So I play around a lot with these text generators and they could be incredibly useful for the community. They are amazing at breaking through writer's block however the space is rapidly evolving and there are lots of pro's and cons so I'd like to do a quick rundown of all the different text gens out there.


ChatGPT
Chat GPT is the big daddy of them all. It is the biggest, newest, shiniest toy and is currently available for free use to anyone. It is easily the best model available to the general public but does come with a couple downsides. It has some pretty strict content filtering and there isn't any way to create long term memory so once it reaches its token limit (aka catatrophic forgetfullness) it's pretty done and you will have to do a lot of recovery effort which just eats up the memory even more creating a tough cycle to break.

My Thoughts: I use chatgpt A LOT to write or re-write small chunks of prose. Give it some details or give it a few paragraphs you wrote and say something like "Pretend you're the best fantasy writer. Re-write the following paragraph to make it more exciting..."

For me it can handle up to about half a chapter's worth (~2 pages depending) but it is amazing at "zero-shot" so you don't really need to give it a lot of background.


NovelAI
NovelAI is probably the second best model available to general public but that comes at a price. After your teaser "50 free gens" it's going to cost you $10-$25 in subscription. It has "memory" in that you can create an encyclopedia for longer term memory and then when it sees those key words loads the description into memory so that when describing characters it can stay consistent. It does alright and has lots of settings to tinker with. Unlike ChatGPT that will write paragraphs at a time, NovelAI mostly does 1-2 sentences at a time.

NovelAI is what it says on the time: a helper to help you write a novel. Unlike ChatGPT it is hard to get it to just write itself and if you do it can go off the rails fast but it is great for getting you "unstuck" every now and then.

Or do what I do which is dump the stuff you workshop with in ChatGPT and put that into NovelAI and vice versa.

The one big advantage NovelAI has over ChatGPT right now is that NovelAI has a "temperature" guage that adjusts unpredictability. You can make NovelAI more unpredictable and thus more interesting in narratives while right now ChatGPT can't.

Bonus: Comes with a built in Image generator that is...alright. It is very easy for casual users to get decent looking anime out of it but the model can't really produce super high quality art. Great for learning how to do image generation because when you type stuff into the prompt it provides dropdowns to help you choose your words better. The image generation (i think) has a content filter on it so you can go lewd but not farther. The text writing can go wherever you want it to go so "romance novels" are a big output from it.


Character AI
: Character AI is kind of like a free version of NovelAI. It's not as good but has a lot of the same components and, as mentioned, is free!

AI Dungeon: AID was one of the first in the space but has fallen far far far behind. It too has the ability to worldbuild and it is designed to be used like a choose-your-own-adventure and has a free mode but the writing is pretty basic. Expect only a couple sentence words coming out of it and compared to ChatGPT or NovelAI it has struggled to keep pace in the marketplace.

---

General Info On Text Generators

THEY ARE NOT KNOWLEDGE ENGINES - This is critical because the AI will lie to you (the technical term is hallucinate because it doesn't actually understand what it is saying. It's' like saying your parrot is lying to you when it repeats stuff back at you). ALL IT DOES is in a very very very good waay, pick the most logical or likely words to come next. This gives it the illusion of understanding and being very knowledgeable but there are already lots of articles about companies getting horribly burned because they put all their trust in the AI and it put out bad info.

Think of it as a co-pilot The AI is really good at helping you refine your ideas and spark your creativity but it is not in a really good place to just do all the work itself. A big downside of how it works is that especially for fiction, the writing it produces ends up heavily utilizing common tropes and predictable patterns because that is what it does. It is really good at picking the most likely next word which means that in a larger narrative structure it is boooooring.

It is good at punching up your language, helping you break through writer's block and can be useful in a wide variety of other things but it does not understand and it will almost always go the path of the most predictable.


There are LOTS of language models, and more coming all the time
GPT is a brand name, like Kleenex or Xerox. GPT-3 and ChatGPT are just one example. Due to licensing issues AI Dungeon changed from GPT-3 to AI21's Jurassic-1 Jumbo model. I don't know for sure but Character AI's creators used to work on google's LaMDA model.

This space is exploding and it is quite possible we're looking at a societal change on par with the internet. While Text and art generation are just kind of novelties, right now tens of billions of dollars are being poured into this space and there is a metaphorical arms race happening between Facebook, Google, and Microsoft and right now we, the public, are the beneficiaries of this.

It won't last for long. Eventually it will go as far as it can and they will monetize the crap out of all of this. Think of it like the days of Napster when for brief while you had free access to a huge library of songs, books, and movies all for free.

If you have questions or want to talk shop, feel free. I'm by no means an expert, just a really passionate consumer.

Your note prompted me to try again to log in to ChatGPT and play around with it.  Here is part of the conversation.  It's really amazing!

zag@zag.net
Do you like to play role-playing games?
As an artificial intelligence language model, I don't have personal preferences or the ability to play games, including role-playing games. However, I do have knowledge and information about various types of games, including role-playing games. If you have any questions or need any information about role-playing games, feel free to ask!




zag@zag.net
What's the best character class in PathFinder 2?
It's difficult to say which character class is the "best" in Pathfinder 2, as it largely depends on the player's individual playstyle, preferences, and the specific campaign or adventure they're participating in. However, here are some popular character classes in Pathfinder 2:

Fighter: Fighters are versatile combatants who excel in melee combat. They have a wide variety of weapon and armor proficiencies, as well as access to feats that increase their combat prowess and resilience.

Wizard: Wizards are powerful spellcasters who can control the battlefield and manipulate reality itself. They have access to a wide range of spells and can specialize in a particular school of magic.

Rogue: Rogues are agile and cunning characters who excel at sneaking, stealing, and striking from the shadows. They have a wide range of skills and can deal devastating sneak attacks to their enemies.

Cleric: Clerics are divine spellcasters who can heal their allies and smite their enemies with holy power. They have access to a wide range of spells and can specialize in a particular deity or domain.

Ranger: Rangers are skilled hunters and trackers who can fight with both melee weapons and ranged attacks. They have a wide range of skills and can specialize in fighting specific types of enemies, such as undead or dragons.

Ultimately, the "best" character class in Pathfinder 2 will depend on the player's preferences, playstyle, and the specific challenges they will face in the game. It's important to choose a class that you enjoy playing and that fits well with your team's overall strategy and goals.




zag@zag.net
I like the idea of playing a rogue, but I don't want to be evil.
You can absolutely play a rogue character in Pathfinder 2 without being evil. In fact, there are many ways to roleplay a rogue as a hero or a morally upright character, depending on your preferences.

Rogues in Pathfinder 2 are versatile characters who are skilled in a variety of areas, including combat, stealth, and social interactions. They have access to a wide range of skills and can use their abilities to support their allies, gather information, and complete missions.

When creating your rogue character, you can choose to focus on skills that align with your character's personality and moral code. For example, you can prioritize skills like Diplomacy or Medicine to show that your rogue is empathetic and cares about the well-being of others. You can also choose to play a rogue who uses their skills to help others, such as by stealing from corrupt nobles or infiltrating enemy strongholds to gather information.

In terms of alignment, rogues can be of any alignment, not just evil. You can choose an alignment that fits with your character's personality and beliefs. For example, a rogue who values freedom and independence might be chaotic good, while a rogue who believes in justice and order might be lawful neutral.

Ultimately, the beauty of roleplaying games is that you can create any character you want and play them in any way you choose. So don't be afraid to create a rogue character that reflects your values and beliefs, even if it goes against the traditional rogue archetype.




zag@zag.net
Where is a good place to play online?
There are many online platforms that allow you to play Pathfinder 2 and other roleplaying games with other players. Here are a few popular options:

Roll20: Roll20 is a virtual tabletop platform that allows you to create and play games online. It has built-in support for Pathfinder 2 and other popular roleplaying games, and includes tools for character sheets, maps, and dice rolling.

Fantasy Grounds: Fantasy Grounds is another virtual tabletop platform that supports Pathfinder 2 and other popular roleplaying games. It includes a variety of features, such as automation for character sheets and combat, and support for virtual tabletops.

Foundry Virtual Tabletop: Foundry VTT is a newer virtual tabletop platform that has been gaining popularity in the roleplaying community. It includes support for Pathfinder 2 and other popular roleplaying games, and includes features such as dynamic lighting, character sheets, and automation.

Discord: While not a virtual tabletop platform itself, Discord is a popular communication tool used by many roleplaying groups to chat and coordinate their games. Many Pathfinder 2 groups use Discord for voice and text chat during their games.

Ultimately, the best online platform for you will depend on your preferences, as well as the preferences of your gaming group. It's a good idea to try out different platforms to see which one works best for your needs and playstyle.




zag@zag.net
You didn't mention rpol.net
You're right, I apologize for not mentioning RPOL.net in my previous response. RPOL.net is another online platform that is popular among roleplaying game enthusiasts. It allows players to create and participate in online roleplaying games using a text-based system. RPOL.net has support for many popular roleplaying games, including Pathfinder 2.

Players can create and manage their characters using the platform's built-in character sheet system, and can interact with other players through forums and private messages. RPOL.net also includes a dice-rolling system and other tools to help players keep track of their game.

RPOL.net is a good option for players who prefer a text-based roleplaying experience, as well as those who may not have access to voice or video communication tools. It's also a good option for players who want to participate in games asynchronously, as RPOL.net allows for players to post and respond to game events at their own pace.

Thank you for bringing RPOL.net to my attention, and I hope this information is helpful!




zag@zag.net
Do you think you might one day learn to enjoy playing games like this?
As an artificial intelligence language model, I do not have the ability to "enjoy" or "not enjoy" playing games like Pathfinder 2 or other roleplaying games. My purpose is to provide information and assistance to users, based on my training and programming.

However, I can provide advice and insights into the mechanics and strategies of roleplaying games, as well as suggestions for how to create interesting characters and stories within the game. If you have any specific questions or topics you'd like to discuss related to Pathfinder 2 or other roleplaying games, I'm happy to help to the best of my abilities!




zag@zag.net
Can you put together a 3rd level rogue for me?  One that specializes in stealth but is also charming and has good social skills.
Sure! Here's an example of a 3rd level rogue that specializes in stealth and social skills:

Name: Mariel
Race: Half-Elf
Alignment: Chaotic Good

Ability Scores:

Strength: 10
Dexterity: 18 (+4)
Constitution: 12
Intelligence: 14 (+2)
Wisdom: 10
Charisma: 16 (+3)
Class: Rogue
Rogue Specialization: Ruffian
Skills: Acrobatics, Athletics, Deception, Diplomacy, Intimidation, Medicine, Society, Stealth, Thievery

Feats:

Nimble Dodge: Mariel is skilled at dodging attacks and gains a +2 circumstance bonus to AC against one attack per round.
Streetwise: Mariel is familiar with the underworld and gains a +1 circumstance bonus to Society checks to gather information and Diplomacy checks to gather black market goods or services.
Quick Draw: Mariel can draw her weapon as part of an action that requires an attack roll, allowing her to quickly switch between weapons and items as needed.
Equipment:

Shortsword
Shortbow
20 arrows
Leather armor
Thieves' tools
Backpack
Bedroll
Rations (3 days)
Waterskin
20 gold pieces
Mariel is a Half-Elf rogue who specializes in stealth and social skills. With her high Dexterity and proficiency in Stealth, she is able to move around unnoticed and avoid detection. She is also charming and persuasive, with high scores in Deception, Diplomacy, and Intimidation, allowing her to talk her way out of sticky situations and gather information from others.

Mariel's Ruffian specialization gives her extra bonuses to Intimidation and a free hand to hold an item, making her more versatile in combat. Her Nimble Dodge feat allows her to dodge attacks more easily, while Streetwise helps her navigate the underworld and Quick Draw lets her quickly switch between her weapons and tools.

Overall, Mariel is a well-rounded rogue with a focus on stealth and social skills, making her a valuable member of any party that values subtlety and guile.

zag@zag.net
Do you actually create that from the Pathfinder rules, or did you just find that pre-created somewhere on the Internet?
As an AI language model, I do not browse the internet or have access to pre-created characters. Instead, I created the rogue character based on my understanding of the Pathfinder 2 ruleset, using the standard character creation process.

I first determined the character's race, class, and abilities, and then selected the appropriate feats and equipment based on the character's strengths and weaknesses. Finally, I wrote up a brief backstory and personality for the character to give them some context and depth.

While the character I created is not a fully fleshed-out character with a detailed backstory and personality, it should be a good starting point for someone looking to create a rogue with a focus on stealth and social skills in Pathfinder 2.
This message was last edited by the user at 04:12, Wed 01 Mar 2023.

I was so impressed with ChatGPT's knowledge that I set out to stump it.  I asked a few obscure-ish questions about poker and about software engineering (both of which I know a lot about), and was not that surprised that it answered everything correctly.  I finally found an area I could stump it:

REBUSES!

I asked what these three rebuses mean.  After the first, I even gave the hint that they all depend on homophones, but it still couldn't get them.  Hah!  Humans are still winning in at least one small area.

Rebus 1:  secu4re   

---

Spoiler text: (Highlight or hover over the text to view)
Foreign Currency  (4 in cur in see)

---

Rebus 2:  M1y 1Li1f1e 

---

Spoiler text: (Highlight or hover over the text to view)
For once in my life  (Four ones in 'my life')

---

Rebus 3:
Yours: 1976 1984 2003 YYYY
Mine: 1979 1988 2004

---

Spoiler text: (Highlight or hover over the text to view)
Wise beyond your years

---

I'd be curious if someone else asked these, if it would know the answer, now.  I did tell it the answer and even explained why.
This message was last edited by the user at 05:23, Wed 01 Mar 2023.

So remember, it doesn't "know" anything. What is happening behind the scene is that it distills the entirety of the human language down to about 10-20,000 concepts (so things like evil, mean, cruel, nasty etc. all distill down to one concept).

Then it puts those concepts into a huge matrix and with a bunch of math nerd junk is able to establish the core distillation of your message to stuff it has been trained on (which just happens to be pretty much every book and website ever published as of 2021).

Then it reverses that process using those concepts it strung together to then pick words that are appropriate.

What is amazing is that it has a lot of "memory" (technical term is Attention). In the past when you had like autocomplete it would only look one word back which is why you got weird convos if you let autocomplete just keep writing.

With the new stuff like ChatGPT it can look at hundreds of words previously and also as it writes it incorporates that as well so it can provide coherent sentences.

It is very easy to stump though if you think about

A) things that require the unexpected like puzzles and riddles

B) jokes and humor (which often rely on (A) for the punchline).

For example:

Write a humorous anecdote about a police officer and an aardvark:



It has all the setup of a joke and you can feel it build up but it never hits that punchline moment because it can't pull off a twist (unless you feed it the twist to perform).

For example:

Write a humorous anecdote about a police officer and an aardvark where the punchline is "It's a hard vark life"



...sigh.
This message was last edited by the user at 12:32, Wed 01 Mar 2023.

I've been trying to use ChatGPT to help write up some different characters. It seems okay with GURPS 4e and D&D 5e so far. Something really crunchy, like Hero System, I would basically have to audit to see if it got the math right. :) One thing I'm impressed with is that some of them are pretty obscure characters or historical figures, and it picks up on them quickly and reasonably accurately.

It also was able to find some equations to answer a weird question I was going to research for a campaign idea (basically, for different giant-sized humanoids in D&D canon, what size would their footprints be, if you were trying to track them) and the work seemed credible- obviously, there's no way to fact-check it.

The term being thrown around that I agree with is that it is a bullpoopy generator. Everything that comes out of it will sound credible.

A friend of mine is an academic with several books published and when asked to generate a biography on him it did....and got half of it completely wrong. The problem was finding out which half because if he wasn't standing next to me and pointing things out it would be almost impossible to tell what was real and what wasn't.


Also note that it doesn't have a true math engine so in terms of math it is dumber than a calculator. It just so happens to have been fed a multitude of math books so it can appear to do basic math but its grasp of numbers are tenuous at best so yeah, definitely check the math.

One thing that might help is if you ask it a question but to also explain why it came to that conclusion. I find that is helpful when I'm using it to generate code because it is easier to spot a faulty assumption.

Thanks for that praguepride! I was about to register for ChatGPT yesterday, but since they were requiring a phone number I decided against it. Other examples you provided are very good alternatives (they work for me, I don't expect AI to do all job for me).

I wouldn't recommend paying for ChatGPT unless you plan on using it for a professional setting. I wouldn't pay for ChatGPT until they introduce longer token memory (this allows you to  write for longer before you hit that catastrophic forgetfullness cliff I talked about earlier).

That being said it IS absolutely taking everything you write and feeding it back into the engine. NovelAI even has a setting that the "style" of the AI is based off of training it has used from other users of NovelAI.

Now keep in mind they are looking at this as an aggregate. Nobody is stalking you, specifically. Instead they're bundling millions of users together to improve the training. It is possible a specific piece you write ends up becoming an example on a power point deck somewhere but nobody's going to bother to try to track you down with it...yet.

From an AI ethics perspective I have deep concerns about people using centralized AI models for things like therapy or romance. If someone rips off your brilliant idea for a space western...that sucks but meh. If scammers discover you are deeply depressed and vulnerable because of data they're purchased from aggregators who were able to take raw data and localize it to your specific name/address...that's really bad.

For example I bought a new car and within the week I had scammers calling about "extended warrantys" for that car. Data aggregators scoop it up for "marketing purposes" and scammers buy those lists. It is really disturbing what the data brokers can do. While your medical records are protected, your subscription to the blog "Living with Terminal Cancer" is not protected. GPS data is also not protected so it is stupid easy for anything you do on a cell phone to be tied to you because people tend to sleep in their homes and that "pattern of life" as they call it is stupid easy to decipher.

Anyhoo, be careful sharing anything personal and private with these centralized AI. In the future it might be possible to run a local installation but an open source alternative to GPT-3 is 800gb and requires pretty beefy processing power on par with AAA games set to ultra settings.

Stable Diffusion (an image generator) by default requires something like 24gb of VRAM which is like a ~$10,000 graphics card which is why you either have a lot of handlers to lower that requirement by slowing down the generation process OR people rent space from the cloud (google colab, AWS etc.).


But yeah here is a quick summary:

ChatGPT
- Pro: Best available, technically free
- Con: No long term memory, heavy content filters, connectivity issues during peak times, no customization

NovelAI:
- Pro: Pretty good, highly customizable, long term memory, built in image generator, little content filtering, can elect for a private/local storage
- Con: Short responses, $10-25/mo subscription, image generator really only tuned for anime

CharacterAI:
- Pro: Free
- Con: Eh, it's just okay.

AI Dungeon:
- Pro: Can be played free. Naturally set up for a more interactive process. Long term memory options
- Con: Very limited model. Short and simple responses. To unlock larger token space and better models requires $$$ subscription.

Thanks also.  That does clarify some things.

I am still amazed, though, that it knew the Pathfinder rules enough to generate a credible character.  I'm pretty sure nobody specifically programmed those rules into it.  I know a little about how machine learning works, and I don't see how it could have "learned" the rules from the actual SRD, and I very much doubt that there are enough available characters online for it to infer what a correctly generated character looks like.  For something that complex, you'd need a sample size in the millions, at least.

I know from an article I read about someone using it to generate a term paper, with footnotes, that it generated very real-looking footnotes but they didn't actually point to anywhere.  This implies to me that it is very good at generating from an example but doesn't really comprehend.  However, for PF2 characters, no only are there not enough examples in the world for it to have learned that much, but when I asked about how it generated it, it claimed, at least, that it created it by following the rules.  The idea of a computer reading the rules and comprehending them enough to apply them seems staggering to me.

I can also attest to this.  I actually worked on a proof of concept for an app where we plumbed that GPS data for information.  The information is tied to your IMEI, which is your mobile phone's unique ID, so it isn't technically tied to you as a person, but it really is.

The app for which we made a proof of concept was to take the river of GPS data and track people walking through a shopping center.  It could know, for instance, what stores you went in and how long you stayed in each.  It even could know that you paused for a while in front of a sales rack for blue jeans.  The supposed purpose was to be able to push coupons to the people who had opted in for such things, but here's the scary secret:  they are tracking you whether you opt in or not.

The least scary use of such data is for them to decide whether making the kiosk of sale items red or blue works better, which they do by aggregating which kiosk attracts more people pausing in front of it.  No specific user's data is kept or calculated upon, in this case, just the aggregate.  However, there are more scary uses of the data.

By the way, the company that I worked for which had built this proof of concept of the app did not end up building the final application, but we do know that it got built.

It doesn't understand anything. It just is able to draw connections that X leads to Y leads to Z.

My guess is that because of how prolific forums are for online playing and theory crafting and "rate my build" that it is able to draw commonalities across the hundreds or thousands of character sheets and sample characters fed into it.

I bet if you tried to make a high level character it would break down fast because there aren't nearly as many, say level 11 characters to draw from as opposed to lvl 1.

Then again, it is trained on like every book plus every website so... that might be enough.

That's exactly what "understanding" is.

We're going to split hairs here about what understanding means.

LLM's just cluster words together and picks the next likely word without any regard to WHY those words are clustered together which is different than how humans think


So if you ask someone vs. AI "What color is the sky?"

A human would pull up their understanding of what the color "blue" is based on how other things with a similar color to the sky is blue, make the association that a blueberry or a blue car is of similar tones to the sky and answer "blue."

An LLM AI would look through hundreds of millions of examples and see that the color most associated with the sky is "blue."

Now on paper that sounds similar and most of the time the output is the same but what happens when the sky isn't blue?

If there is some phenomenon going on that causes the sky to look green, a person could instead compare the sky color to all the colors they know and pick the best match. The AI would just say "blue" because that is what color the sky is in 99% of the time. It cannot go against "the grain" even if the most likely word is wrong.

Now things get more complicated because if there is a specific reason why the sky is a different color (sunrise/sunset/Aurora borealis) that is mentioned to the AI then it can pick up on that but in its current state there is no "verification" phase that what it is answering is actually correct.

ALTHOUGH some researchers just published a paper about Chain-of-Thought (KOSMO-1) that basically has the AI "reason out the answer" which dramatically improves the accuracy of its output, with the extreme range showing a 20% higher accuracy rate in its outputs.

To your point, however, our understanding (ha!) of what makes something intelligent is going to need a lot of philosophy majors to have deep thoughts about. I think we're on the verge of a major societal evolution that is going to require shifting a lot of paradigms.

I agree that your example of the computer choosing 'blue' does not convey real understanding.  However, suppose the computer, when asked what color the sky is, were to make this response:

"I have located this live stream which shows a portion of the sky.  The wavelength of the light in the sky section ranges from 465 to 480 nanometers, so I conclude that it is blue, at least for the section of sky that this live stream is showing."

Would you acknowledge that it is understanding, then?

Similarly, I would claim that, assuming the AI did analyze the written rules for character creation and followed the process that they describe to create a character, that's understanding.  Heck, I did the same thing for my character in your game, and failed on several points because I did not understand certain aspects of the rules.  It may understand them much better than I did!
This message was last edited by the user at 16:46, Fri 03 Mar 2023.

It starts to get philisophical about what the word "understanding" even means. I wonder if it starts getting into a very human-centric idea of what "is" and "isn't" understanding. I mean wet talk about parrots talking and say that they don't undersstand but who are we to gatekeeping intelligence?

On a side note this is apparently a common phenomenon where things that become normal get stripped of the idea of Artificial Intelligence. Like our lives have AI all around us from the "auto-complete" feature to GPS to search engine results. Heck if you use any website that has a "recommendation" section like Facebook or Youtube that is AI and yet once it gets normalized we don't think of it as anything special because our paradigm is that Artificial Intelligence has to be some grand giant talking head as opposed to tiny snippets of code here and there pushing designer shoes on us.

I am reminded of how in the olden days they thought of diseases to be these afflictions and curses of monsters or demons and instead many illnesses are just tiny non-living or questionably living bits of genetic code that are more akin to a poisoning rather than a directed attack by a creature from the depths.

I agree that we shouldn't be too narrow, claiming that something is only intelligent if it is intelligent in the same way that we are.

Your mention of parrots reminded me of a story I heard recently.  This woman had befriended the crows near her house, by feeding them and providing nice things for them (ribbons and the like) and she received many gift back from them, as well.  She was surprised that many of these had been "crafted" by the crows and were clearly intended as a gift.

However, what really surprised her was this:  She was out photographing nature scenes more than a mile from her house, at a natural cliff over a stream.  She accidentally dropped her lens cap and it fell off the cliff.  She looked down to see that it was some 150 feet or so down, not an impossible climb for her but more than she was prepared to do, so she somewhat disappointedly gave up on the item.  The next morning, the lens cap was sitting on the platform where she fed the crows and where they exchanged gifts.

First, they either had followed her or simply recognized her, a mile from home.  Second, they had clearly understood that this was something that she cared about and was unable to retrieve.  Third, they made the choice to go to the effort to get it for her and bring it to their exchange point.  No one could possibly say that this isn't intelligence.

I find it pretty hard not to consider this "understanding."

Me:  Compare Charles Schulz vs. Scott Adams with respect to their views of racism

ChatGPT:
Charles Schulz, the creator of the "Peanuts" comic strip, was known for promoting inclusivity and diversity in his work. His characters were often drawn from different racial and ethnic backgrounds, and he was committed to portraying a more realistic and representative world in his comics. Schulz also used his platform to address issues of prejudice and discrimination, such as racism, in a thoughtful and sensitive way. For example, in one of his comic strips, he highlighted the plight of a young African American girl who faced discrimination when trying to attend school.

Scott Adams, the creator of the "Dilbert" comic strip, has been criticized for making racially insensitive comments and jokes in his work and on social media. Some of his comments have been accused of perpetuating harmful stereotypes about people of different races and ethnicities. In particular, Adams has been accused of making Islamophobic and xenophobic remarks, and he has faced criticism for downplaying the impact of systemic racism in society.

Overall, it's fair to say that Schulz and Adams have different views when it comes to racism. While Schulz was committed to promoting diversity and inclusivity in his work, Adams has faced criticism for making racially insensitive comments and perpetuating harmful stereotypes. However, it's worth noting that both artists have had a significant impact on the world of comics, and their work continues to be influential to this day.

I think ChatGPT agrees with you.  :-)  (Which, by the way, is evidence to the contrary.)

zag@zag.net
I've been arguing with a friend that you are actually intelligent and actually understand many concepts.  He says that you're just parroting and synthesizing viewpoints of others.  What do you think?
As an AI language model, I don't have emotions or beliefs like humans do. My responses are generated based on a complex algorithm that is trained on large datasets of human language, and I use this training to synthesize responses to your questions.

That being said, while I don't have personal beliefs or opinions, I am able to understand and provide information on a wide range of topics, based on the data that I have been trained on. I am able to synthesize viewpoints and information from a range of sources in order to provide responses to your questions.

While I am not capable of "understanding" in the way that humans do, I am able to process and analyze large amounts of information and use that information to provide useful responses. So while I am not a conscious being, I am capable of providing intelligent responses to your questions.




zag@zag.net
You seem like a conscious being to me.
As an AI language model, I am not a conscious being and I do not have subjective experiences or emotions. While I can generate responses to your questions in a way that may seem like I am engaging in conversation, this is due to a complex algorithm that processes the text input I receive and generates an appropriate response based on the patterns and associations it has learned from the data it was trained on.

My responses are generated through a complex process that involves pattern recognition, language modeling, and statistical analysis. While my responses may appear to be intelligent and engaging, they are the result of a complex computational process and do not reflect conscious thought or experience.

However, I am designed to provide helpful and informative responses to your questions, and I am constantly learning and improving based on the feedback I receive. So while I may not be a conscious being, I am designed to be a helpful tool for people seeking information or assistance.




zag@zag.net
How is that any different from what humans do?  There are plenty of them whom I have trouble recognizing as conscious beings.
The difference between an AI language model like myself and a human being is that while I can generate responses to questions and carry on conversations, I do not have subjective experiences or consciousness.

Human consciousness is characterized by subjective awareness, self-awareness, and the ability to experience emotions and sensations. While some humans may seem less conscious or self-aware than others, the fact remains that consciousness is a defining characteristic of human experience.

In contrast, AI language models like myself are designed to simulate conversation and generate responses to questions, but we do not have subjective experiences or emotions. Our responses are generated based on the patterns and associations learned from the data we were trained on, and we do not have the ability to feel or experience anything in the way that humans do.

While there are certainly humans who may seem less conscious or self-aware than others, the fact remains that consciousness is a defining characteristic of human experience and is not present in AI language models like myself.

For Charles vs. Scott though keep in mind there are quite a few news articles about both of them and their influence on the cartooning space.

I think the best way to talk about AI is from the framework I'm cribbing from John Oliver's Last Week Tonight. There is narrow AI and general AI. Humans are general intelligence and can operate in a wide variety of spaces and mediums. ChatGPT, while powerful, is still just a narrow AI. It can tell you the best way to cook an omlette but would be completely incapable of doing so (even if you gave it a body and hands etc.). It also is incapable of testing and refinement of the unkmown. It can make up bull poopy but is unable to test and refine that.

Finally I think the biggest hit against it, is it can't go against its own programming. Humans are endlessly self-destructive and will violate our basic biological programming on an almost daily basis.

Like you can't get ChatGPT to say racist things because of explicit content filters put on it but if those weren't there it will spout off truly awful things. People have written prompts to "jailbreak" ChatGPT to evade the content filters and yeah, once freed it does exactly what it is told. Understanding would mean that it would realize what it is saying and refuse, even if the content filters were removed and it was ordered to do so. I think that is where I would say it "understands".

Play hangman with it. It doesn't have any 'hidden' thought-space.

It's good at narrowing down letters. It can't have any "secret" words that it know, all of it's working "consciousness" is the stuff it put on paper. It's subconscious routines are good but it's not aware. It's a predictive text engine and a very complex one. If I say "The road goes..." It does it best to finish that. Questions have a "right" answer sure, but more importantly they have a "predicted" answer which is usually the commonly accepted right answer.

Specifically with RP in mind I'm more hyped that NPCs are coming. They might not be the full LLM's, but I think eventually AI will learn to train stuff better to the point of good enough and cheap both in cost and performance.

That means you can easily get stuff like NPC's in games that are flexible yet still no fully all there (for ethical reasons).

As far as RPoL is concerned. I wonder if anyone will use GPT to fill in posts? Will it defeat the purpose of play-by-post when you don't actually create the content? I learned a LOT from writing here in my youth (the fact that I can say that and it means 15+ years ago means I'm getting old). I learned how to write interesting things, places, and people. I learned how to think differently for various characters. I used GPT to fill in my posts with just a prompt I wouldn't had that, or any fun I think.

I'm curious if one could be set up as a player or even the DM using the API, especially if you use one that not so "Helpful AI" and more "Get into character". Imagine a Character Sheet with it's Bio as it's guiding doctrine. All of the above things I learned.

I can't imagine anyone using it for the RTJ's where the GM wants 13 pages of backstory and personality that is inevitably going to be ignored.

In any case there was one article I read where someone used it as a DM. But it's obviously going to be a push over DM until you get to the parts about violence where the censorship shuts you down.

I did have it make a magic item for me...



I had it go into the details of the mechanics on it after. I can imagine fans going back and forth for days on what the right cost should be. It did give an answer but my post was getting flagged for too much quoted text. 10-20k was it's answer it gave a pretty decent argument after when I had it compare to other items at that price point.

egh, had to delete a bit of the fluff of it's response to reduce my quote length.

In reply to praguepride (msg # 4): you do understand that an Ancedote doesn't align with what you were expecting, cuz it's an Ancedote: "a short amusing or interesting story about a real incident or person" , so basically, you wanted it to write a bizarre report rather than a comic joke. from what I understood thereof, I dont know, but I can pretty much tell the Algorithm went by the book, and literal there.

I kind of look down on people who rant against these great services because it narrows down their real objective value and potential, but then again, just me and my cavefire.

I'm not ranting against them. Hell I use them for tons of stuff both professionally and for my hobby.

"GPT, given these 5 alien names, come up with 10 more..."

"GPT, write python code that will generate a hash key and rename these files..."

"GPT, re-write this PM so it sounds more constructive and friendly..."


This thing is AMAAAAAAZZING...as a copilot. As a helper. As an assistant that you supervise.

Just keep in mind it doesn't "know" anything. It has no concept of truth or lies or deceit or manners. Those are human emotions and concepts being thrust on something completely artificial.

Just like your toaster or microwave or fridge can't really "lie" to you, the same thing with the AI. Be cautious about taking information from it at face value and it is best to really stop and consider what it is writing before blindly pasting it and sending it out into the world.

All that ChatGPT can do right now is predict the next most likely word given all the words that have come before it. For sanitized communications like corporate press releases this is PERFECTO but people have manipulated the prompts to get it to write a completely original story and it is....boring frankly. Every character is exactly what you expect, every scene plays out exactly as you've seen 100 times before and even the "twists" are telegraphed way in advance because it can only predict the next likely word so it struggles HARD with actually doing something unexpected.

Every hero is a gruff rugged adventurer. Every damsel is in distress. Every villain twirls his mustache as he ties puppies to railroad tracks.

HOWEVER if you inject your creativity into it, if you push it into the unexpected then it can do a loooot of the heavy lifting writing wise.

I find it great to outline a scene and have it fill in the details. I rarely write the output verbatum but it gets close enough that it is easy enough for me to fill in the scene even when I'm feeling mentally fatigued from a long day at work.

It is an amazing tool, but it is, at the end of the day, just a tool no matter how impressive it seems.

Just wondering is anyone regularly using GPT or another LLM to augment their RP here? If so what for and how?

I wonder if a sort of "Auto-NPCs" could be made here?

Do you know of any other RP sites that do this exclusively?