@Dullahan@Cocoa@Zergling_man Agreed. TBF I virtually only game online once in a blue moon. So a few more tests to see if maybe some tweaking is in order, and see if I can save myself the cost I think is worth the potential turnoff I might trigger in other people.
@realman543@Cocoa@Dullahan Anyway, this sort of scenario is also why I built @Rance. It is a game that can be played on any ping. When complete, it will be a game for like 8-10 players. I have no motivation to complete it because nobody has any interest in what it is currently.
@realman543@Rance@mint@Cocoa@Dullahan@NEETzsche I'd be amazed if that doesn't already exist. One of the projects I have is to eventually extract a board game framework from rancebot so I can reasonably quickly shove stuff live after only implementing game rules.
@mint@Cocoa@Dullahan@NEETzsche@Rance@Zergling_man Anni at least is blocked at domain level. So they do block. Probably blocked all the naughty word instances just so their game could have wider appeal, but it hasn't even been used since Wed, 25 Oct 2023 07:18:35 GMT
@NEETzsche@realman543@Dullahan@Cocoa@Rance@Zergling_man Zerg brought up his WIP RPG battle bot which I thought might be of your interest, then the conversation shtfted to game bots in general. The post in question refers to castling.club, a standalone ActivityPub chess game server.
A lot of the context messages in this thread never federated for me. A chess bot is interesting. RPG bots could be interesting as well, but honestly at this point my interest in TTRPGs is primarily in writing prose for PbP, and using dice bots to simulate, well, dice.
Checking if a move is legal on a chess board is a surprisingly elaborate process and this may have convinced people to not write an effective chess bot on fedi.
It's not an impossible task by any stretch, it's just a pain in the ass. Also, if I were to write a chess bot on fedi, I wouldn't just put all of the content in the content field. I'd create some new ActivityPub document types to accommodate a chess game. I'd probably do things like make the a Move object, and make it so that there's from and to field, each an (x,y), then after the fact write a little function to convert that into chess algebraic notation.
I really would hate to make the bot something that just puts a bunch of human-readable shit into content and leaves devs wanting for some way to manipulate/analyze the game data other than regex hacking the fucking HTML or whatever. That would suck.
@Zergling_man@NEETzsche@Dullahan@Cocoa@mint@Rance Either the bot is busted or the blocks aren't being picked up because I tried putting myself out there for challenges and challenged dcc and neither worked.
I actually love regex. I learned it playing MUDs writing "triggers," which are basically scripts tied to a regex that some MUD clients had. When the client receives text that meets the regex, a script is run, and the regex captures are put into variables like $1 or something like that, depending on the scripting language.
Really elaborate combat systems were written using these, as global variables were a thing, which essentially automated the strategy game element of games like Aetolia. One such system was one of my first programming projects as a 13yo.
Calling vidya RPGs "roleplaying games" triggers my autism because you're not usually doing more roleplaying than, say, Duke Nukem 3D. Pressing X on your PlayStation controller for more dialog is not roleplaying. They're more like strategy games which model individuals with much more granularity than, say, RTS, 4X, or "grand strategy" games.
I'm not going to create a big thing with this it's just a remark I'd like to make.
Regex is pretty sweet in certain circumstances but now that I think on it, in the context of processing HTML in a content field, if you're using Regex to scrape out details from JSON or XML or something like that you are doing it wrong. Unless you're writing something that, for whatever reason, needs to process literally thousands of chess moves per second, just take the memory footprint L and use one of those processing libraries that turn your JSON into an associative array or dictionary or whatever your preferred meme languages calls it. It's just not worth the headache.
Regex shines when you're dealing with MUDs, though. It was perfect for that:
/^(\w+) slashes your face with (his|her) dirk, dealing you (\d+) damage\.$/
Which I think is probably the true intended use case of regex, stuff like that.
I use AI to generate boilerplate code and then just proofread/fix it. The ways it breaks is usually one or two critical points, like some logic error or something like that. It does improve productivity but it absolutely does not replace the programmer.
@NEETzsche@Dullahan@Cocoa@mint@Rance@Zergling_man Also AI generated most programming snippets don't work. I've told this story before once but when I asked it's help with a simple programming problem the AI pooped out a piece of code the required a non-existent class from a non-existent file from a non-existent folder.
Well, the reason it likely won't work for now is the context window is limited. That's the big limiter. For a while, the big limit was based on quantity of training data, but we've reached diminishing returns on adding more training data, and now the output is actually pretty good given whatever limited context it has. So now the big deal is how many tokens you can give it at any given moment. The reading comprehension of modern LLMs is actually very good, but if you only give it two classes and not the whole project, it can only infer so much.
For example, if you want to automate away the role of GM with something at the reading comprehension of GPT-4, you'd probably need millions of tokens. You'd need enough context to put in the manual and all of the supplements. You'd need it to have a log of all events so far in the game. You'd need it to record past GM rulings to remain internally consistent.
If you deny the LLM these things in its context window, can you really blame it for not knowing some very important detail from early on in the game, that's just too far back for it to remember? No. You really can't. So the development of these LLMs will mostly be about increasing its limit in this respect.
Personally I'm just waiting for the Russians or Chinese to throw the requisite GPU cycles to make their own English-language LLM that competes with GPT and Gemini etc. That way we'll be dealing with their content management jannying instead of the kind that's pertinent to us here. So instead of not being allowed to say "nigger" we won't be allowed to say that Taiwan is a country or refer to Russia's invasion of Ukraine as a "war"
@NEETzsche@realman543@Dullahan@Cocoa@mint@Rance@Zergling_man everyone's already wearing a mask online anyway in anon space so it's tough to distinguish gpt bots from actual retards sometimes if they put on a semi convincing performance and aren't marked as bots.
GPT-4 does open links now but they didn't do it for a while. One thing I notice a lot of people doing is arguing with ChatGPT or trying to abuse it. I ran a bot in my channel a while ago and people did both. They treated it like a person. It was insanity.
In reality it's just a fucking computer program that can, sometimes, automate the drudgery of writing, somewhat. The output still needs a lot of editing, and always will.
@NEETzsche@Dullahan@Cocoa@mint@Rance@Zergling_man >but if you only give it two classes and not the whole project, it can only infer so much. Actually the entire code is open source and AFAIK they weren't one of the niggers demanding not to have their code scrapped. That said the codebase is ass and incredibly complex which is where I think the issues came in. Also I think it wasn't actually referencing the specific project, it was making inferences based on my prompt and just guessing what would actually work.
@realman543@Rance@mint@Cocoa@Dullahan@NEETzsche People already demonstrated that the bot doesn't open links. Someone did it with like research papers by Taylor Swift or something and gave links to her publications, and only one of the four the bot actually said was fake even though they all were.