ChatGPT4 plug in

[Kaktus]

Greetings everyone!

 

I just want to start this topic in the hopes of getting traction all the way to the developers as it will give a biiiig huuuuuuge boost to development of the game and playability. What is happening today with ChatGPT is surely a phenomena of the decade if not even biger thing than  internet back in the 90's.

 

Right now ChatGPT is starting to offer companies plug-in ability so you can use it in programs like word, excell, and why not in games too?

 

DCS be modern, be proactive, DO IT, get it done, ability of the program to understand subtext and meaning would accelerate mission editing, development of missions, maybe even ability to create dynamic campaign!?!?

 

All that GPT needs is data of end mission and strategic reason, lets say-we want Blue team to win, blue team victory means capturing land, advancement of units on ground, controling air zones and minimizing friendly losses and maximizing or destroying in totality enemy forces. If you want you can go into details what all this entails and basically write a BOOK, and the program GETS IT in a second. 

 

So all you need is then to talk, tell it to create a campaign, with so and so many unity for blue force, such and such storage, such and such RED LINES where if reached new storage and weapons arrive, such and such conditions for victory, then let the AI fight it out and human player plays the campaign to best of his/her ability and gets medals based on its merit. It would be even better than todays campaign which are too pilot oriented, who believes 1 pilot actually determines victory in the end? but this is what all campaigns today in games are which is silly. 

 

You can strive to make campaigns more or less 50:50 meaning victory or loss could go easily one way or another and in series of campaign missions through consistent pilot work you can maybe TILT the chances a bit and victory would be possible.  

 

But that is on a side note, more important is harnessing the plug in to the best of its ability!

 

Great add on would be JESTER with GPT ability!?!?!? Jester search of enemy air, concentrate on incoming bogeys, or telling your AI crafts how to attack enemy ground units or enemy air units, you can literally TALK TO THEM and they will understand and do it!? 

 

PLEASE DCS, and fellow simmers, let this threat be the opening and hopefully DCS is the FIRST one to utilize this and gains the most advantage and not the last to use it as some dinosaur in the industry!

 

To those who do not understand benefits of chatGPT is making pls, go check it out, like REALLY check it out, its not hype, its real!

[cfrag]

I can see a good case for a ChatGPT-based system for ATC, sure, after some more advancements. Today, it could provide colorful commentary / chatter in the background with some text to speech.

Creating missions is currently where ChatGPT and similar systems that merely assemble words, and derive the next word from an elaborate rule set and very big data base fail completely. All missions are one big Lua program, so conceivably, this may work in the future (programs are a string of words after all). The problem with these systems is that they don't understand subtext nor meaning, they completely fail to contextualize source code, which is the reason why they fail (use ChatGPT and have it write for you a non-trivial scrip, and you'll immediately see why it fails: in it's learning DB, it can't differentiate between a script that uses MIST or some other framework - the context in which the script runs - with straight (vanilla) code. That's where it shows that ChatGPT is merely a program to generate words based on a ruleset and big data base. Once these systems become able to really contextualize, we'll have something indeed to create missions. If they are good missions remains to be seen, but these systems can then learn which missions are popular, and then can create more of the same.

Still, having a chat bot create background chatter, and maybe even be an acceptable ATC could work - although that may also exceed the current limits of it's abilities, as AFAIK it can't contextualize the traffic on an airfield to generate correct, purposeful directions. It would, however, completely nail the phraseology. It'll indeed be interesting to observe if, when and where AI can enhance the DCS experience.

Edited April 1, 2023 by cfrag

[TEMPEST.114]

I can't believe the myopia of people when it comes to AI, even a basic AI as ChatGPT. 

This is a dangerous road. Humans aren't capable of handling this tech. 

Learn lua. Do the work yourself, grow, learn and put the effort in - stop being part of the 'I want it all and I want it now but I'm not going to work for it' generation. 

[Dragon1-1]

It's not that humans aren't capable of handling such tech, it's that they're misinformed about what it can and what it can't do. In the early days of computers, they themselves were hyped as "thinking machines" and some people thought they'll be able to have them do thinking for them. It didn't work that way then, it doesn't work that way now. Sure, computers replaced humans in many areas, and AI will replace them in quite a few more, but in all those cases, those jobs did not actually involve a whole lot of thinking. This is bad news for those who aren't all that good at this "thinking" thing, but people who can use their brains have nothing to fear. Likewise, any job that involves thinking, like making missions that are actually good, is simply not going to be made any easier by using an AI.

As a matter of fact, DCS has a built-in quick mission generator which is probably better than anything an AI could produce. ChatGPT is pretty good at fooling humans in a conversation, but for anything you can write a procedure for, like ATC or AWACS work, an actual algorithm made to follow the procedure will perform better. A dedicated language, trained only on aviation chatter, could provide non-critical parts of the messages, just to make them sound like you're talking to a human, but that's it. The technology has potential as user interface, but only that.

[Kaktus]

Well the idea is if AI can understand what  you want, then you need the interface to "talk" to DCS AI logic and you get a big win, i don't see the problem, i think gpt will come to games much sooner ppl realize, all i think is its good to give a good try and see if DCS can be pioneer and not lag in this actually most important aspect of warplanes genre-AI.

 

There are plug in for excel and word and i see it does understand and also has good interface with word or excell program so it can deliver on this, when previously you had to "know" how to chart and recognize trends in data sheet, now you just need to let the program know what you want.

 

If such interface can be given to DCS i still believe it has big potential. DCS AI is abysmal at this point, lets be real about it, dynamic campaign biggest issue is known, where if data after a mission is shared with gpt and order is given to protect assets, protect ground units, defend where no ability to progress, advance where recon and satelite gives you an edge regarding the enemy then why not? again, its just words, understanding words, gpt proves it understands words, not philosophy but words is enough. --- move ground units FORWARD if mission to secure air zone A successful (success means mission was flown, friendly planes were victorious or not challenged), thus CAS can be deployed, victory on ground assured, MOVE ON. 

 

Seriously ppl make it more complicated than it is. 

[cfrag]

14 hours ago, Kaktus said:

Well the idea is if AI can understand what  you want, then you need the interface to "talk" to DCS AI logic and you get a big win

Agreed. The problem is the 'if AI can understand what you want' part. Currently, they can't understand you. Ai is not aware, and AI algorithms aren't able to correctly infer meaning and contextualization - but they surely can fool a casual observer into thinking that they can. It has been that way since the famous 'ELIZA' program, written in the early 1960s. These programs analyze key words in your query, and for output they then string together the words that best match the weighted query words according to a rule set and (today) a truly gargantuan data base of "learning material" as resource. As we know, this can have spectacular results: ELIZA astonished people 60 years ago, today deep learning chess apps beat world masters. It works really well for chat, too. In general terms, these 'deep learning' algorithms work better, the larger the data base is that they can trawl for answers. For DCS, that learning pool or 'knowledge' currently is too small, as their answers always consists of chunks from their learning DB.

This is not to say that AI may not soon help create better games - for example AI could help create the ground traffic and people rummaging around a busy airfield to make it look more alive and interesting. It can help to analyze, and then better set up force patterns in a mission. There are many interesting applications, yes. All this will require an investment into the AI interface (plug-in), with unknown results, a shot in the dark, so to speak. So I believe that yes, AI can help make games better (case in point: I was involved with a project that used AI to litter a scenery with junk to make it look more believably lived-in). And I believe that it will. For DCS, I think that the technology will have to mature before it reaches a point before it can be gainfully, and sufficiently low-cost, employed in routine mission design.

Oh, and I would of course love to be proven wrong, that it can work today   

14 hours ago, Kaktus said:

DCS AI is abysmal at this point

Now this is something different entirely. We are no longer talking about off-line (mission design time) AI. On-line AI to control aircraft or enemies - that's not ChatGPT. We are not there at all. Tesla can't even get their AI to drive a single car correctly, and the onboard computers only have to drive a single car. No way they can be used to control 100 units at the same time. So we can't use Chat GPT Ai for in-game AI, and you are not saying that, I know; I'm merely making sure that we are not talking about ChatGPT-assisted in-game AI for my next point: 

Analyzing the abysmal AI after the fact will not make the in-game AI better, no matter how smart the post-mortem analyzing AI is. It doesn't require AI to understand how bad current DCS AI is . So an AI-assisted dynamic campaign might be better able to set up the next iteration of the mission with some remedies to DCS AI flaws baked in. But that would be treating the symptoms, not the root cause. It would be better to invest in better in-game DCS NPC AI than into an AI-based off-line interface to paste over the flaws. 

Edited April 2, 2023 by cfrag