The state of Post-OSR content

Here's Chat GPT's strange answer for why it thinks there's an "R" in "South Dakota" (these LLMs are weird little creatures):

1754630523610.png
1754630672481.png
 
lol Gemini's self-loathing. Those gothy teenage years can be a bitch even for AI, apparently. Someone needs to direct it to the nearest Vampire the Masquerade group (or whatever the emo kids are playing this year)
 
So, my hypothesis for why the one blogger thought 5 was so amazing, where everyone else I see is saying "Meh," is that they have allowed 5 to use smaller data sets as examples in particular conditions. So a few people, or maybe even one person, wrote an "amazing" post with those characteristics, and the comments were about how amazing it was, so 5 could draw on that small data set to give a response. And that it is why it is better but not perfect at "counting" or "identifying States with an R in the name," because it has seen a small number of those questions being responded to, so the predictive text is more likely to spit out correct answers.

If my guess it correct, that suggests it is training itself in common human fallacies before it is able to actually reason. By this I mean, it is gaining the very human ability to confidently draw conclusions about the way things work using a sample size of one. Which would make it very vulnerable to various fallacies as well as conspiracy theories. It will be interested to see how often, and how badly, it hallucinates or gives illogical answers, although it would be tough to top the meltdown XGrok had a couple of weeks ago.

Add that to the response DP got, and it sounds like we are building machines to mimic human stupidity.
 
(these LLMs are weird little creatures):
To me, they are not; just about every strong point and weak point are a result of the technology being fundamentally a deeply nested set of random tables. Granted, it is an insane number of random tables interconnected in hundreds, if not thousands, of ways. However, just like rolling up a Traveller subsector can produce total garbage. So it not surprising that the output of even ChatGPT 5 can be garbage.


What is the utility of that? We already have a method to build actual humans. As tools go, "makes the same mistakes I make" seems kind of pointless.
So on the flip side of things, LLMs are very good at processing and using patterns. The trick to using them effectively is that you can accomplish 70 to 80% of the work with any repetitive task. And in some cases, nearly all the work. Used wisely, it allows a person to do more in less time. Yeah, it makes mistakes like people do, but so what? It's there 24/7 and doesn't have an ego about its mistakes.

The problem, of course, is that the hype is out of control, and people expect it to act like the computer on Star Trek and just do anything.

Doesn't mean there won't be some amazing things done with the technology. However, my personal prediction based on my experience developing software, is that the main benefits will be realized by coming up with ways to mix up traditional methods and using LLMs. In short a LLM will never get to the point where it can do everything well. But it can do some things well and, with the proper interface, really work out well.

For example for Tabletop Roleplaying I would like to see a LLM coupled together with software where the rules of a system like D&D or GURPS is baked in. So anytime a rule question comes up it always gives the right answer however anytime we get something "fuzzy" like coming up with adventures, encounters, or characters, the LLM is looped in.
 
To me, they are not; just about every strong point and weak point are a result of the technology being fundamentally a deeply nested set of random tables. Granted, it is an insane number of random tables interconnected in hundreds, if not thousands, of ways. However, just like rolling up a Traveller subsector can produce total garbage. So it not surprising that the output of even ChatGPT 5 can be garbage.
I think the fact that LLMs are trained on human-generated information, and humans are inherently fallible, then it would follow that LLMs would be equally as fallible. The complexity of these interconnected "tables" no doubt exacerbate the situation. My guess is that there are aren't enough measures in place for these LLMs to verify the validity of whatever information it assimilates.
 
I think the fact that LLMs are trained on human-generated information, and humans are inherently fallible, then it would follow that LLMs would be equally as fallible. The complexity of these interconnected "tables" no doubt exacerbate the situation. My guess is that there are aren't enough measures in place for these LLMs to verify the validity of whatever information it assimilates.
Yup, LLMs excel at dealing with patterns, and among other things, humans make mistakes in a predictable ways, hence another pattern that can be picked up by an LLM.
 
So on the flip side of things, LLMs are very good at processing and using patterns. The trick to using them effectively is that you can accomplish 70 to 80% of the work with any repetitive task. And in some cases, nearly all the work. Used wisely, it allows a person to do more in less time. Yeah, it makes mistakes like people do, but so what? It's there 24/7 and doesn't have an ego about its mistakes.

The problem, of course, is that the hype is out of control, and people expect it to act like the computer on Star Trek and just do anything.

Doesn't mean there won't be some amazing things done with the technology. However, my personal prediction based on my experience developing software, is that the main benefits will be realized by coming up with ways to mix up traditional methods and using LLMs. In short a LLM will never get to the point where it can do everything well. But it can do some things well and, with the proper interface, really work out well.

For example for Tabletop Roleplaying I would like to see a LLM coupled together with software where the rules of a system like D&D or GURPS is baked in. So anytime a rule question comes up it always gives the right answer however anytime we get something "fuzzy" like coming up with adventures, encounters, or characters, the LLM is looped in.
I agree with all this. My comment was aimed at what DP seemed to be saying about building in fallibility in order to mimic humans. I see no utility to that.

I am not a luddite. I work in an industry that has both repetitive and creative tasks, and has been using machine learning to assist with repetitive tasks for a long time. I see its value for assisting with repetitive tasks. I am frustrated that various applications I need to use have been building in AI in intrusive ways, for tasks that either don't require it, or for which it it will probably be able to do the job someday, but is currently not yet up to the task. Like, please don't put your wrong summaries in front of the information I actually need to see.

I am extremely skeptical of the idea that LLMs are going to spontaneously develop AGI. I do think it is possible that the interfaces are going to get sophisticated enough that a lot of people will think they are now AGI. Including a good number of the proponents of LLMs, who if we are being honest have among the worst social skills on the planet and should not be making any judgments about whether a machine is or is not thinking or behaving like a human. The efforts of one particular proponent to make his AI share his values without saying the quiet parts out loud being a particularly poiniant example of this happening, not to mention the dangers associated with it.

I feel like a lot of proponents don't really understand LLMs, even if they work on them, and also don't really understand humans, or even themselves, or very much about how the world works or any industry other than their own, and so they make announcements or design changes without having a clue about anything.

I suspect that in the long run, the value of LLMs won't be in their ability to analyze data for the average person, who will always have inssues with (a) knowing what to ask for and (b) being unable to determine if an answer is correct or not. I think the broader application would be for making UIs that can interpret and execute vague instructions, where the task does not require machine learning to execute. If I could tell a computer in plan, spoken language to "Find me that rule on drowning, I think it might be in the Rules Compendium", that would be considerably more useful to me than "write me a rule on drowning" or "adjudicate this drowning scenario".

Which is how the computers on Star Trek actually worked, the magic was in the oral UI, not in the tasks the computer was asked to do. It was "play me Mozart", not "compose a piece in the style of Mozart". It was "Show me a list of people with security clearance. Remove people who are known not to have been in the city on [date]. Which of those have written or spoken publicly against the legal reforms?" That last sentence requires some interpretation by the computer, in assessing whether words are "against" reforms, but for the most part the interpretation of information is done by the humans. This was the difference between Lieutenant Commander Data and the ship's computer. And the ship's computer was never stated to be AI, that label was limited to androids (not counting Discovery, which was a different situation).
 
I was trying to find the thread where someone - I think DP - was talking about how D&D isn't great for horror. I mention it because of James Maliszewski's latest post, which might show a way.

I'm not familiar with The Whispering Vault but I suspect what it has in common with conventional horror is that there are things that lie outside the PC's control, with potentially awful consequences. I think most attempts at horror in D&D fail because they are trying to make the PCs fear for their personal safety (or perhaps, to make players for their PCs' safety), which is a non-starter in a game where the PCs are designed to fight monsters, and fighting monsters is one of the most frequent aspects of the game.

I also don't think it is a fix to make a D&D module where PC combat ability is made irrelevant, because at that point you aren't really playing D&D anymore. But uncertainty as to who or what to fight, or where to fight them, or how to fight them, while bad things are happening that are outside their control until/unless they find a solution? Where players know that they will be allowed to fail, that the risk of failure is significant, and maybe they don't even have a good way of assessing that risk? I think that should be doable.

Classic D&D dungeon crawls, and perhaps even more OSR D&D dungeon crawls, are already a bit of a survival horror game. I think it would be possible to turn it into a sort of fetch quest, with time pressure to find or do something in the dungeon that prevents catastrophe, so the players can't afford to pixel-bitch and they have to take risks. That could be a really tense game.
 
I think most attempts at horror in D&D fail because they are trying to make the PCs fear for their personal safety (or perhaps, to make players for their PCs' safety), which is a non-starter in a game where the PCs are designed to fight monsters, and fighting monsters is one of the most frequent aspects of the game.
That's one facet of the issue, sure. I think the bigger one is that a description of something scary doesn't quite hit the way it used to. We have collectively grown jaded to tension, jumpscares, and film gore; narration just doesn't have the same verisimilitude as it did pre-internet. Trying to scare someone using D&D is like trying to scare someone by reading them a copy of Dracula or The Legend of Sleepy Hollow or whatever. These aren't the days of campfire ghost stories; it's just not enough to keep a grown adult awake at night.
 
How can you be worried about death when you have access to resurrection spells? How can you be worried about death when you know that the afterlife exists and that (based on your alignment) you know where you're going?

@DangerousPuhson is correct. We've become jaded. This is all commonplace now. Even the Cthulhu mythos doesn't inspire fear like it used to. Even then, the best of Lovecraft's works hit the hardest because it was ambiguous. He didn't fully describe the monster. If you know Cthulhu has 400 hp and AC -10 he is still powerful, but his abilities are defined.

(Or, silly enough, with PFRPH you know he is CR 40, with +43 acrobatics (!?!) and all his other skills and feats listed out. :D)

The Heretic
 
How can you be worried about death when you have access to resurrection spells? How can you be worried about death when you know that the afterlife exists and that (based on your alignment) you know where you're going?

@DangerousPuhson is correct. We've become jaded. This is all commonplace now. Even the Cthulhu mythos doesn't inspire fear like it used to. Even then, the best of Lovecraft's works hit the hardest because it was ambiguous. He didn't fully describe the monster. If you know Cthulhu has 400 hp and AC -10 he is still powerful, but his abilities are defined.

(Or, silly enough, with PFRPH you know he is CR 40, with +43 acrobatics (!?!) and all his other skills and feats listed out. :D)

The Heretic
Just pointing you over to this conversation.

I don't think we are jaded so much as we usually consume our horror in a medium that doesn't really takes its horror very seriously. It's generally campy, over the top teen slasher movies; ugly, over the top torture porn; stylized, over the top ghost stories; or the spectacle of over the top body horror alien/Lovecraftian ancient evils. You can't do jump scares in D&D, and the other stuff only makes you fake-scared, not real scared.

These movies (or maybe it's just American horror movies) rarely get you invested enough in them to actually be fearful for any of the characters. You could absolutely make a movie terrifying, but it would be so dark I don't expect North American audiences would want to see it. The same goes for D&D. This is the post I was going to make before I thought the better of it:

But also, it isn't just about character death, it's about loss. It's about pain, not physical pain that PCs shrug off, but the suffering of others who aren't so fortunate. And not some bullshit edgy a-million-is-just-a-number abstraction of people suffering, but realistic situations that make the players feel something. There aren't enough clerics to resurrect a population, and some of them might not be dead so much as gone, missing or kidnapped or altered beyond recognition. All bad enough that your players absolutely do not want it to come to that.
And then I realized, nobody wants to play in that game. It's not that you can't write a horror adventure, it's that people don't actually want to feel scared, or lost, or demoralized, or powerless. And in a D&D campaign, as opposed to a one-shot, you are often much more invested in the world than you will be in a horror movie. So maybe you could watch some genuinely creepy Japanese vehicle for a couple of hours, or a really scary novel; but who wants to feel scared, or lost, or demoralized, or powerless on a regular basis, for the length of a D&D session, over and over again?

I don't think people consume horror to be scared, actually. So the question is, outside of jump scares, what is it about horror media that you actually want to replicate? And I don't know the answer to that question.
 
That's one facet of the issue, sure. I think the bigger one is that a description of something scary doesn't quite hit the way it used to. We have collectively grown jaded to tension, jumpscares, and film gore; narration just doesn't have the same verisimilitude as it did pre-internet. Trying to scare someone using D&D is like trying to scare someone by reading them a copy of Dracula or The Legend of Sleepy Hollow or whatever. These aren't the days of campfire ghost stories; it's just not enough to keep a grown adult awake at night.

So you say you don't like trigger warnings?* Visceral descriptions ahead.

Reading this again I thought of another point. Most D&D games are pretty abstracted from reality, and people want it that way. Think about what an accurate portrayal of a typical adventuring day would look like, and compare that to how a DM typically describes it. As a starting point, how about, "I search the body". Your character is touching, in a very intimate way, a still warm, unexpectedly limp and unmistakably dead person, whose head and limbs roll in unnatural ways as you jostle it. It is lying on its belt pouch, which is starting to become soaked with already cooling urine and sticky blood. When you turn it over to get to the pouch, its guts spill out through the wound that killed it. Your hands are soaked with blood and piss and shit, which you wipe off as best you can with its cloak, so you don't foul the contents of the pouch when you rummage through them.

When a battle turns in your favour, do any of your opponents beg for your lives? Do they tell you they have a family that will miss them, and children who will become orphaned? Do they weep with terror?

Most people want their D&D to be pretty cartoony. It's pretty hard to squeeze horror from Looney Tunes. I think you could do horror, I just don't know that anyone playing D&D actually wants to do horror.

My games are pretty earnest, I want my players to feel stuff. Because of that, I'm careful about my content, which is why I don't do things like sexual violence or on-screen infant mortality. As a rule, I don't subject my players to the above descriptions. And I don't generally like horror movies, so this isn't something I would intentionally bring into my game. Maybe that means I'm not the best person to talk about horror in D&D. But I still think the real question is, what are you trying to do, really? If you tell me the reason horror doesn't work in D&D is because you can't make your players scared, I don't believe you. What is it from horror media that you actually want to bring to D&D?


*No, not you DP, I don't know your position on this one way or another, it's an abstract "you".
 
With regards to Ravenloft, is it possible that people are looking for the trappings of horror? Like it's not so much scary, but gothic and Hot Topic spoopy?
Like Dark Sun is grimdark, survivalist, but most players will find a way around the serious impediments (lack of water and metal) pretty early on, and then it's just D&D in the wasteland. The flavour remains though. Player as tourist.

Because otherwise, the best way to inspire fear is through powerlessness, and I think folks are feeling that way plenty irl these days.
Also, taking away power reeks of DM fiat, and will lead to an adversarial relationship with your players -your friends- fast.
 
I don't think we are jaded so much as we usually consume our horror in a medium that doesn't really takes its horror very seriously. It's generally campy, over the top teen slasher movies; ugly, over the top torture porn; stylized, over the top ghost stories; or the spectacle of over the top body horror alien/Lovecraftian ancient evils. You can't do jump scares in D&D, and the other stuff only makes you fake-scared, not real scared.
Straight up: D&D doesn't work for horror because we've become desensitized to horror. These days movies have to do some groundbreaking work to get you to feel fear - I'm talking paradigm-shifting, zeitgeist-changing, never-before-seen things done by entire teams of people whose sole job is to scare an audience. A singular DM isn't going to be able to accomplish this with some dice and choice descriptions.

You know what scares me? The thought of being buried alive. You know what doesn't scare me? Somebody telling me "ok, so you're buried alive now, what do you do?", in so many words, no matter the descriptive spin they put on it. It's just the wrong medium. We are past that; we are jaded. We have been shown some very visceral footage of scary things - we are past the point of mere words doing justice to any of it.
 
Straight up: D&D doesn't work for horror because we've become desensitized to horror. These days movies have to do some groundbreaking work to get you to feel fear - I'm talking paradigm-shifting, zeitgeist-changing, never-before-seen things done by entire teams of people whose sole job is to scare an audience. A singular DM isn't going to be able to accomplish this with some dice and choice descriptions.

You know what scares me? The thought of being buried alive. You know what doesn't scare me? Somebody telling me "ok, so you're buried alive now, what do you do?", in so many words, no matter the descriptive spin they put on it. It's just the wrong medium. We are past that; we are jaded. We have been shown some very visceral footage of scary things - we are past the point of mere words doing justice to any of it.
But were audiences ever really scared? A century ago, even half a century ago, people in Western culture had much greater exposure to IRL death and suffering. When the first horror movies were coming out, people were still washing their loved ones' bodies when they died, maternal and childhood death rates were still pretty high, medicines was not nearly as effective, not just at saving lives, but of mitigating the long term effects of injury. Most horror movies were pure camp in comparison. I don't think people who go to horror movies ever went to be scared.
 
Is that not the whole point? Are you watching horror movies for the riveting character development?
They go for the camp, and the spectacle. I think the not-realness is part of the appeal. People die in over-the-top, unrealistic ways, it's not like Band of Brothers where people who are shot just crumple.
 
Back
Top