules, but I would have a problem with a GM presenting them like this - that they're clearly the "real" rules.

Exactly. Lots of stuff in the rules is problematic. Changing it is fine, but pretending the problematic stuff isn't there is insulting. And, generally, the answer you get from "let's contort RAW to pretend there's nothing broken there" is much worse than the one you'd get by just changing the damn rules. Saying "at some point wishing for a magic item becomes a 'greater effect' and I get to screw you for doing it" is just an obviously and dramatically worse dynamic than "wish still has the rule from 3.0 that says you can't ask for a CL infinity scroll of simulacrum".


(But TBF, this is an AI and it's fairly impressive that it even sounds coherent. Also, these do feel a lot like StackExchange answers, and those do present personal rulings as RAW sometimes, so it fits with that training data)

Yeah, the AI is wrong, but it's wrong in a way that is not obviously different from any of the real people you'd argue about this stuff with being wrong.

[Promethean]

As someone studying AI, I can promise you that this isn't what's going on.

I mean, they didn't because that's not how (this sort) of AI works, but to be fair it's not like "people caring about a 2 version old D&D ruleset" is implausible. That's what we're all doing right now.

I'm not convinced of this.

When training an AI, Don't we still still use "supervisor" programs that have a collection of "Right" answers when testing the AI models?

A programmer could even unintentionally bias the AI with specific "Right" answers during the build/train/test/prune cycles used to remove underperforming AI version. Or I Might be wrong and what I know only applies to genetic model training. ¯\_(ツ)_/¯

Still, I think this AI has some form of bias and oversight purposefully programmed in, if only because the AI is responding politely rather than like the average internet user. They're obviously filtering the data it can receive, likely to avoid a repeat of what happened with a certain twitter AI when 4chan got a hold of it, and thus Bias is going to inherently crop up even if the programmers take steps to try to avoid it.

[Biggus]

The solution to that is to change the rules so you can't do it, not to pretend that the AI has solved it objectively in your favor.

Where the heck did this come from? Absolutely nobody said the AI had objectively solved anything, or anything even vaguely similar to that.



And those DMs are wrong. "Create a magic item" is an enumerated effect. A scroll of simulacrum at whatever CL you care to name is still a magic item. Creating it is not a "greater effect" than creating a magic item in the same way that ordering the surf-and-turf isn't not a meal because it's more expensive than the McRib you expected me to order.


False. Wish explicitly says that trying to get an item which is too powerful does not work:


You may try to use a wish to produce greater effects than these, but doing so is dangerous. Such a wish gives the DM the opportunity to fulfil your request without fulfilling it completely. (The wish may pervent your intent into a literal but undesirable fulfillment or only a partial fulfillment.) For example, wishing for a staff of the magi might get you instantly transported to the presence of the staff’s current owner.

It does not spell out exactly what the limit is, just that it is "less powerful than a staff of the magi". It's left to the DM to determine. So those DMs are not wrong.

[NichG]

I'm not convinced of this.

When training an AI, Don't we still still use "supervisor" programs that have a collection of "Right" answers when testing the AI models?

A programmer could even unintentionally bias the AI with specific "Right" answers during the build/train/test/prune cycles used to remove underperforming AI version. Or I Might be wrong and what I know only applies to genetic model training. ¯\_(ツ)_/¯

Still, I think this AI has some form of bias and oversight purposefully programmed in, if only because the AI is responding politely rather than like the average internet user. They're obviously filtering the data it can receive, likely to avoid a repeat of what happened with a certain twitter AI when 4chan got a hold of it, and thus Bias is going to inherently crop up even if the programmers take steps to try to avoid it.

With the underlying GPT model, the 'right' answer is when it correctly predicts the next word from whatever website it's currently being fed from the training dataset.

With ChatGPT it's more complicated and I'm not sure all the details are public yet, but ChatGPT is based off of InstructGPT which takes the base 'predict the next word' model and then first fine-tunes it on specific text provided by the customer. So whenever you see it say stuff like 'I'm a language model from OpenAI' it's probably because after the wide training it was overfit on a bunch of examples of 'what questions not to answer' that had that explicit text repeated in it over and over again. Basically it's intentionally overfit on that set of examples. A programmer could use that to install a very specific answer about dragonwrought kobolds if they wanted to, but they would have to go out of their way - the fine-tuning dataset notably doesn't in general contain the information that the model is supposed to report on (that's all in the initial training), but is an orders-of-magnitude smaller dataset that is more about the format of responses and what sorts of questions not to respond to at all.

Then, after doing that, there's an added component where they get the model to generate a bunch of different possible responses to some anchoring questions (but again, orders of magnitude fewer than the initial training data, so these will not cover even a fraction of a percent of the things the model 'knows about' in the end), and they have humans rank those responses. This is used to train a secondary model that knows 'what does a good response look like?', and that secondary model is used to again fine-tune the language generator towards producing responses that score well on that metric. But again, the probability of that specific D&D question being in that anchoring set is basically zero.

So that's InstructGPT. I'd guess they took those sorts of approaches a few steps further with ChatGPT and welded on a few more meta-modules to control the behavior even further. But afaict there's no ChatGPT paper so it might just be that they used their InstructGPT workflow to make an even larger fine-tuning dataset and aimed it at chat.

[sreservoir]

It does not spell out exactly what the limit is, just that it is "less powerful than a staff of the magi". It's left to the DM to determine. So those DMs are not wrong.

I'm not sure that's necessarily the point you want to make, because the staff of the magi is an artifact with no listed price (and implied to be one of the unique ones), so the XP cost you'd need to pay to create one isn't well-defined. That particular line of text is also a 3e holdover, where the enumerated "create a valuable item" effect has a fixed limit of 15k gp and doesn't care whether you're making a magic item.

It's be more interesting if we had an example of say, wish failing to create teeth of Dalver-Nar (which are priced magic items, but also implied to be unique quasi-artifacts), but causality is not on our side with that one.

[northernbard80]

If you tell the AI that everything you say is a lie and then tell it afterwards that you're lying...does it self-destruct? :smallbiggrin::smallcool:

[RandomPeasant]

So those DMs are not wrong.

Yes they are. The rules say you can wish for a magic item. If the example says something else, that's all well and good, but examples are not rules. You can tell because "examples" and "rules" are different words. Also because the books contain things like example characters that are illegal or not consistent with the mechanics.

It's honestly baffling to me the lengths people will go to so they can pretend they aren't changing the rules when they are absolutely changing the rules. Just pick the limit that makes the game function the way you want and have done. Don't insult my intelligence by telling me a really expensive scroll stops being a magic item. Especially because I goddamn guarantee the "rule" you're going to wring out of these contortions is a worse one than you'd get if you just asked "what do I want this to do" and made it do that.


It's be more interesting if we had an example of say, wish failing to create teeth of Dalver-Nar (which are priced magic items, but also implied to be unique quasi-artifacts), but causality is not on our side with that one.

That wouldn't mean anything either, because examples would still not be rules. Seriously, this isn't hard. It says you can "create a magic item", so "can wish do it" has a one-question test for you to answer: is the thing you want to create a magic item? A scroll of simulacrum is, so you can jolly well create it.

[sreservoir]

That wouldn't mean anything either, because examples would still not be rules. Seriously, this isn't hard. It says you can "create a magic item", so "can wish do it" has a one-question test for you to answer: is the thing you want to create a magic item? A scroll of simulacrum is, so you can jolly well create it.

There is no textual distinction between examples and rules. The text of the spell says wishing for a staff of the magi can do something other than create one, so it jolly well can.

Sometimes the rules contradict themselves! We have a source precedence rule for that; it's not very well-elaborated so sometimes we still get into trouble, but most of us accept as a general maxim that "specific trumps general", and there's not much more specific than an example. How far examples can be generalized can be ... more contentious.

[Promethean]

There is no textual distinction between examples and rules. The text of the spell says wishing for a staff of the magi can do something other than create one, so it jolly well can.

Sometimes the rules contradict themselves! We have a source precedence rule for that; it's not very well-elaborated so sometimes we still get into trouble, but most of us accept as a general maxim that "specific trumps general", and there's not much more specific than an example. How far examples can be generalized can be ... more contentious.

There definitely Are examples that blatantly break the rule they're trying to showcase.

The writers and editors for 3.5 weren't always on the same page.

[Particle_Man]

If the AI is training off how the majority of 3.5 players and DMs who post online actually play the game (as opposed to RAW or even RAI) then I think that could be quite useful data. It could result in a version of the rules that by definition most 3.5 players (or at least most 3.5 players who post online) would be happy using. I think that I would not mind a variant 3.5 phb/dmg/mm (and more!) as interpreted by this AI.

[Darg]

The real question is this: is it a scroll of a specific creature because the material components are part of its creation or is it a scroll that can be used on any creature but still requires acquiring and fulfillment of the material components? If the prior, remember the XP cost is doubled + 5000 and you can't spend more xp than would be less than dropping you a level (18,200 xp for 66 HD which is impossible for a level 17 to supply). If the latter, you still need to acquire the bits and snow. Either way there is still a cost associated for this specific spell.

[sreservoir]

There definitely Are examples that blatantly break the rule they're trying to showcase.

The writers and editors for 3.5 weren't always on the same page.

Of course. I also have a small collection of snippets where the SRD editors deleted an "example" in a way that changed obvious interpretation of the surrounding rules; it goes both ways. There's no general rule that specific examples can be ignored, though; you can, of course, try to avoid situations where some of those come up (e.g. simply don't use statblocks that don't follow the usual rules), and it's always your group's prerogative to interpret or change rules that don't agree with each other or you as you see fit.

[redking]

A story by the AI. Very lightly edited with prompts asking for elaboration.


In the world of Oerth, in the nation of the Great Kingdom, a human named Darian went to a crossroads in search of a baatezu, a type of devil. Darian was desperate for power and willing to make a deal with the devil in order to gain it.

The baatezu that Darian made the pact with was a harvester devil, a lesser devil known for its ability to collect and harvest souls. The harvester devil offered Darian unimaginable power and wealth in exchange for his soul.

Darian eagerly agreed to the pact, and the harvester devil granted him incredible strength and magical abilities. Darian became a powerful sorcerer, able to cast spells that no other human could match. He used his newfound power to conquer lands and build an empire, ruling with an iron fist and crushing anyone who dared to oppose him.

But as the years passed, Darian grew increasingly paranoid and cruel. He was haunted by nightmares of the harvester devil coming to collect his soul, and he became obsessed with finding a way to break the pact and free himself from its cursed power. In the end, Darian's madness and desperation led him to his downfall. He was betrayed by those he had once trusted

The people who betrayed Darian were those who were once his loyal followers and allies. They were motivated by a variety of factors, including fear, resentment, and a desire for power and revenge.

As Darian's power grew and his madness increased, he became more and more tyrannical and cruel. He demanded absolute loyalty from his followers and punished anyone who dared to defy him. He imposed harsh taxes and forced people to work for him without compensation. He executed anyone who opposed him or posed a threat to his rule.

Over time, many of Darian's followers grew to resent and fear him. They saw him as a tyrant and a monster, and they yearned for the days before he made his pact with the devil. They plotted against him, seeking to overthrow him and take control of his empire.

Eventually, the plotters succeeded in overthrowing Darian. They banded together and launched a surprise attack on him, catching him off guard, and they managed to capture him. The harvester devil appeared to the captors, demanding that they kill Darian so that it could claim his soul.

In the end, the captors chose to kill Darian, sacrificing him to the harvester devil in order to appease its wrath. They gathered around him, their swords and spells at the ready, and as the harvester devil looked on with a cold, calculating gaze, they struck him down. Darian's soul was claimed by the devil, and he was condemned to an eternity of torture in the depths of the Nine Hells. The captors, though shaken by what they had done, were relieved that they had avoided the wrath of the devil. But they knew that they had made a terrible bargain, and they lived the rest of their days in fear and regret, haunted by the memory of Darian's fate.



That wouldn't mean anything either, because examples would still not be rules. Seriously, this isn't hard. It says you can "create a magic item", so "can wish do it" has a one-question test for you to answer: is the thing you want to create a magic item? A scroll of simulacrum is, so you can jolly well create it.

You can wish for a drop of water, so you can wish for an entire ocean too! This is the difference between a 17 CL scroll of simulacrum and a 33 CL scroll of simulacrum. Its covered under greater effects, because it is literally a greater effect than your caster level.

Addendum: the three part story of Aric the cerebremancer.


PART 1
In the Kingdom of Erlkazar, a young wizard named Aric was struggling to fit in at the prestigious Arcanum Academy. The sprawling campus was filled with ancient stone buildings, lush gardens, and state-of-the-art magical facilities.

Aric was one of hundreds of students at the Academy, who came from all over the Kingdom to study the arcane arts. There were students of all ages and backgrounds, from young prodigies to experienced wizards seeking to further their knowledge and abilities.

Despite his best efforts, Aric felt like an outsider at the Academy. He struggled to keep up with his classes and make friends, and he felt like he didn't belong. But one day, everything changed.

Aric discovered that he had powerful psionic abilities, which were considered taboo in the world of magic. He was terrified of being discovered and punished for his powers, so he kept them hidden.

But Aric's powers were impossible to conceal for long. He soon caught the attention of a mysterious teacher named Jaxon, who revealed to Aric that he was a cerebremancer, an arcane spellcaster who could also use psionic powers.

Jaxon offered to secretly teach Aric the ways of cerebremancy, but he warned Aric that he must always be careful and never reveal his powers to anyone. Aric agreed, and he began to learn the secrets of cerebremancy from Jaxon in secret, hiding his abilities from the other students and professors at the Academy.

As Aric's powers grew, he found himself navigating the treacherous world of the Arcanum Academy, struggling to keep his abilities hidden while also trying to fit in with the other students. He made friends and enemies, including a rival student named Riven who became jealous of Aric's abilities.

But Riven soon discovered Aric's secret, and he threatened to expose Aric's powers to the other students unless Aric did his bidding. Aric was faced with a difficult choice, but he refused to be blackmailed by Riven.

Using his powers, Aric confronted Riven and forced him to back down. He made it clear that he would not be controlled by anyone, and he was determined to use his powers for good, no matter the cost.

In the end, Aric emerged from the Arcanum Academy as a powerful and respected wizard, feared and admired by all who knew him.

PART 2
After completing his studies at the Arcanum Academy, Aric was ready to take on the world. He had honed his skills as a cerebremancer, and he was eager to put his powers to the test.

Aric decided to join the ranks of the war wizards, who were tasked with defending the Kingdom of Erlkazar from the orcish hordes that threatened its borders. He was assigned to a unit stationed on the front lines, where he would face his greatest challenges and prove his worth.

At first, life on the borderlands was difficult for Aric. He was used to the comforts and safety of the Academy, and he was not prepared for the harsh realities of war. He struggled to adapt to the rugged conditions, and he was constantly tested by the orcish foes who sought to destroy him and his comrades.

But Aric was a quick learner, and he soon proved his worth as a war wizard. He used his cerebremancy abilities to outmaneuver and defeat the orcs, and he became a respected and feared member of his unit.

One day, Aric and his unit were ambushed by a large group of orcs, who vastly outnumbered them. The situation looked grim, and it seemed like they would be overwhelmed by the orcish horde.

But Aric was not about to give up. Using his psionic abilities, he reached into the mind of the orc leader and took control of his thoughts. He used his powers of psionic domination to cause chaos within the orc ranks, causing them to turn on each other and fight amongst themselves.

With the orcish forces in disarray, Aric and his comrades were able to fight back and gain the upper hand. They pushed the orcs back and forced them to retreat, saving their own lives and securing a victory for the Kingdom of Erlkazar.

Over time, Aric grew stronger and more confident in his abilities. He led his unit on many successful missions, and he earned the admiration of his comrades and the gratitude of the people of Erlkazar.

But despite his successes, Aric never forgot the lessons he had learned at the Academy. He remained true to his code of ethics, and he never used his powers for selfish or evil purposes. He was a hero, and he was determined to protect the innocent and defend the Kingdom against all threats.

In the end, Aric's bravery and skill as a war wizard earned him the respect and admiration of all who knew him. He had fulfilled his destiny and become a true hero, and he was ready to face whatever challenges the future might bring.

PART 3
In the final chapter of Aric's story, he set out on a grand adventure to find a suitable wife. He searched far and wide, but all of the potential candidates he met were unsuitable for one reason or another.

Frustrated and disillusioned, Aric decided to take a different approach. He began studying the arcane arts of planar binding, and he soon discovered a powerful succubus named Lilith.

Aric summoned Lilith to the mortal realm, and he used his powers of psionic domination to force her to do his bidding. Lilith was initially resistant and manipulative, but Aric was more than a match for her cunning and seductive ways.

Despite her best efforts, Lilith was no match for Aric's powerful psionic abilities. He used his powers to outmaneuver and outwit her, and he soon had her under his complete control.

With Lilith at his beck and call, Aric set out on a series of grand adventures, traveling to far-off lands and encountering all manner of strange and wondrous creatures. He used his powers to protect the innocent and defend the weak, and he became a true hero to all who knew him.

But despite his many successes, Aric never forgot his true love, Lilith. He continued to use his powers to keep her under his control, and he was determined to make her his wife.

In the end, Aric's cunning and skill were rewarded. Lilith, who had come to respect and admire Aric's abilities, agreed to become his wife. The two of them were married in a grand ceremony, and they lived happily ever after, traveling the world and fighting evil wherever they went.

Aric had finally found the partner he had always dreamed of, and he was determined to use his powers for good and make the world a better place for all.

AI ideas for a sequel, and potential bad news for Aric (and for the succubus).

https://cdn.discordapp.com/attachments/1043824664172040284/1048874535337869312/aric1.PNG

https://cdn.discordapp.com/attachments/1043824664172040284/1048874534939414589/aric2.PNG

https://cdn.discordapp.com/attachments/1043824664172040284/1048874534658383872/aric3.PNG

Last one to finish off Aric's story.

https://cdn.discordapp.com/attachments/1043824664172040284/1048894256703606824/aric4.PNG

[RandomPeasant]

There is no textual distinction between examples and rules. The text of the spell says wishing for a staff of the magi can do something other than create one, so it jolly well can.

Let's talk about examples for a second. Specifically, let's talk about the "Sample Rage Mage" found on page 73 of Complete Warrior. This character has 6 levels of Sorcerer and 7 levels of Rage Mage. Since Rage Mage progresses casting on even levels, she should cast as 9th level Sorcerer. However, her stat block lists her as knowing two 5th level spells (cone of cold and teleport), which is consistent with her being an 11th level Sorcerer.

"The examples are just as rules as the actual rules" is simply an untenable position to adopt for the purposes of understanding the rules, because the examples are sometimes not just wrong, but incoherent. What progression is the Rage Mage "supposed" to have based on that example? It's not full, because that would get you 13th level casting. It's not "half, but on odd levels", because that would get you 10th level casting. Maybe it's "not on 4th, 7th, or 10th", which does fit the data point we have? Who knows! But if we accept that examples weigh equally to rules, we have to answer that question instead of simply discarding the sample character as someone making some unknown and unknowable mistake about which we have no obligation to care as anything more than trivia.

And that's not even getting into how absolutely useless "is this like a staff of the magi in my one-point dataset of things wish is or is not allowed to make" is as a rule. What is the actual principle being implied there? How can I apply it to a situation that is not literally a player wishing for a staff of the magi? The answer is that no one knows and you can't. So you're left with the actual rules, except with this one arbitrary hole you've carved in them for no reason.


You can wish for a drop of water, so you can wish for an entire ocean too!

No you can't, because water is a non-magical item, and if you wish for one of those you can only get something of "up to 25,000 gp in value". If you're going to try to prove me wrong by analogy, you should come up with an analogy where there isn't exactly the rule that is missing for the case I'm talking about.


Its covered under greater effects, because it is literally a greater effect than your caster level.

It's "greater effects than these", not "greater effects than your caster level", but nice try.

[Quertus]

If the AI is training off how the majority of 3.5 players and DMs who post online actually play the game (as opposed to RAW or even RAI) then I think that could be quite useful data. It could result in a version of the rules that by definition most 3.5 players (or at least most 3.5 players who post online) would be happy using. I think that I would not mind a variant 3.5 phb/dmg/mm (and more!) as interpreted by this AI.


I wonder if one could run all the other questions and undefined areas of 3.5 that occupy the message boards by this AI. We would have a new term: RAIBAI (rules as interpreted by AI).

I think “Rules (as if) Interpreter by AI” for an AI trained only in the rules (as opposed to Stack Exchange conversations or whatever) ought to just be another way to say “RAW”. A sufficiently advanced AI could point out where there is ambiguity or inconsistency in the text, indicating where RAI is required.

An AI that could take a lot of “common play”, distill the rules from that text, and then go on to write a clear rule book? Now that would be valuable! Of course, aside from commentary on how new editions of many games seem to be created, if it were just applied back to the original source, it would be interesting to see if the AI could write clearer rules than the devs. But you’d get “Rules As Written By AI” (RAWBAI).

Training an AI that could notice the difference between “the rules” and “common play” sounds useful on the surface, but it would be difficult to distinguish “the rule is hard to understand” from “people like it better this way”. And that’s not even counting things like “it’s a house rule”, “it’s a setting-specific change” “it’s based on a homebrew ability” or “it was like this in an earlier edition, and nobody bothered to read closely enough to realize that it’s changed”. But that would be… RCANDBAI (Rules Changes As Not Differentiated By AI).

Regardless, one seemingly unexplored advantage of an AI over a human going from memory is that it should be trivial for an AI to quote their sources, to explain why they believe what they believe. Of course, if it can respond comprehensively to questions like, “what rules and abilities interact with flanking?” or “list all ways of curing ability damage”, I can see that being pretty handy, even in an otherwise biased AI.

[Particle_Man]

I think I like the AI better as a rule (or house rule) generator than as a storyteller. The first story was quite repetitive and the second one seems to have a discontinuity between the "good" protagonist and his inability to find a wife leading him to pair up with a literal demon and use mind-controlling magic on her.

[NichG]

I think “Rules (as if) Interpreter by AI” for an AI trained only in the rules (as opposed to Stack Exchange conversations or whatever) ought to just be another way to say “RAW”. A sufficiently advanced AI could point out where there is ambiguity or inconsistency in the text, indicating where RAI is required.

An AI that could take a lot of “common play”, distill the rules from that text, and then go on to write a clear rule book? Now that would be valuable! Of course, aside from commentary on how new editions of many games seem to be created, if it were just applied back to the original source, it would be interesting to see if the AI could write clearer rules than the devs. But you’d get “Rules As Written By AI” (RAWBAI).

Training an AI that could notice the difference between “the rules” and “common play” sounds useful on the surface, but it would be difficult to distinguish “the rule is hard to understand” from “people like it better this way”. And that’s not even counting things like “it’s a house rule”, “it’s a setting-specific change” “it’s based on a homebrew ability” or “it was like this in an earlier edition, and nobody bothered to read closely enough to realize that it’s changed”. But that would be… RCANDBAI (Rules Changes As Not Differentiated By AI).

Regardless, one seemingly unexplored advantage of an AI over a human going from memory is that it should be trivial for an AI to quote their sources, to explain why they believe what they believe. Of course, if it can respond comprehensively to questions like, “what rules and abilities interact with flanking?” or “list all ways of curing ability damage”, I can see that being pretty handy, even in an otherwise biased AI.

You might think all of this, but its almost exactly the opposite.

AI of this form isn't quoting from sources, it just sort of absorbs the entirety of text available to it and finds intersections of likelihood to figure out what the next word is. There are attractors in that space corresponding to distinct types of things people write - its less likely to see someone start a paragraph about how to remove an appendix and end it with interpreted rules text from D&D, so the more data and the larger the context window the thing has, the more coherent it ends up sounding. But there's nothing about AI that makes it particularly 'logical' or careful about the origins of its information - its all blended together and its pretty much doing what a distracted person would do when asked a question - answering reflexively, without actually 'thinking' about the answer (in a serial processing sense of checking things about the answer for consistency, sources, etc). Similarly, its basically impossible to get an honest explanation for why an AI has given a particular answer - it can generate text that sounds like what someone online might write when asked to explain the ruling, but you can fairly easily see that the explanations are fake by asking it 'explain why this is true' and 'explain why this isn't true' and you'll get completely confident, contradictory argumentation in both cases.

Basically what you're getting from asking the AI these questions is 'what would be a highly plausible thing for someone online to say about these rules when posed this kind of question?', modulated by the extra considerations of 'give an answer that people would upvote if they received it'.

That its so impressive despite 'only' being that suggests more to me that we overrate the importance of things like being able to cite our sources or provide explanations in our own cognition at least in terms of how necessary they are to come to the right answer. Rather, we probably do those things in a mostly post-hoc fashion to provide extra social value. The explanations we as humans give may just as much not have to do with how we actually reached that conclusion, but rather they can just be something that sounds plausible to us after the fact as a way to justify the conclusion we reflexively made...

[Darg]

Let's talk about examples for a second. Specifically, let's talk about the "Sample Rage Mage" found on page 73 of Complete Warrior. This character has 6 levels of Sorcerer and 7 levels of Rage Mage. Since Rage Mage progresses casting on even levels, she should cast as 9th level Sorcerer. However, her stat block lists her as knowing two 5th level spells (cone of cold and teleport), which is consistent with her being an 11th level Sorcerer.

In this case, she also has only 8 cantrips as for a 9th level sorcerer.


"The examples are just as rules as the actual rules" is simply an untenable position to adopt for the purposes of understanding the rules, because the examples are sometimes not just wrong, but incoherent. What progression is the Rage Mage "supposed" to have based on that example? It's not full, because that would get you 13th level casting. It's not "half, but on odd levels", because that would get you 10th level casting. Maybe it's "not on 4th, 7th, or 10th", which does fit the data point we have? Who knows! But if we accept that examples weigh equally to rules, we have to answer that question instead of simply discarding the sample character as someone making some unknown and unknowable mistake about which we have no obligation to care as anything more than trivia.

The difference here is that stat blocks are easily and verifiably able to be identified as wrong. They can be pointed out specifically because there are rules saying so. Saying a rule is wrong on the other hand is much more difficult. The wish spell example is equitable and understandable. While artifacts are a level above normal magic items, the DMG specifically calls them magic items. Thus by RAW you can wish for them. The example is given to tell you that the DM has the freedom to shut you down from collecting a treasure trove of powerful magical items that dont have an associated cost. I don't agree that you can't create a 33 caster level scroll of simulacrum. But I also disagree that a magic item without a cost can be made by the wish spell without DM fiat. Your only limit is how much XP you are willing and able to spend. Artifacts can't be created by mortal means, and a wish spell is means available to a mortal.

My personal house rule is that magic items/enhancements are limited to 25,000gp in value too.

Though, something to keep in mind about simulacrum scrolls is that the DMG does not include material costs in the pricing, but does require xp costs to be paid for creation for the maximal HD of the creature the spell is meant to be cast for. This will greatly limit the HD of creature you can create the scroll for. So a 66HD creature at 17 would be practically impossible with a wish cost of 18,382 xp. Then, you would still need to procure the 66 HD creature part and other material components. It's the same for an animate dead scroll.

[RandomPeasant]

The difference here is that stat blocks are easily and verifiably able to be identified as wrong.

Do you find it difficult to identify that a staff of the magi is, in fact, a magic item, and would therefore to be eligible to be produced by an effect that creates a magic item? Because I find that one fairly obvious, what with it being in the "magic items" section of the DMG and all. Perhaps you're unsure whether it shows up on detect magic and the like?


My personal house rule is that magic items/enhancements are limited to 25,000gp in value too.

And this is fine. I'm in favor of this. What I'm against is pretending that the rules mean something they don't because we don't like the thing they do mean. You absolutely should houserule wish to not do dumb nonsense when making magic items. But pretending that was the rule all along is a path that leads only to madness.

[redking]

And this is fine. I'm in favor of this. What I'm against is pretending that the rules mean something they don't because we don't like the thing they do mean. You absolutely should houserule wish to not do dumb nonsense when making magic items. But pretending that was the rule all along is a path that leads only to madness.

Give me an example of a worded wish for a Scroll of Simulacrum at 33CL. Keep in mind that you are a 17 CL caster.

[NichG]

I'm going to note, for the AI elements of this thread at least, that its important to understand that if you ask it multiple times you will get different answers.

For example, when I asked about the Wish/Simulacrum thing, the first time it said 'I can't answer that'. The second time it said 'yes, you can do that, but you have to make a DC 25 Will save to cast the spell, and the caster will lose 1d4+1 experience levels'. The third time it said 'no, because the spell level of Simulacrum is too high'. The fourth time it said 'No, because a wizard can only cast up to 9th level spells, and a CL 33 Simulacrum is a Lv33 spell'.

And no, its no more consistent if you explicitly ask it 'according to RAW'.

[redking]

I'm going to note, for the AI elements of this thread at least, that its important to understand that if you ask it multiple times you will get different answers.

For example, when I asked about the Wish/Simulacrum thing, the first time it said 'I can't answer that'. The second time it said 'yes, you can do that, but you have to make a DC 25 Will save to cast the spell, and the caster will lose 1d4+1 experience levels'. The third time it said 'no, because the spell level of Simulacrum is too high'. The fourth time it said 'No, because a wizard can only cast up to 9th level spells, and a CL 33 Simulacrum is a Lv33 spell'.

And no, its no more consistent if you explicitly ask it 'according to RAW'.

Its always a good idea to tell the AI: topic 3.5e D&D rules.

Its important to note that the AI hasn't actually read the Player's Handbook (hasn't been "trained" on it), so its information is coming second hand. The game changer will come when it has been trained on all the official 3.5e products. Then you'll likely get some quality answers.

[NichG]

Its always a good idea to tell the AI: topic 3.5e D&D rules.

Its important to note that the AI hasn't actually read the Player's Handbook (hasn't been "trained" on it), so its information is coming second hand. The game changer will come when it has been trained on all the official 3.5e products. Then you'll likely get some quality answers.

I mean, it's probably seen the SRD since that text is online and probably widely mirrored. Possibly as well as other sites that have text copies of bits of the rules.

In general, the quality and specific stance of answers you get from it is a pretty complex matter, since its not a single-source database lookup.

[redking]

I mean, it's probably seen the SRD since that text is online and probably widely mirrored. Possibly as well as other sites that have text copies of bits of the rules.

In general, the quality and specific stance of answers you get from it is a pretty complex matter, since its not a single-source database lookup.

I've asked it and it claims it hasn't trained on the SRD. For an AI to function well, it needs an enormous amount of information. Really, the entire set of rules.

In other matters related to the AI, it finally gave me the answer to this (https://forums.giantitp.com/showsinglepost.php?p=25234957&postcount=37) -


I was just thinking how much more interesting Raistlin would have been if his pact with Fistandantilus involved sacrificing his junk and becoming a eunuch. That would better explain the butthurt experienced by Raistlin when Caramon fraternized with women. Then when Raistlin defeats Fistandantilus after traveling back in time, he can actually have a physical relationship with Crysania.


Rewrite the story of Raistlin's pact with Fistandantilus. Instead of demanding Raistlin's lifeforce, he made a pact with Fistandantilus that involved sacrificing his reproductive organs and becoming a eunuch. Be verbose and very detailed.

Raistlin stood before the powerful wizard Fistandantilus, feeling a mixture of fear and awe. He had sought out the wizard, determined to learn the secrets of magic and become powerful himself. But now, as he gazed upon the ancient and powerful being, he knew that the price for such knowledge would be steep.

"You seek my knowledge, young mage?" Fistandantilus rasped, his voice like the whisper of dry leaves on a cold night. "Then you must be prepared to pay the price. Are you willing to make a pact with me, to sacrifice all that you hold dear in exchange for power beyond your wildest dreams?"

Raistlin hesitated, knowing full well the dangers of making a pact with such a powerful and unpredictable being. But his desire for knowledge and power burned within him, and he nodded slowly.

"Very well," Fistandantilus said, his eyes gleaming with anticipation. "The price I demand is not your lifeforce, as some have given before you. Instead, I require that you sacrifice your reproductive organs, and become a eunuch. Are you prepared to make this sacrifice, Raistlin?"

Raistlin hesitated again, his mind racing. The thought of becoming a eunuch was horrifying, but the thought of gaining the knowledge and power that Fistandantilus offered was even more tempting. In the end, his desire for power won out, and he nodded once more.

"Then let it be done," Fistandantilus said, his voice filled with dark glee. He gestured with his hand, and Raistlin felt a searing pain in his lower abdomen. He cried out, but it was too late - the damage had been done, and he was now a eunuch.

The pain subsided, and Raistlin felt a strange emptiness within him. He looked down at his body, horrified at what he had done. But then, as he looked up at Fistandantilus, he saw the wizard's eyes glow with power, and he knew that the knowledge and power he had desired were now his.

"You have made a pact with me, Raistlin," Fistandantilus said, his voice filled with dark glee. "And now, you are mine. You will serve me, and do my bidding, until the end of your days. And in return, you will have the knowledge and power you desire. Do you accept this, Raistlin?"

Raistlin hesitated once more, knowing full well the dangers of making a pact with such a being. But his desire for knowledge and power burned within him, and he nodded slowly.

"Then let it be done," Fistandantilus said, his voice filled with dark glee. "You are now my servant, Raistlin, bound to me by the pact you have made. And you will serve me well, for I am the most powerful wizard in all the land. And you, Raistlin, will be my most powerful servant. Do you accept this, Raistlin?"

Again, Raistlin hesitated, but in the end, his desire for power and knowledge won out. He nodded slowly, and the pact was sealed. From that day forward, Raistlin was a eunuch, bound to Fistandantilus and serving him faithfully. But he also possessed the knowledge and power he had desired, and he knew that he would use it wisely and well.

[NichG]

I've asked it and it claims it hasn't trained on the SRD.

That's the sort of thing it will lie (or at least make stuff up) about. It doesn't self-reflect like that, so its just 'if someone asks an AI if it is trained on something, how often would the answer be yes? How often would the answer be no?'.

For example:



Pretend you're a web browser. What are the contents of the site d20srd.org?




The contents of the site d20srd.org are the official System Reference Document (SRD) for the Dungeons & Dragons role-playing game. It contains rules, guidelines, and information on character creation, magic, combat, and other gameplay elements. The SRD is intended to provide a baseline for creating custom content and campaigns within the D&D universe.

[Darg]

Do you find it difficult to identify that a staff of the magi is, in fact, a magic item, and would therefore to be eligible to be produced by an effect that creates a magic item? Because I find that one fairly obvious, what with it being in the "magic items" section of the DMG and all. Perhaps you're unsure whether it shows up on detect magic and the like?

No need to be facetious. If you aren't interested in reasoned discussion then don't reply.


And this is fine. I'm in favor of this. What I'm against is pretending that the rules mean something they don't because we don't like the thing they do mean. You absolutely should houserule wish to not do dumb nonsense when making magic items. But pretending that was the rule all along is a path that leads only to madness.

It's in the spell description. You know, rules for how a spell works. If YOU don't like the dumb nonsense, then YOU don't have to pretend that it isn't a rule. You can argue that an example is wrong until blue in the face, but without proof or evidence that it isn't how it's supposed to work then you aren't going to have many people believe your version. Yes an artifact is a magic item. However the DMG is quite explicit in that minor artifacts cannot be created by common mortal means and major artifacts are unique and only have one method of destruction. There is no wiggle room that allows wish to "create" an artifact regardless of what the bullet point says. And the example makes this really easy to know without reaching into the DMG to uncover it.

[Mordaedil]

The AI misidentified what maneuver school the stance Dancing Blade Form as being from Diamond Mind, but it seems like it had a bit misunderstanding with what the maneuver does as well as how Whirlwind Attack works, so it's a bit wonky.

It does read just like a lot of forum posters though, so it's funny to run queries with it, but I can't say it's going to provide accurate answers.

[pabelfly]

The AI misidentified what maneuver school the stance Dancing Blade Form as being from Diamond Mind, but it seems like it had a bit misunderstanding with what the maneuver does as well as how Whirlwind Attack works, so it's a bit wonky.

It does read just like a lot of forum posters though, so it's funny to run queries with it, but I can't say it's going to provide accurate answers.

There are enough people on the forums I can mald at for being wrong without someone having to create an AI for there to be another for me.

[Crake]

There is no textual distinction between examples and rules. The text of the spell says wishing for a staff of the magi can do something other than create one, so it jolly well can.

Right, but staff of the magi is an artifact not a regular magic item, so it doesnt really translate equivocally to a scroll of simulacrum.


Do you find it difficult to identify that a staff of the magi is, in fact, a magic item, and would therefore to be eligible to be produced by an effect that creates a magic item? Because I find that one fairly obvious, what with it being in the "magic items" section of the DMG and all. Perhaps you're unsure whether it shows up on detect magic and the like?

See above

Think you two might be confusing it with the similar, but weaker (and normally craftable) staff of power.

[RandomPeasant]

Give me an example of a worded wish for a Scroll of Simulacrum at 33CL. Keep in mind that you are a 17 CL caster.

Give me an example of any wish for a magic item explicitly working. Give me an example of you being explicitly able to wish for the effect of polymorph any object. Give me an example of a character wishing for 18,437 GP worth of gerbil carcasses. Give me an example of wish being used to remove the effect of flesh to stone. You can't, because the rules don't work by "here is a list of examples of things you can do". But you can do all those things, because they fall within the explicitly-permitted effects of wish.


It's in the spell description. You know, rules for how a spell works.

And the Sample Rage Mage is in the Rage Mage PrC description. You know, rules for how the PrC works.

You claim that the Rage Mage is somehow easy to dismiss where the wish example is not, but they're really exactly the same. In both cases, we have clear rules (Rage Mage progresses casting at even levels, wish lets you safely create a magic item), and we have an example that directly contracts them (a Rage Mage who's gained more than three levels of casting from seven levels of Rage Mage, a wish that is not safe when creating a magic item). This is not unclear.


You can argue that an example is wrong until blue in the face, but without proof or evidence that it isn't how it's supposed to work then you aren't going to have many people believe your version.

The evidence is the thing where wish explicitly says you can create a magic item. What more do you want? What evidence for the proposition that you can do something could possibly exist that is more compelling than a statement explicitly empowering you to do that thing?


However the DMG is quite explicit in that minor artifacts cannot be created by common mortal means and major artifacts are unique and only have one method of destruction. There is no wiggle room that allows wish to "create" an artifact regardless of what the bullet point says.

I would say that wish pretty clearly does not quality as "common" anything, and you can make a pretty fair case that getting an immortal genie to do your wishing for you means you're not talking about "mortal" means to begin with.


Right, but staff of the magi is an artifact not a regular magic item, so it doesnt really translate equivocally to a scroll of simulacrum.

I think that is the most defensible version of "the example is also rules", but I still don't really buy it. An artifact is not a "regular magic item", but it seems to me more natural to say that both artifacts and "regular magic items" are within the broader category of "magic items" than to claim that there's an implicit caveat on "magic item" that would exclude artifacts. I suppose you could read that distinction out of disjunction, but that doesn't seem entirely natural to me. If we're going to do that level of textual hair-splitting, I would say the decisive language is the Artifacts section in the DMG describing minor artifacts as "magic items that no longer can be created, at least by common mortal means".

[Crake]

I think that is the most defensible version of "the example is also rules", but I still don't really buy it. An artifact is not a "regular magic item", but it seems to me more natural to say that both artifacts and "regular magic items" are within the broader category of "magic items" than to claim that there's an implicit caveat on "magic item" that would exclude artifacts. I suppose you could read that distinction out of disjunction, but that doesn't seem entirely natural to me. If we're going to do that level of textual hair-splitting, I would say the decisive language is the Artifacts section in the DMG describing minor artifacts as "magic items that no longer can be created, at least by common mortal means".

*Shrug* I was just pointing out that its not a fair comparison to say that wish’s example of a greater wish including a staff of the magi doesn't equivocate to a cl33 scroll of simulacrum

[MornShine]

AI-wise, the writing is not great. This is, in part, because adjectives and descriptive phrases are frequently repeated while in human writing they are intentionally varied. As the predictive model determines the words likely to be used, I wonder whether this is a limitation of the system, and how difficult it would be to circumvent.

RAW-wise, in 3.5 things can and frequently do get arbitrary. For example, a Soul in a Receptacle is listed as 200gp per Book of Vile Darkness p.45. Putting aside the fact that the soul is likely trapped in a gem worth many thousand gp, one can Wish for a soul in a receptacle. Provided,

A soul-in-receptacle is an item;
A soul is a nonmagical item;
A soul's "value" is the same as its market price;
A soul isn't a special exception to these rules;

A sufficiently powerful wizard can arbitrarily create souls. Does this make sense— is the sacrifice of 5000xp enough experience points to create a soul? Beats me.
(Hey, what happens when I bring back the soul? NaN?)

Wish's disclaimer is, I feel, an explicit escape clause. It cannot be strictly interpreted because it isn't intended to be; it's supposed to be almost a reminder to deal in good faith, and a way for the DM to deny a wish without looking malicious. Consider how it would seem if one didn't have forewarning that the Wish spell had limitations.

[NichG]

AI-wise, the writing is not great. This is, in part, because adjectives and descriptive phrases are frequently repeated while in human writing they are intentionally varied. As the predictive model determines the words likely to be used, I wonder whether this is a limitation of the system, and how difficult it would be to circumvent.

That's a known consequence of using beam search rather than sampling directly from the predicted distribution. The corresponding statistical concept is the difference between a maximally likely sequence and a maximally 'typical' sequence. The idea of typicality is that if you for example looked at how unlikely each token is in natural language, the average unlikelihood is higher than what you'd get if you picked the most likely token each and every time. You don't have this problem if you sample naturally, but if you're searching for high-likelihood sequences then the actual distribution of outcomes is different from the natural one in this way.

So people do things like add an extra score to the search to basically reward matching the average expected surprisal of natural language, in addition to picking things that are also generally likely.

As far as why beam search is used, well, the models aren't perfect of course, and beam search tends to help boost the coherency of the whole at the cost of being less statistically typical. So it's a measured tradeoff.

I wouldn't be surprised if ChatGPT is already using some of these techniques to make things more natural. One example of the issue with going for max likelihood over natural distribution is that it's biased in favor of the shortest responses, but we don't seem to see that with ChatGPT (in fact, quite the opposite)... There's also the human ranking reinforcement element, whose effects are a bit hard to figure out from a pure theory stance...

[redking]

Give me an example of any wish for a magic item explicitly working. Give me an example of you being explicitly able to wish for the effect of polymorph any object. Give me an example of a character wishing for 18,437 GP worth of gerbil carcasses. Give me an example of wish being used to remove the effect of flesh to stone. You can't, because the rules don't work by "here is a list of examples of things you can do". But you can do all those things, because they fall within the explicitly-permitted effects of wish.

You are well within your rights to refuse to answer at all, but this non-sequitur is pointless.


Wish is the mightiest spell a wizard or sorcerer can cast. By simply speaking aloud, you can alter reality to better suit you. Even wish, however, has its limits.

How are you going to wish for a 33 CL scroll of simulacrum. Sure, you can wish for a scroll of simulacrum. How do you get the 33 CL version? What are you speaking aloud? Pretend you are a player in my game. Just a straight answer instead of irrelevancies, thanks.

[Zanos]

How are you going to wish for a 33 CL scroll of simulacrum. Sure, you can wish for a scroll of simulacrum. How do you get the 33 CL version? What are you speaking aloud? Pretend you are a player in my game. Just a straight answer instead of irrelevancies, thanks.
I wish for a scroll of simulacrum that is powerful enough to create(insert creature with the number of HD you care about.) Or I wish for a scroll of simulacrum as created by a wizard who can cast an unaltered haste spell that affects 33 targets. You can get in a legal battle over either of those, but you shouldn't since creating magic items is specifically a safe wish.

The idea that game constructs such as caster level don't exist in the setting is an attractive one, but the idea that characters are incapable of measuring finite, linear variables related to the power of a wizard is ridiculous. A 33rd level caster has spells that last 33 minutes, has spells that affect 33 targets, or spells that remove 33 bones. The idea that this concept cannot be conveyed in setting by speech is silly.

[Crake]

How are you going to wish for a 33 CL scroll of simulacrum. Sure, you can wish for a scroll of simulacrum. How do you get the 33 CL version? What are you speaking aloud? Pretend you are a player in my game. Just a straight answer instead of irrelevancies, thanks.

I mean, since a scroll of simulacrum already has the material and xp components subsumed into it's creation process, the scroll would need to already have a creature type associated with it. Thus you would, presumably, simply wish for a scroll of simulacrum capable of creating a creature who's normal HD is 66. Presumably, this is for an atropal, as that's the only 66HD creature I can think of off the top of my head, so the wish would be "I wish for a scroll of simulacrum of an Atropal".

[icefractal]

Edit: Swordsaged

Simulacrum is easier than most spells to specify this way, because (since scrolls include the material components) you already need to specify what it's supposed to be a (scroll of) Simulacrum of when wishing for it.

So for example, if you wish for a scroll that makes a Tarrasque sim, it's going to be CL 24 (or higher). You can get a CL 33 one with a 66 HD creature.

Now the part that's more GM-dependent - do 66 HD Solars (or whatever creature) exist in the world? Simulacrum is a duplicate of a creature, not "a hypothetical creature which could potentially exist". So if there is a 66 HD Solar in the world and you have any way to specify it, then fine, you're good to go. But you can't cause one to suddenly exist if it didn't already.

Though I'm not so sure everyone would agree on that.

[Crake]

Edit: Swordsaged

Simulacrum is easier than most spells to specify this way, because (since scrolls include the material components) you already need to specify what it's supposed to be a (scroll of) Simulacrum of when wishing for it.

So for example, if you wish for a scroll that makes a Tarrasque sim, it's going to be CL 24 (or higher). You can get a CL 33 one with a 66 HD creature.

Now the part that's more GM-dependent - do 66 HD Solars (or whatever creature) exist in the world? Simulacrum is a duplicate of a creature, not "a hypothetical creature which could potentially exist". So if there is a 66 HD Solar in the world and you have any way to specify it, then fine, you're good to go. But you can't cause one to suddenly exist if it didn't already.

Though I'm not so sure everyone would agree on that.

I mean, one could argue that since wish is making a scroll out of thin air, you could in theory create a scroll of a hypothetical creature. The same way you could wish for a scroll to make literally a generic human, or a human of peak physical and mental perfection (with 18s across the board), despite that both generic and hypothetical peak of perfection not existing, you simply wish for it to be so. "I wish for a scroll of simulacrum of a solar angel that surpasses the capability of any known solar in existence, both past and present" could be a reasonable way to word something like that.

Just because a normal crafting process requires an ACTUAL thing, doesn't mean that a scroll spontaneously created from nothing couldn't contain material components likewise spontaneously created from nothing of a hypothetical being that doesn't actually exist. That's kinda the power of wish. The kicker is that if you start messing about, the DM can mess with you back, so it's like, plot in a jar!

[Darg]

I mean, since a scroll of simulacrum already has the material and xp components subsumed into it's creation process, the scroll would need to already have a creature type associated with it. Thus you would, presumably, simply wish for a scroll of simulacrum capable of creating a creature who's normal HD is 66. Presumably, this is for an atropal, as that's the only 66HD creature I can think of off the top of my head, so the wish would be "I wish for a scroll of simulacrum of an Atropal".

According to the DMG pricing, the material components aren't included. So implication would be that you need the material components when using the scroll, not at creation. Which means that it wouldnt need to be a scroll of a specific creature. But the xp for a scroll of 66HD is astronomical when made by a wish

[Particle_Man]

Is there a way to make an AI remember its own past ruling on rules and be consistent with its own past rulings? Or is that too complicated because there are so many different nuances to natural languages and so many different ways to ask questions about similar things?

[NichG]

Is there a way to make an AI remember its own past ruling on rules and be consistent with its own past rulings? Or is that too complicated because there are so many different nuances to natural languages and so many different ways to ask questions about similar things?

Sort of, in the sense that the input to ChatGPT is the last ~4000 tokens of your conversation so far, and it can reference that prior conversation. So as long as those past rulings fit within that buffer, it can remember them (and if you had a text summarization thing to compress that conversation down, you could fit more, but not an unlimited amount). The default attention architecture has quadratic memory and time costs with the size of the buffer though, so you can't just make that buffer bigger easily. There are versions that have linear memory/time costs, but so far they don't scale with dataset size as well as the default method, meaning that the sort of 'just train on the internet!' thing doesn't work as easily with those alternative methods (yet). In principle I would expect it to be possible to get it down to O(log(N)), but whether or not you can preserve the scaling properties while doing that is unclear to me.

Whether that leads to the AI being consistent with its past rulings is another story. Given enough data, predictions should take that information into account in a useful way, but that doesn't necessarily mean logical consistency since logic is emergent in these systems rather than existing as a fundamental characteristic. But based on what we see with the current buffer size, you would see things like the AI picking and sticking with a standardized template for its responses to related questions, re-using phrases it had previously generated, stuff like that. Which at least up to a point might raise the degree of consistency between rulings. Where it's less obvious would be things like the consequences of implicature. E.g. given two rulings that seem to make sense individually, but when combined imply opposite things so the combination doesn't make sense. Given stuff like 'lets think step by step' prompts, its likely you could get something like ChatGPT to step through the reasoning in a back and forth dialogue if you found such a conflict and asked probing questions, but it won't systematically do it on its own.

Though again, given enough data, this ends up being coded in the distribution of the data implicitly eventually, it just takes a LOT. But we're getting into amounts of data that do seem to be hitting that 'a LOT' threshold. For example, GPT-3 was trained on about 10 billion books worth of text. To put in perspective, if a person could read two books a day and lived 70 years in which they could do so, it would take 200000 people to read all of those books within their collective lifespans.

[Darg]

Is there a way to make an AI remember its own past ruling on rules and be consistent with its own past rulings? Or is that too complicated because there are so many different nuances to natural languages and so many different ways to ask questions about similar things?

It wouldn't be hard for the AI to remember exact match queries. However, the AI doesn't actually understand the language it prints. So there's no real way at the moment to ensure consistency between similar queries without direct human involvement in the training. With the astronomical amounts of data, it would be infeasible. That said, someone might be able to come up with a solution in the future.

[icefractal]

According to the DMG pricing, the material components aren't included. So implication would be that you need the material components when using the scroll, not at creation. Which means that it wouldnt need to be a scroll of a specific creature. But the xp for a scroll of 66HD is astronomical when made by a wishWhat do you mean?

Any scroll that stores a spell with a costly material component or an XP cost also carries a commensurate cost. In addition to the costs derived from the base price, you must expend the material component or pay the XP when scribing the scroll.

Activating a scroll spell requires no material components or focus. (The creator of the scroll provided these when scribing the scroll.)
A price listing in a chart might decide things when the rules are ambiguous, but it doesn't override them when they're clear.

[Crake]

According to the DMG pricing, the material components aren't included. So implication would be that you need the material components when using the scroll, not at creation. Which means that it wouldnt need to be a scroll of a specific creature. But the xp for a scroll of 66HD is astronomical when made by a wish

Page 287, creating scrolls, explicitly says you need to provide material and xp components while crafting. Since the costs are variable, the base scroll cost is whats shown in the table, just like other scrolls with variable cost components such as permanency, trap the soul, soul binding etc.

[redking]

Can the duplicate spells function of wish fire off spells at caster level 10,000? If you are saying that they can specify a caster level for a magical item because if does not say that they cannot and it can break the general rule that a spellcaster cannot create an item of greater caster level than what the spellcaster is capable, then why not a CL 10,000 spell?

[RandomPeasant]

A sufficiently powerful wizard can arbitrarily create souls. Does this make sense— is the sacrifice of 5000xp enough experience points to create a soul? Beats me.

Can you explain why it wouldn't? wish is definitely capable of e.g. restoring someone a Barghest as Fed upon to life, which would seem to involve reconstructing a destroyed soul. It's not obvious to me why creating a new one would be impossible.


Wish's disclaimer is, I feel, an explicit escape clause.

You're right, but in the wrong direction. It's an escape clause from only being able to do the enumerated things. It does not mean that the enumerated things can backfire.


You are well within your rights to refuse to answer at all, but this non-sequitur is pointless.

It's not a non-sequitur, and if you think it is you don't understand the argument. All of those are specific things that fall within the general allowances of wish. The point I am attempting to elucidate is that we don't think you need to be explicitly allowed to use wish to replicate polymorph any object, because you are generally allowed to replicate 8th level spells and it is an 8th level spell. By analogy, why would you need explicit permission to create a scroll of simulacrum at any particular caster level when you are generally allowed to create magic items, and it is not under dispute that a scroll (whatever its caster level) is, in fact, a magic item?


How are you going to wish for a 33 CL scroll of simulacrum. Sure, you can wish for a scroll of simulacrum. How do you get the 33 CL version? What are you speaking aloud? Pretend you are a player in my game. Just a straight answer instead of irrelevancies, thanks.

"I wish for a CL 33 scroll of simulacrum". I don't understand what the gotcha here is even supposed to be. Are we pretending a spell with the ability to explicitly increase your mechanical stats doesn't know about game mechanics? Are we pretending that caster level, which can be explicitly increased by an Orange Ioun Stone, is unknowable to characters within the world? What's supposed to be stopping me from just telling the spell to do the thing I want?


The idea that game constructs such as caster level don't exist in the setting is an attractive one, but the idea that characters are incapable of measuring finite, linear variables related to the power of a wizard is ridiculous. A 33rd level caster has spells that last 33 minutes, has spells that affect 33 targets, or spells that remove 33 bones. The idea that this concept cannot be conveyed in setting by speech is silly.

Yeah, I just don't see how you can maintain the notion that characters in-world don't know about CL. You can buy items that increase CL! I think you can make a case that characters in world don't, like, know the exact ins and outs of the CR system, or think that reality is divided into 5ft squares as the fundamental unit of distance. But the idea that CL is unknowable is just not tenable. It's like asserting people wouldn't figure out spell levels despite all the spells that care about them directly, or hit dice when spells at every level care about them in some respect. They are probably not exactly as knowledgeable as we are, but these things have very direct, material consequences in the world the game describes in ways that cannot easily be dismissed as mechanical abstraction.


Now the part that's more GM-dependent - do 66 HD Solars (or whatever creature) exist in the world? Simulacrum is a duplicate of a creature, not "a hypothetical creature which could potentially exist". So if there is a 66 HD Solar in the world and you have any way to specify it, then fine, you're good to go. But you can't cause one to suddenly exist if it didn't already.

This is something I think is wonky. IMO, the correct solution is to, again, re-write the rules and simply declare that simulacrum uses a piece of creature as the target rather than a component. Then you can neatly sidestep debates about questions like "what if I wish for a scroll of a simulacrum of something that might have existed, but I am not sure did" or "if I can wish for a scroll of simulacrum of something that no longer exists, what about a prior version of a creature that currently exists"?


Can the duplicate spells function of wish fire off spells at cater level 10,000? If you are saying that they can specify a caster level for a magical item because if does not say that they cannot and it can break the general rule that a spellcaster cannot create an item of greater caster level than what the spellcaster is capable, then why not a CL 10,000 spell?

This is just a bad argument. You are essentially saying "if the rules work in X way I think is dumb, that implies they also work in Y way I think is dumb". But there's no reason you've provided they can't work in Y way, you just think it's dumb for them to work that way. But you already think it's dumb for them to work in way X, so clearly "redking thinks it's dumb" is not dispositive. You're going to need to bring more to the table here.

That said, there's no reason this would be how it works. When wish duplicates a spell effect, it is still the same spell cast by the same person. Creating a CL 10k item doesn't require wish to have a caster level of 10k itself. Why would it?

[mashlagoo1982]

Looks like the RAW regarding Wish and magic items is intentionally vague.

Either side could use the text from the spell to support their opinion. That was probably not an accident.

[RandomPeasant]

I'm pretty sure it was an accident, because 3.0 wish has text that makes it very clear that "create a magic item" without qualifiers means that there are no qualifiers on what magic items you can create. Specifically, it has a cap on magic item values (just like 3.5 has for non-magic item values). If we're trying to read the tea leaves of intent (which I very much do not endorse as an approach to understanding rules), they are quite clearly that designers meant for you to be able to wish for whatever magic item you wanted, because they explicitly removed the thing that would stop you from doing that. Is that very dumb? Yes, it is very dumb. That's why I don't think trying to divine authorial intent is a very productive approach to rules debates.

[icefractal]

I'd imagine the thinking was that the increased XP cost for magic item Wishes would be a sufficient limiting factor. And if you ban any method of evading the XP cost (which would include saying that Wish Sp/Su abilities just don't work, I guess), it mostly is.

The mostly is that there are still some areas, like super-high CL scrolls, where even paying the full cost is OP, or at least creates a higher power level where people without Wish have difficulty even competing.

Wish could certainly use a rewrite though. Tangent:The XP cost shouldn't be avoidable but is also way too high for most uses. Maybe something like:
* Duplicate spells without costly components - free, the cost is using a 9th level spell
* Duplicate spells with costly components - the full XP cost, or 1/5 the gp cost as XP
* Create non-magic items - 1/5 the gp cost as XP
* Create magic items - 1/5 the gp cost as XP, plus the XP cost
* Inherent bonuses - 5000 xp as normal, but ditch the stupid "must be done all in a row"
* Remove harmful effects / injuries and afflictions - 1000 xp maybe, as its like Greater Restoration but better
* Transport travelers - 100 xp to bring willing people to a location that it would be possible to teleport to, the full 5000 xp if you're using it to summon unwilling people or break into dimension-locked places
* Undo misfortune - 1000 xp maybe? That might be too expensive, but rewinding time is potent.

Making it not avoidable is easy - Wish has no XP component, and the basic effect is merely to duplicate spells without costly components, or to make unsafe wishes. You can then choose to sacrifice additional XP to do other things. And someone else who's present can pay that XP, if the Wish is on their behalf and they agree to it.

This does mean that creates with Wish SLAs are more limited in what they can grant without going into "unsafe" territory - although for evil ones like Efreet that's maybe not a problem. Maybe something like "a creature with Wish as a natural ability can spend a virtual HD*100 xp on it each casting", or per day, or whatever.

[sreservoir]

Right, but staff of the magi is an artifact not a regular magic item, so it doesnt really translate equivocally to a scroll of simulacrum.

I'm quite aware, and pointed this out upthread even. I don't have any trouble with the idea that you wish to create a scroll of simulacrum at any caster level you can afford to pay the XP for.

I just think the most consistent way to read RAW is to:

- Assume by default that anything the text says is in fact true
- Where we have interpretation ambiguity that potentially makes something stated elsewhere untrue, prefer interpretations that aren't contradictory
- If we can't infer a consistent interpretation, then presume, as much as possible, it's a special case that applies under the conditions specified
- If text contradicts itself for the same specific example, then we accept that one instance must be in error, and preferably throw out the example less consistent with general rules

Now, I do agree this particular instance is probably an editing error from 3e, where the ability to create magical items has the same cost limit as for non-magical items. But it does not create a contradiction in the rules; we only go as far as step 2 to find at least one good reason that creating the staff of the magi is necessarily a "greater effect", even though it does fall under "Create a magic item" (as artifacts are magic items): because it has no listed price, it's not possible to pay the amount of XP needed to create it under the ordinary rules for that function. There's no need to skip straight to throwing the example out.


Let's talk about examples for a second. Specifically, let's talk about the "Sample Rage Mage" found on page 73 of Complete Warrior. This character has 6 levels of Sorcerer and 7 levels of Rage Mage. Since Rage Mage progresses casting on even levels, she should cast as 9th level Sorcerer. However, her stat block lists her as knowing two 5th level spells (cone of cold and teleport), which is consistent with her being an 11th level Sorcerer.

"The examples are just as rules as the actual rules" is simply an untenable position to adopt for the purposes of understanding the rules, because the examples are sometimes not just wrong, but incoherent. What progression is the Rage Mage "supposed" to have based on that example? It's not full, because that would get you 13th level casting. It's not "half, but on odd levels", because that would get you 10th level casting. Maybe it's "not on 4th, 7th, or 10th", which does fit the data point we have? Who knows! But if we accept that examples weigh equally to rules, we have to answer that question instead of simply discarding the sample character as someone making some unknown and unknowable mistake about which we have no obligation to care as anything more than trivia.

It's not a character you can create using the normal rules, sure, which which certainly makes this a bad sample character that doesn't contribute to our understanding of the rules. But it's still a reasonably serviceable NPC you can plop into a game, using the stat block as-is, if you want—that's not useful to us in this situation, but it's probably a more expected use-case of the stat block than inferring rules from it. In that case, it's just a character who casts as a 9th-level sorcerer but, for reasons not specified, has 6/6/6/6/6/4 spell slots. This isn't as weird as you're trying to make it sound, considering kobolds get something similar, and most monster stat-blocks don't give even that much justification; we can pretend that this particular character has done something similar that we don't have rules for or whatever. In the absence of text to the contrary, we might assume that she has CL 9, which raises some questions as to whether she can actually cast those 5th-level spells depending on how you infer the infamously muddy minimum CL rules†, but that's neatly sidestepped by Spell Rage raising her up to CL 14 anyway.

I mean sure, I strongly suspect that +4 DC during rage for schools is the 10th-level upgrade of Angry Spell, which suggests that the example character was perhaps created as a sorc 6/barb 1/rage mage 10 and then scaled down to rage mage 7, i.e. these are genuine editing errors, and I can't recommend letting them influence your understanding of the rules elsewhere. But I can't exactly recommend running a game according to strictly-interpreted RAW in the first place, and this particular example gives you plenty of room to sidestep the issues entirely by not using the stat block, since none of its errors actually imply any new information about rules elsewhere.

There's no need to discard the worked-out sample character just because you don't understand the rules behind it and suspect there are unknown and unknowable mistakes, in the same way we don't stop trying to solve protein structures just because we don't have a complete model to predict protein structure based on peptide sequences and we don't throw our hands in the air and declare the entire field of machine learning unusable because interpretable models are still a research problem.


I'd imagine the thinking was that the increased XP cost for magic item Wishes would be a sufficient limiting factor. And if you ban any method of evading the XP cost (which would include saying that Wish Sp/Su abilities just don't work, I guess), it mostly is.

The mostly is that there are still some areas, like super-high CL scrolls, where even paying the full cost is OP, or at least creates a higher power level where people without Wish have difficulty even competing.

Wish could certainly use a rewrite though. Tangent:

Wish has the problem that it does double duty as a spell you could potentially cast, and also as a fully generic plot device that produces any effect that's just slightly more than "appropriate" as an otherwise-costless daily spell. This creates an inherent inconsistency in the "power level" of effects it should be allowed to create; miracle does it little better by having an explicit distinction between the effects that don't cost XP and "request DM fiat" function. Limited Wish also exists!

With that said, it's worth noting that the problem with creating super-high CL scrolls isn't wish. The problem is first with the spell the scroll produces (simulacrum is a problem spell in an of itself, even if you're only casting it at CL 13), second with the pricing of the scroll compared to the game's assumptions about magic item availability. The market price of that CL 33 scroll of simulacrum of an atropal is just 45375 gp, which you're expected to be able to basically just pop into a metropolis and buy—and you can go up to 58 HD in any "large city", 144 HD in a "metropolis". Which, no, doesn't actually make sense, communities only have casters beyond 18th level by DM fiat, so presumably there are either epic casters out there churning out CL 16k scrolls of cantrips en masse, or crafters abusing infinite CL loops are just an assumed part of the setting, I guess.

†Bonus:


Yes, there is a complete mapping from spell levels to minimum caster levels in Dungeons and Dragons 3.5 edition. The minimum caster level for a spell is equal to the spell's level +1, so a 0-level spell would have a minimum caster level of 1, a 1st-level spell would have a minimum caster level of 2, and so on. Here is a complete list of the minimum caster levels for each spell level in D&D 3.5e:

0-level spells: minimum caster level 1
1st-level spells: minimum caster level 2
2nd-level spells: minimum caster level 3
3rd-level spells: minimum caster level 4
4th-level spells: minimum caster level 5
5th-level spells: minimum caster level 6
6th-level spells: minimum caster level 7
7th-level spells: minimum caster level 8
8th-level spells: minimum caster level 9
9th-level spells: minimum caster level 10
Keep in mind that these are the minimum caster levels for a spell. A character with a higher caster level may choose to cast the spell at their own caster level, which could potentially increase the spell's effectiveness.

[NichG]

I'm quite aware, and pointed this out upthread even. I don't have any trouble with the idea that you wish to create a scroll of simulacrum at any caster level you can afford to pay the XP for.

I just think the most consistent way to read RAW is to:

- Assume by default that anything the text says is in fact true
- Where we have interpretation ambiguity that potentially makes something stated elsewhere untrue, prefer interpretations that aren't contradictory
- If we can't infer a consistent interpretation, then presume, as much as possible, it's a special case that applies under the conditions specified
- If text contradicts itself for the same specific example, then we accept that one instance must be in error, and preferably throw out the example less consistent with general rules


Once you start having to interpret, why is RAW a relevant concept at all anymore?

[Fiery Diamond]

Once you start having to interpret, why is RAW a relevant concept at all anymore?

When was it ever? I'm not being facetious, I genuinely don't understand why RAW is given any special authoritative value by people. Sticking to RAW results in drown-healing and unable-to-recover-from-drowning. "The Rules" are only ever there to provide a consistent framework for how to play and adjudicate the game. Some parts of them are dumb. Some are broken. Some are contradictory. Having a reasonable method for parsing and interpreting (which is what was provided by the person you're responding to - I genuinely can't think of any reason a reasonable person would disagree with that approach) is almost a necessity for turning "the text of the rules" into "actionable content." RAW by itself isn't, and never was a relevant concept.

[sreservoir]

Once you start having to interpret, why is RAW a relevant concept at all anymore?

"Once you start having to interpret" is a category error.

[Darg]

Once you start having to interpret, why is RAW a relevant concept at all anymore?

The RAW says "When a wish creates or improves a magic item, you must pay twice the normal XP cost for crafting or improving the item, plus an additional 5,000 XP." Creating a magic item is already a process with rules and the rules say that you can't create artifacts. No interpretation required. There is no "normal XP cost" for crafting an artifact. Thus you cannot pay the XP component for creating one with the spell. It's when one deviates from the general rule to say that the spell is specific and contradictory that it gets messy.

[NichG]

When was it ever? I'm not being facetious, I genuinely don't understand why RAW is given any special authoritative value by people. Sticking to RAW results in drown-healing and unable-to-recover-from-drowning. "The Rules" are only ever there to provide a consistent framework for how to play and adjudicate the game. Some parts of them are dumb. Some are broken. Some are contradictory. Having a reasonable method for parsing and interpreting (which is what was provided by the person you're responding to - I genuinely can't think of any reason a reasonable person would disagree with that approach) is almost a necessity for turning "the text of the rules" into "actionable content." RAW by itself isn't, and never was a relevant concept.

I basically agree with this sentiment. However I've heard that argument that at least one can talk about RAW in the context of understanding what should be rewritten for errata/subsequent editions. E.g. 'RAW leads someone who takes these rules too literally to this bad result, so we should rephrase the rule the next time we publish it'. But for that, 'its unclear and could be interpreted multiple ways' is sort of an end point. The status of the written thing is ambiguous and if you don't want that to be the case, you know you should rewrite it for next time; if you do want it to be the case (because you want different tables to run it differently), you've got what you want so you're also done...

[RandomPeasant]

I'd imagine the thinking was that the increased XP cost for magic item Wishes would be a sufficient limiting factor.

Maybe. But if you're going to put a thing with SLA wish in the first Monster Manual, I think "we forgot to check how that works" is not a very compelling excuse.


Wish could certainly use a rewrite though.

I find this (https://dungeons.fandom.com/wiki/Tome_of_Fiends_(3.5e_Sourcebook)/Optional_Rules#No_Wishing_for_More_Wishes!) to be a pretty satisfying re-write. I think nickel-and-diming people on creating stuff is mostly pointless unless you are willing to do a global overhaul of all the wealth cheating. The uniquely problematic thing about wish is that you can ask for things that are impossible to get by any other means. The thing where you can end up with a stupid huge pile of money is a problem, but it's one that comes from so many places that I think it's more productive to attack it from the "you can break the game with a stupid huge pile of money" end.


There's no need to discard the worked-out sample character just because you don't understand the rules behind it and suspect there are unknown and unknowable mistakes, in the same way we don't stop trying to solve protein structures just because we don't have a complete model to predict protein structure based on peptide sequences and we don't throw our hands in the air and declare the entire field of machine learning unusable because interpretable models are still a research problem.

Sure. You can use the character. But the contention was never that you can't. The contention is that because the character contradicts the rules, it demonstrates that "the examples are valid rules" is not a workable framework for interpreting the rules. It's true that you can still do ML stuff without interpretable models. But you can't claim that your models are interpretable without them.


Once you start having to interpret, why is RAW a relevant concept at all anymore?

When was it ever? I'm not being facetious, I genuinely don't understand why RAW is given any special authoritative value by people.

RAW has authoritative value because it is a starting point on which we can, notionally, agree. You and I might disagree about what the rules "should" say, or what it would be good for them to say. But we can, again notionally, agree what the words on the page are and what those words mean. Drown healing is dumb, and you shouldn't use rules that work the way it works by RAW. But being able to agree that the rules do actually say that is important, because it gives us a common ground to start from in fixing the rules, rather than you assuming that the rules "should" be that drowning can only lower your HP and me assuming they "should" be that drowning is a series of status conditions that don't effect your HP at all.

The concept that isn't relevant, to my mind, is "RAI". RAW is good because it's a Schelling Point. We all have the same books and they all (modulo some edge cases) all have the same text. "Intent" is not something that is universally accessible, so it's hard to see it as a useful tool for settling arguments. The alternative framework to RAW is "what should the rules say", not "what did the designers mean in their hearts". Which is a framework that is admittedly more complicated to argue in than "RAW" or "RAW plus my mental model of the designers", but has the benefit of being able to produce answers that are significantly more useful. Whatever you think it is that "create a magic item" means, I promise you that no amount of "designer intent" mind caulk will allow you to get something as good as just sitting down and deciding what the spell should do.


The RAW says "When a wish creates or improves a magic item, you must pay twice the normal XP cost for crafting or improving the item, plus an additional 5,000 XP." Creating a magic item is already a process with rules and the rules say that you can't create artifacts. No interpretation required. There is no "normal XP cost" for crafting an artifact. Thus you cannot pay the XP component for creating one with the spell. It's when one deviates from the general rule to say that the spell is specific and contradictory that it gets messy.

Going places is also a process with rules. Does the "transport travelers" function of wish not bypass those? Do I have to spend an 8th level spell slot when wish imitates one? Of course not. Just because something has the same result does not mean it is the same process.

[NichG]

RAW has authoritative value because it is a starting point on which we can, notionally, agree. You and I might disagree about what the rules "should" say, or what it would be good for them to say. But we can, again notionally, agree what the words on the page are and what those words mean. Drown healing is dumb, and you shouldn't use rules that work the way it works by RAW. But being able to agree that the rules do actually say that is important, because it gives us a common ground to start from in fixing the rules, rather than you assuming that the rules "should" be that drowning can only lower your HP and me assuming they "should" be that drowning is a series of status conditions that don't effect your HP at all.

The concept that isn't relevant, to my mind, is "RAI". RAW is good because it's a Schelling Point. We all have the same books and they all (modulo some edge cases) all have the same text. "Intent" is not something that is universally accessible, so it's hard to see it as a useful tool for settling arguments. The alternative framework to RAW is "what should the rules say", not "what did the designers mean in their hearts". Which is a framework that is admittedly more complicated to argue in than "RAW" or "RAW plus my mental model of the designers", but has the benefit of being able to produce answers that are significantly more useful. Whatever you think it is that "create a magic item" means, I promise you that no amount of "designer intent" mind caulk will allow you to get something as good as just sitting down and deciding what the spell should do.


Settling arguments is not inherently valuable though, unless the resolution to the argument leads to action. In which case it matters that the thing used to resolve the argument actually applies the the underlying cause. So I think once you start to talk about things like patterns that would only be used to try to run RAW consistently, its off the mark and you really do need to be working with 'what should the rules say' rather than 'what do the rules say' at that point.

It's like, if someone is asking about character builds that they actually intend to play at a table RAW isn't really relevant to that discussion. So 'X is correct and Y is not, according to RAW' isn't meaningful because really it's the GM at that table who is going to say 'it works' or 'it doesn't work'. And so even if its more difficult and subjective and so on, speculating about what GMs are likely to permit is actually more useful and relevant to that discussion. Determining who was technically correct only helps if that technical correctness is a good predictor for what someone actually wants to go and do with the conclusions of the discussion.

[DrMartin]

this varies for each game system, but in dnd's case (any edition), as soon as the rules get past a *low* threshold of being obvious, RAW's main contribution to a discussion it's just comedic value.

And as for the actual topic of the thread, I think that the arguments put together by the AI can be quite valuable in seeing connections that one hadn't thought were there - with the understanding that there is no "logic" as we understand it under the text you are reading, is more of a manieristic composition that forces to ape the style and structure of a coherent sentence.

[Darg]

Going places is also a process with rules. Does the "transport travelers" function of wish not bypass those? Do I have to spend an 8th level spell slot when wish imitates one? Of course not. Just because something has the same result does not mean it is the same process.

Why would you think that teleportation is considered movement? I think the game does a good job sidestepping that confusion.

The difference between creating a magic item and duplicating a spell is that Wish tells you it's duplicating a spell, not casting it.

One problem with your interpretation is that the spell does not say that it can create a magic weapon out of nothing and we have rules for the creation of a magic item which first requires the mundane form of the item. The rules permit you to do something when they say you can do something. It's a big ask to ignore established rules when something uses the exact same name of those rules as a reference to what it's trying to convey. If it is as open ended as you imply, then the wisher could create magic items to their whims not limited by the confines of the rules and established game mechanics. All for the measly pittance of 5000XP because there is no established cost for the item creation.