If you have a problem with moderation, you know how to address it. This is not it. This is now your second warning in this thread. One more and you will be asked to leave the thread.
AI/LLMs AI art bans are going to ruin small 3rd party creators
- Thread starter ECMO3
- Start date
Maxperson
Morkus from Orkus
Between 17 and 34 million Americans still believe that! Next time you are in a large crowd, look around and realize that between 5 and 10 percent of them think the moon landing never happened.
Wikipedia could theoretically be 100% accurate. According to a study by OpenAI last fall, even when LLMs are trained on perfect data, hallucinations are mathematically inevitable, and that cannot be corrected. So there is an inherent limit to accuracy for LLM AI.
Be a little careful of that, though.
Such numbers are obtained by polling, and then how you ask the question matters, but not everyone is thoughtful about how they report on the numbers.
As and example, the questions, "Do you believe X occurred?" And "Do you believe it is possible that X occurred?" Are not really the same question, but are sometimes both lumped into the same category as believing the conspiracy theory.
Bill Zebub
“It’s probably Matt Mercer’s fault.”
Oh, well, that entirely changes the story then. I guess the analogy I made...comparing the scare tactics around accuracy that were used against Wikipedia 20 years ago to similar tactics being used against AI today...falls apart because of that theoretical difference.
But I guess the good news is that we can stop worrying about the environmental impact of AI because theoretically fusion energy production will solve that problem. Hurray for theory!
You are comparing apples and oranges here. Large-scale fusion power is still a theoretical industry. Human editing/oversight of Wikipedia is a fact and is happening today. The question of the site's accuracy is simply a matter of how many people are brought to bear on the problem. That is the theoretical aspect of what I pointed out. Assuming OpenAI's study is accurate (and they certainly do not have an anti-AI bias), LLM AIs have a built-in limit to their accuracy, which no degree of work can correct, short of a completely novel AI model.
Bill Zebub
“It’s probably Matt Mercer’s fault.”
Do you believe that Wikipedia will eventually be 100% accurate, entirely from human effort?
If your answer is 'yes' then I will happily concede the debate.
No, I do not believe it is practically achievable for Wikipedia to be 100% accurate. But if a Wikipedia page is known to be accurate as of a certain date and it has not been edited since that date, it can be relied upon. Every response from an LLM AI needs to be double-checked every time. Also, I would guess that Wikipedia is overall more accurate than LLM AIs, but that is just an estimation -- I have no numbers or studies to back that up (but would welcome seeing some if anyone knows of them). Also, I would say that most Wikipedia errors are not going to fabricate and reference entirely fictional things (aside from certain polarized topics and personalities).
Mecheon
Sacabambaspis
That's not Wikipedia's end goal. Its trying to be an encyclopedia. In the dinosaur space, that means a lot of stuff needs to be put in that we know isn't accurate, but was considered back in the day. Consider Dawndraco. We're pretty sure that's just a random indistinct Pteranodon and the person who said we should split it was wrong. So, it has a page that has to go 'yeah its maybe a speciles (The vast majority of people in the field do not think its a species, only this one guy). See also Triceratops and Torosaurus needing to include the well debunked theories about them being the same.
(though technically based on stuff done for Brontosaurus and Stygimoloch they really should get rid of the Dawndraco page at this point.)
Wikipedia isn't trying to be 100% accurate. So, no, it won't be, because that's not what it is. Its an encyclopedia that includes historically relevant stuff and provides the context for it, so you can go and look at the other sources. Wikipedia's really dang good at providing those alternative sources, because that's what it focuses on, making sure everything has a source you can go to and compare. A source you can research from.
AI on the other hand? AI doesn't provide sources and its notoriously prone to just making stuff up if you want deeper research. Accuracy is pointless in this when it has no way to actually back it up, it's just always untrustworthy as it hides where it gets everything from. It will never be useful for anyone actually doing serious research into something, especially when you actually research and find out the thing it was so sure about it either made up (as a lot of lawyers infamously discover) or was just some urban legend spread on Reddit or the like.
Warpiglet-7
Hero
If the premise of the op argument is true…how do we explain evidence to the contrary?
Kickstarter? Luck?
Not being facetious…really asking. I bought shadowdark and all the zines so far and Dragonslayer off dtrpg…
Kickstarter? Luck?
Not being facetious…really asking. I bought shadowdark and all the zines so far and Dragonslayer off dtrpg…
Similar Threads
Recent & Upcoming Releases
-
June 16 2026 -
June 16 2026 -
September 16 2026 
Arcana Unleashed(Dungeons & Dragons)
Rulebook featuring "high magic" options, including a host of new spells.
Replies (250) -
September 16 2026 -
October 1 2026 -
October 6 2026 -
January 1 2027 -
January 1 2027
