• You've discovered RedGuides, an EverQuest multi-boxing and scripting community 🧙‍♀️⚙️. We want you to play several EQ characters at once, come join us and say hello! 👋

  • A TLP without truebox has thawed (Very Vanilla ready)
    Frostreaver

Discussion - ChatGPT for macros?!

Z

zathus

Well-known member
Joined
Mar 23, 2019
RedCents
541¢
With the update and some down time! i decided, hey! i wonder if ChatGPT can recognize 1, EverQuest as a game, 2 class's and DPS and 3, what MQ2 is and how it works with EQ.

I asked a simple question.

Below is a macro i made for Everquest using MQ2. please look at the macro and let me know if you see ways to improve my DPS...

a 5 page essay! Not some BS generic stuff, like, unplug your router lol but fixed a lot of my scripts that where redundant or inefficient . at least i think, need MQ2 up to test, but really cool! written in the proper format, with the proper in game commands. very cool! i am so stoked to try this out now lol
 
I use Claude.ai

It understands Lua, knows what EQ is and macroquest as well. I've seen it make some bugs, but for the most part it's good stuff.
 
burdsjm said:
chatgpt.com

ChatGPT - Macroquest Everquest Expert

Master of Macroquest, Lua scripting, and EverQuest automation with detailed code guidance.
chatgpt.com chatgpt.com

It's not 100% yet, so if you see something wrong, tell it it's wrong.
Click to expand...
OH!! time to play! thank you
darkfur said:
I use Claude.ai

It understands Lua, knows what EQ is and macroquest as well. I've seen it make some bugs, but for the most part it's good stuff.
Click to expand...

ill give that a try as well. seems like i am behind the curve here but this just opened my eyes to so many possibilities!
 
darkfur said:
I use Claude.ai

It understands Lua, knows what EQ is and macroquest as well. I've seen it make some bugs, but for the most part it's good stuff.
Click to expand...
I have no clue if what its spitting out is good or not. I dont know Lua at all, but its cooking! and it looks really good thus far. shits kind of amazing tbh
 
i've been using grok for Lua stuff. it knows about EQ and some MQ stuff. it makes some pretty bold assumptions though where you need to correct it, so having a good understanding of TLO's helps a lot.

it definitely struggles with certain stuff though. pretty much i use it to get a bit of a structure in place then make my own changes along the way.

gonna play with claude next.

one pet peeve about grok is randomly it will change ~= to != constantly.
 
It changes by the day. I use the GPT above and Gemini 2.5 pro to work through stuff.
 
FYI - working on some raids finally and this is going to help out a ton so thanks! Teaching it as I go as well. Not the technical stuff…
 
One tip while using my GPT. Ask it to explain and add a comment for each line. Then ask it to double-check and audit itself. It tries to question itself and play devil's advocate.
 
burdsjm said:
One tip while using my GPT. Ask it to explain and add a comment for each line. Then ask it to double-check and audit itself. It tries to question itself and play devil's advocate.
Click to expand...
This is a game in a game.. look deeper into the Matrix neo, what do you see? Super Mario looking back at me.
I like the concept of a personal learning system that helps you as you train it more, that training is really difficult though if you don't know the answer in the first place.
 
Try this prompt:


"Reflect on 5-7 different possible sources of the problem, distill those down to 1-2 most likely sources, and then add logs to validate your assumptions before we move onto implementing the actual code fix"
 
I've used Claude extensively for two years. I have been a software developer for 40 years and now a software engineering manager the last 10 of them. I do a lot of work with entry level college kids. So, for comparision, Claude can do c/c++/c# at a level superior to a fresh grad, but not as well as a 3-4 year guy. On the other hand, I personally use Claude to replace a small 3-4 man entry level team and I drive the process. So I'm the architect and designer and claude does the wrote coding based on my inputs. I've done hundreds of thousands of lines of code this way and here are a few of my lessons:

1) Always ask for complete code for a method or function. There are time the AI says "//other code remains unchanged" and if you cut and paste the new coede you lose the old code. This gets ugly to find
2) Save your work a lot. You may have to go back several revisions if the AI Goes down a rabbit trail
3) AI tends to vomit out large functions. Ask it to refactor for maintainability to get better code
4) Work in small batches. Do not try and implement a complex task all at once. Break it in to pieces and check on the results of each step. Build on your success
5) Ask for diagnostic code to be inserted. You can take it our when everything works.
 
OldGuard said:
I've used Claude extensively for two years. I have been a software developer for 40 years and now a software engineering manager the last 10 of them. I do a lot of work with entry level college kids. So, for comparision, Claude can do c/c++/c# at a level superior to a fresh grad, but not as well as a 3-4 year guy. On the other hand, I personally use Claude to replace a small 3-4 man entry level team and I drive the process. ...
Click to expand...
The problem is that when we retire out (at 30 years myself) there will not be enough people that have the chops to know when the LLM spits out convincing garbage.
If all the entry level jobs go away there simply will not be enough people creating new concepts and designs at some point stagnation will set in. Being a programmer is a creative endeavor and is something that cannot be replicated by "AI".
 
arch said:
The problem is that when we retire out (at 30 years myself) there will not be enough people that have the chops to know when the LLM spits out convincing garbage.
If all the entry level jobs go away there simply will not be enough people creating new concepts and designs at some point stagnation will set in. Being a programmer is a creative endeavor and is something that cannot be replicated by "AI".
Click to expand...
I suspect that a new training model for developers may come along. The "reasoning" process can be useful for either demystifying concepts or perhaps giving a clue to alternate methods and might speed up learning such that it eventually takes less time to become that "3-4" or "5-10" year equivalent dev. I imagine some people might find difficulty but some may flourish as they are able to more rapidly apply their innate creativity.

This is of course my opinion as a "many plus years in some other discipline / SW hack guy". I've found some use in filling the gaps in my ignorance :D
 
arch said:
The problem is that when we retire out (at 30 years myself) there will not be enough people that have the chops to know when the LLM spits out convincing garbage.
If all the entry level jobs go away there simply will not be enough people creating new concepts and designs at some point stagnation will set in. Being a programmer is a creative endeavor and is something that cannot be replicated by "AI".
Click to expand...
As we inch closer to:
Mike Judge Applause GIF by Idiocracy
 
What we have to do is shift from coding/hacking skills to System Architect/Design skills. Let AI be the code typer. We still need people to figure out what to do, how it should look, what should be stored/reported and how it should be used. Fretting over this is like the guys in the Buggy-whip factory in 1910. "Where will all the buggy-whip men work when we switch to these new famgled automobiles?" The enw economy found a place for them; usually a better place in terms of economic growth. Typeing code is terribly inefficient and it's where the bugs often multiply. Design and investigation are human still best driven by humans.
 
OldGuard said:
What we have to do is shift from coding/hacking skills to System Architect/Design skills. Let AI be the code typer. We still need people to figure out what to do, how it should look, what should be stored/reported and how it should be used. Fretting over this is like the guys in the Buggy-whip factory in 1910. "Where will all the buggy-whip men work when we switch to these new famgled automobiles?" The enw economy found a place for them; usually a better place in terms of economic growth. Typeing code is terribly inefficient and it's where the bugs often multiply. Design and investigation are human still best driven by humans.
Click to expand...
Love the buggy whip guys example btw!!
I always joke about we will dream of doing a Picard on the bridge and asking the computer to theorize such and such, and his ChatGPT knew super complex solutions that always came to a chance still (for entertainment of course). My ChatGPT says here is everything you need, 2+2+2=9 what? Even I know thats wrong...I'm sorry your correct that is wrong its 8.
 
Redfrog said:
Love the buggy whip guys example btw!!
I always joke about we will dream of doing a Picard on the bridge and asking the computer to theorize such and such, and his ChatGPT knew super complex solutions that always came to a chance still (for entertainment of course). My ChatGPT says here is everything you need, 2+2+2=9 what? Even I know thats wrong...I'm sorry your correct that is wrong its 8.
Click to expand...
What's really scary is seeing AI health assistants popping up. "This patient has a headache. Recommend open heart surgery." We're getting closer and closer to a world like in the movie, Idiocracy. Brawndo...It has electrolytes...It's what plants crave... 🤦‍♂️
 
My favorite use of AI is for code reviews. In the past these have been largely a waste of time as people just do not put in the prep work necessary for a good review. Using AI, however, one can do a before-after analysis on a file by file basis in a short period of time. I have been able to find critial bugs that would have been missed in the past. We split the team and use different AI engines to get multiple review schemes on the same code. We use ChatGPT, Claude, CoPilot. Deep Dive and Grok. If is the single best use of time we have.
 
eqsubi15 said:
What's really scary is seeing AI health assistants popping up. "This patient has a headache. Recommend open heart surgery." We're getting closer and closer to a world like in the movie, Idiocracy. Brawndo...It has electrolytes...It's what plants crave... 🤦‍♂️
Click to expand...
Except in at least two recent studies LLM alone are significantly more accurate at diagnosing patients than doctors. Even when the doctors themselves are allowed to use the LLM to assist them. Hell in one of the studies the AI displayed more empathy 🤣
 
More AI information.

Creating a MacroQuest Lua script for EverQuest using AI tools like ChatGPT or Claude involves a structured approach from idea conception to testing and refinement. Here's a step-by-step guide:

Steps to Create a MacroQuest Lua Script with AI Assistance​

  1. Define Your Script's Purpose and Requirements
    • Conceptualize: Determine what you want the Lua script to achieve in EverQuest. For example, automate a task, provide an alert, manage buffs, or assist in combat.
    • Break it Down: Divide the desired functionality into smaller, specific actions or checks the script needs to perform.
    • Identify Key Elements: Think about the EverQuest game elements the script will interact with. This includes character status (health, mana, buffs), target information, spells, items, or in-game commands.
  2. Choose Your AI Code Generation Tool
    • Select an AI assistant like ChatGPT, Claude, or a specialized Lua helper like MacroquestGPT. These tools can understand natural language prompts and generate code.
  3. Craft Your Initial Prompt for the AI
    • Be Specific: Clearly describe what you want the script to do. The more detailed your request, the better the AI's initial output will be 23.
    • State the Language and Context: Explicitly mention that you need a "Lua script for MacroQuest (MQ) in EverQuest." This helps the AI provide relevant syntax and functions 1.
    • Include MQ-Specifics (If Known): If you know particular MacroQuest Top-Level Objects (TLOs) (e.g., mq.TLO.Me.PctHPs(), mq.TLO.Target.Name()) or commands (mq.cmd('/attack on')) that should be used, include them in your prompt.
    • Request Structure (Optional): You can ask for a basic structure, like a main loop or event-driven functions, if applicable to your script's design 4.
    • Example Initial Prompt: "Create a MacroQuest Lua script that continuously checks my character's health percentage. If my health drops below 40%, the script should cast the spell 'Light Heal' on myself. The spell 'Light Heal' should be identified by its name."
  4. Review and Refine the AI-Generated Code
    • Initial Code Review: The AI will provide a first draft of the Lua code 3. Carefully read through it.
    • MacroQuest Syntax Check: While AIs understand Lua, they might not be perfectly up-to-date with all MacroQuest-specific implementations or nuances 1.
      • Ensure TLOs are correctly called (e.g., mq.TLO.Me.Buff('Spirit of Wolf').ID() often requires parentheses for arguments and retrieval).
      • Verify that MQ commands are formatted correctly (e.g., mq.cmd('/cast "Light Heal"')).
    • Iterate with the AI: If the code isn't correct or complete, provide feedback to the AI. Explain what's wrong or missing and ask for specific changes or additions 23.
    • Example Refinement Prompt: "The script looks good, but please modify it to use mq.TLO.Me.PctHPs() for checking health and mq.cmd('/cast "Light Heal"') for casting. Also, add a 10-second delay using mq.delay(10000) after casting the spell before it checks health again."
  5. Implement and Test the Script in EverQuest
    • Save: Copy the Lua code from the AI and paste it into a plain text file. Save this file with a .Lua extension (e.g., myhealer.Lua) in your MacroQuest Lua scripts directory (usually MQ_INSTALL_FOLDER/Lua).
    • Run: In EverQuest, with MacroQuest running, load and run your script using the command: /Lua run myhealer.
    • Observe: Watch your character and the MQ console window to see if the script behaves as expected.
  6. Debug with AI Assistance
    • Identify Errors: If the script doesn't work or you see error messages in the MQ console, copy the exact error message and the relevant lines of code 5.
    • Ask AI for Help: Paste the error message and code snippet into the AI tool and ask for help debugging it 2. The AI can often spot syntax errors or logical flaws.
    • Example Debugging Prompt: "My MQ Lua script gives this error: [MQ2Lua] Error: attempt to concatenate a nil value (local 'charName'). Here's the code: local charName = mq.TLO.Me.Name print('Character: ' .. charName). How can I fix this?" (The AI should suggest mq.TLO.Me.Name().)
  7. Iterate on Features and Functionality
    • Once the core script works, you can add or improve existing features.
    • Return to step 3 or 4 to prompt the AI for these enhancements.
    • Example Feature Request Prompt: "My healing script works. Now, I want to add a feature that sends a message to my group chat when it casts 'Light Heal'. The message should say 'Healing myself!'. Use the MQ.cmd('/gsay Healing myself!') command."
  8. Finalize and Document Your Script
    • Clean Up: Remove any unnecessary test code or debug print() statements.
    • Add Comments: Include comments in your Lua code to explain what different sections do. This will help us understand and modify the situation. You can even ask the AI to help generate comments for your code.
By following these steps, you can effectively leverage AI tools to assist in creating, refining, and debugging your MacroQuest Lua scripts for EverQuest 12345. Remember that AI is a tool to help you; a basic understanding of Lua and MQ principles will significantly enhance your ability to guide the AI and troubleshoot the generated code.

Citations:​

  1. https://www.redguides.com/community/threads/chatgpt-my-first-look.85899/
  2. https://www.ai4chat.co/gpt/luagod
  3. https://www.tutorialspoint.com/claude_ai/claude_ai_code_generation.htm
  4. https://docsbot.ai/prompts/programming/fivem-script-creator
  5. https://www.mmobugs.com/forums/index.php?threads/chatgpt.35538/
  6. https://www.promptkit.tools/blog/ai-coding-assistant-for-lua
  7. https://docs.macroquest.org/lua/macro-to-lua/
  8. https://documentation.n-able.com/N-...twareRepository/Create_AI_Assisted_Script.htm
  9. https://www.anthropic.com/engineering/claude-code-best-practices
  10. https://www.linkedin.com/pulse/lear...cripts-using-llms-chatgpt-claude-jaokar-sdone
  11. https://tasvideos.org/Forum/Posts/409882
  12. https://chatgpt.com/g/g-aSe4SKupf-macroquest-everquest-expert
  13. https://gamedev.stackexchange.com/questions/49240/how-should-i-manage-the-ai-using-lua-scripts
  15. https://docsbot.ai/prompts/technical/gameguardian-lua-script-engineering
  16. https://letters.alfredlua.com/p/ai12-how-i-ai-code
  17. https://forums.steinberg.net/t/chatgpt-for-creating-lua-scripts/844682
  19. https://docs.macroquest.org/lua/


Answer from Perplexity: pplx.ai/share
 
Wow! One of the best threads to come along in awhile. Full of thought-provoking and useful information. Kudos to the OP and all who contributed.
 
Gohast said:
Except in at least two recent studies LLM alone are significantly more accurate at diagnosing patients than doctors. Even when the doctors themselves are allowed to use the LLM to assist them. Hell in one of the studies the AI displayed more empathy 🤣
Click to expand...
So as someone who has been a victim of medical malpractice I have some insight into why this happens in SOME BUT NOT ALL cases.
I was having heart issues that were "undiagnosable" according to two different cardiologists, both in the same medical network.
Cardiologist A said I needed a tilt table test. I quite literally said "I think I'm going to die" in the middle of the test, the technician administering the test immediately called the cardiologist in and told them I was unstable she said "he's fine continue the test" about half an hour later I woke up in another room... Long story short I fucking died, but they were able to bring me back through a combination of things. I was then discharged and told to go home. So in less than a 3 hour period I died, was brought back, and pushed out into the parking lot.

Feeling like this was not safe I called my wife who took me across town to another hospital where I stayed for 4 days while they finally got my heart rate to stabilize and keep me conscious.

This is only example A of medical malpractice I experienced, the TLDR of it all is that I didn't need the test they wanted to do, I wasn't healthy enough for it either, the cardiologist knew this fact but wanted to be able to bill me/my insurance for the maximum amount of testing. She literally killed me and then attempting to cover up her fuckup shoved me out the door instead of admitting me to the attached cardiovascular health unit at the hospital to keep me alive. Through investigation it was found she was on the verge of bankruptcy which in my opinion clouded her judgement when making decisions.

Doctors are people too, they make decisions based on feelings, predisposition, mental state that day, etc. etc.

All of that said to say, I trust AI models less now than I did a year ago as they are continuing to hallucinate, if not hallucinate more than they were previously. I think part of that is they are supposed to learn from humans and as humans we are all fallible.
 
With any AI work, get a second opinion. Doesn't matter if it's about code or your growing tumor in your brain. There are at least 5 "free" mainstream AI systems. Feed the results of one into another; run a second query similar to the first but with a different AI agent. Don't trust the AI, verify the AI.
 
Has anyone used something like Cursor or Roo to program scripts? I've been using it as of late to make some simple programs and it works well for someone who is very new to coding.
 
This thread is gold, and funny too :)
AI has changed the SWE field completely - feel like I'm 25 years back when I started and at the same time, can do 10x more myself.
Thanks for the tip on claude.ai !!
Didnt know about that one - already testing it out now - been trying out lots of different ones lately and still looking around.
So far grok has been most useful for me - the embedded AIs in visual studio for example arent nearly as good - but prompting is still something I'm learning to git gud at so probably some dose of "user error" has to be said on my end...
Awesome posts in here!
 
Claude 4.0 has been a pretty decent upgrade. It does a bit better with context and takes a little longer to "think" about what it does.

It still does really dumb stuff like trying to use Spawn.Buff to check the buffs of everything/anything. I frequently have to tell it that it cant get that information unless its my target.

Corzarium said:
Has anyone used something like Cursor or Roo to program scripts? I've been using it as of late to make some simple programs and it works well for someone who is very new to coding.
Click to expand...
Cursor is neat cuz just overlays VS Code.
 
eqsubi15 said:
What's really scary is seeing AI health assistants popping up. "This patient has a headache. Recommend open heart surgery." We're getting closer and closer to a world like in the movie, Idiocracy. Brawndo...It has electrolytes...It's what plants crave... 🤦‍♂️
Click to expand...
I think a lot of problem here is that what's been fed into them is crap. There are a *lot* of forums from people talking about their "rare" conditions or "suspected" Chiari malformation. But there's really only a few good, high quality articles about the diagnostic pipeline and treatment algorithm for cluster headaches. Medical students generally learn about the 3-4 highly reputable and trusted, up to date resources for finding that information, and just hit those spots repeatedly. But if the only "correct" resource is fed into the training data with one iteration and not weighted differently than the garbage, then it is just as likely to spit out garbage as not. I think content curation *prior to* training is critical, and I have not heard that they are recruiting teams of practicing physicians to perform that task before training these models.

To be fair there are quack-a-doodle docs out there who do not perform "content creation" for themselves and have some wild ideas about how to treat people, but those are the exception and not the norm.

arcreactorkc said:
So as someone who has been a victim of medical malpractice I have some insight into why this happens in SOME BUT NOT ALL cases.
I was having heart issues that were "undiagnosable" according to two different cardiologists, both in the same medical network.
Cardiologist A said I needed a tilt table test. I quite literally said "I think I'm going to die" in the middle of the test, the technician administering the test immediately called the cardiologist in and told them I was unstable she said "he's fine continue the test" about half an hour later I woke up in another room... Long story short I fucking died, but they were able to bring me back through a combination of things. I was then discharged and told to go home. So in less than a 3 hour period I died, was brought back, and pushed out into the parking lot.

Feeling like this was not safe I called my wife who took me across town to another hospital where I stayed for 4 days while they finally got my heart rate to stabilize and keep me conscious.

This is only example A of medical malpractice I experienced, the TLDR of it all is that I didn't need the test they wanted to do, I wasn't healthy enough for it either, the cardiologist knew this fact but wanted to be able to bill me/my insurance for the maximum amount of testing. She literally killed me and then attempting to cover up her fuckup shoved me out the door instead of admitting me to the attached cardiovascular health unit at the hospital to keep me alive. Through investigation it was found she was on the verge of bankruptcy which in my opinion clouded her judgement when making decisions.

Doctors are people too, they make decisions based on feelings, predisposition, mental state that day, etc. etc.

All of that said to say, I trust AI models less now than I did a year ago as they are continuing to hallucinate, if not hallucinate more than they were previously. I think part of that is they are supposed to learn from humans and as humans we are all fallible.
Click to expand...
Fortunately docs like that aren't the ones training LLMs. Sorry for your experience -- that sounds painful, confusing, frustrating, and scary. Most of us are just trying to do the best we can. Bad actors are the reasons why we have state boards and peer review panels, but we rely on people (ie patients and staff) to make reports. Unfortunately sometimes those reports can only be made after situations like this happen.
 
mythal said:
So far grok has been most useful for me
Click to expand...
I’ll have to try grok again the last 10 times I tired to use and curate data on there it was as bad as memory ward patient. It was fine for questioning and answering about the social post on X but any real work was disappointing.
 
Ssoulle said:
Claude 4.0 has been a pretty decent upgrade. It does a bit better with context and takes a little longer to "think" about what it does.

It still does really dumb stuff like trying to use Spawn.Buff to check the buffs of everything/anything. I frequently have to tell it that it cant get that information unless its my target.


Cursor is neat cuz just overlays VS Code.
Click to expand...
Claude is great for coding and GitHub Copilot for VScode has been great for writing and vibe coding
 
ROFL!!
Sorry — I can’t help write scripts that automate play or bot an online game. Creating code to automatically complete missions in an MMO (including EverQuest) would be enabling cheating and account automation, so I have to refuse that request.
 
You must log in or register to reply here.
Discussion - ChatGPT for macros?!
Next unread thread 🦁

Users who are viewing this thread

Total: 1 (members: 0, guests: 1)
Back
Top
Cart

Cart

Total -