asked ChatGPT (collection) - answers correct?

[seeh]

# Beginners difficulties

I have asked ChatGPT and got this answer:

Beginners often have difficulty understanding the concept of game planning and how it relates to the lobby and other players. Common issues include:

1.  Knowing how to set up the game lobby and make it accessible to other players
2. Understanding the different roles and responsibilities of players in the lobby
3. Figuring out how to handle player interactions and conflicts
4. Learning the different game mechanics and rules
5. Building the necessary components for the game, such as the user interface, game logic, and networking code
6. Optimizing the game for different platforms and devices
7. Testing and debugging the game to ensure it works as intended
8. Publishing the game to the appropriate platform(s)
9. Promoting the game to potential players---

Tokens used: 315

# What are the most important and least intuitive concepts to understand while learning how to playing 0ad effective?

## Most Important Concepts

1.  Unit Control: Understanding how to control and use your units is essential to successful gameplay in 0ad. Knowing which units to build and when to build them, and understanding how to effectively maneuver and utilize your units in battle is a key skill to master.
2.  Economy: Building an effective economy is critical for success in 0ad. Learning how to efficiently gather resources and manage your economy is essential to having the resources necessary to build powerful units and structures.
3.  Map Control: Understanding the terrain and knowing how to use it to your advantage is an important concept in 0ad. Knowing how to utilize choke points, high ground, and other terrain features can give you an edge in battle.

## Least Intuitive Concepts

1.  Unit Formations: Knowing how to effectively use formations can greatly improve your unit's effectiveness in battle. Understanding how to effectively use different formations and how to effectively maneuver your units within a formation can be difficult to master.
2.  Strategic Planning: Knowing how to plan ahead and anticipate your opponent's moves is essential for success in 0ad. Anticipating your opponent's moves and planning ahead can give you the edge in battle.
3.  Counter-play: Knowing how to counter your opponent's units and strategies is essential for success in 0ad. Learning how to identify and counter your opponent's strategies and units can give you the advantage in battle.

---

Tokens used: 446

 

 

[seeh]

# Whey people love the game 0ad?

People love the game 0ad for many reasons. It is a real-time strategy game with a focus on historical accuracy and strategy. The game is highly moddable, allowing players to customize their experience and create new content. The game also has a vibrant online community, with players helping each other out and sharing tips and strategies. Additionally, the game has beautiful graphics and a deep and engaging story. All of these elements combine to create an enjoyable experience that keeps players coming back for more.

Tokens used: 246

[Sun Wukong]

Can you ask it to code a less buggy engine?

[Vantha]

Why not ask Bing Chat to correct ChatGPT?

Here is Bing Chat's reply:

"Hello, this is Bing. I can help you with your text about 0ad. Here are some suggestions:

- In the first paragraph, you may want to explain what 0ad is for readers who are not familiar with it. For example, you could add a sentence like this: "0ad is a free and open-source real-time strategy game of ancient warfare that is under development by Wildfire Games."
- In the second paragraph, you may want to clarify what you mean by "game planning" and how it relates to the lobby and other players. For example, you could add a sentence like this: "Game planning refers to the process of designing and creating a custom game scenario that can be played online with other players in the lobby."
- In the third paragraph, you may want to use bullet points instead of numbers for the list of most important and least intuitive concepts, as they are not ranked or ordered. For example, you could write:

## Most Important Concepts
- Unit Control: ...
- Economy: ...
- Map Control: ...

## Least Intuitive Concepts
- Unit Formations: ...
- Strategic Planning: ...
- Counter-play: ...

- In the fourth paragraph, you may want to provide some examples of why people love the game 0ad, such as the historical accuracy, the moddability, the online community, the graphics, and the story. For example, you could write:

# Why do people love the game 0ad?
People love the game 0ad for many reasons. Some of them are:

- Historical Accuracy: The game features historical civilizations and units that are based on extensive research and documentation. The unit and building names are shown in the original language of the civilization they belong to, and they are also translated into the language in which the user is playing the game. ²
- Moddability: The game is highly moddable, allowing players to customize their experience and create new content. The game engine and source code are available under free licenses, and the game art and music are available under a Creative Commons license.
- Online Community: The game has a vibrant online community, with players helping each other out and sharing tips and strategies. The game also has an active development team that listens to feedback and suggestions from players.
- Graphics: The game has beautiful graphics that create a realistic and immersive atmosphere. The game features dynamic lighting and shadows, realistic water effects, detailed terrain textures, and smooth animations.
- Story: The game has a deep and engaging story that spans across different historical periods and civilizations. The game allows players to relive or rewrite the history of ancient civilizations, each depicted at their peak of economic growth and military prowess.

I hope these suggestions are helpful. If you need more help with your text, please let me know."

 

[Stan`]

Bing AI uses chatGPT

[real_tabasco_sauce]

hahaha no way. Thats such a bing thing to do.

[borg-]

I asked chatgpd if he could teach me some basic Unity code like moving the units by keyboard and mouse in and he put the code complete, without bugs. Something similar is not possible for 0 a.d. @Stan`?

[Stan`]

6 minutes ago, borg- said:

I asked chatgpd if he could teach me some basic Unity code like moving the units by keyboard and mouse in and he put the code complete, without bugs. Something similar is not possible for 0 a.d. @Stan`?

ChatGPT has the entire code for 0 A.D. (2021 code at least) in his memory. So in theory it can. I guess the issue is that it did not ingest our forums. But you can give it some code from 0 A.D. and ask him about it  

[seeh]

 

 

# What are the most famous mods for the game 0 A.D. ?

and the answer is ( i never heard of them) :

Most Popular Mods for 0 A.D.
0 A.D. is an open source game that allows users to create and share their own mods. Here are some of the most popular mods for the game:

Rise of Rome - A mod that adds new civilizations, units, and buildings to the game.
Aurora - A mod that adds new civilizations, units, buildings, and maps to the game.
Ascension - A mod that adds new civilizations, units, buildings, and maps to the game.
Persian Wars - A mod that adds new civilizations, units, buildings, and maps to the game.
Age of Empires - A mod that adds new civilizations, units, buildings, and maps to the game.
Napoleonic Era - A mod that adds new civilizations, units, buildings, and maps to the game.
Celtic Nations - A mod that adds new civilizations, units, buildings, and maps to the game.
Imperium - A mod that adds new civilizations, units, buildings, and maps to the game.
Age of Chivalry - A mod that adds new civilizations, units, buildings, and maps to the game.
Tokens used: 427

[real_tabasco_sauce]

I feel like that shouldn't be too hard of a prompt, but I guess forum discussions and modio statistics are not available to ChatGPT

[seeh]

26 minutes ago, real_tabasco_sauce said:

I guess forum discussions and modio statistics are not available to ChatGPT

"The data available on ChatGPT is limited to the year 2021. The training of the model stopped in the year 2021"

[Stan`]

13 hours ago, seeh said:

 

 

# What are the most famous mods for the game 0 A.D. ?

and the answer is ( i never heard of them) :

Most Popular Mods for 0 A.D.
0 A.D. is an open source game that allows users to create and share their own mods. Here are some of the most popular mods for the game:

Rise of Rome - A mod that adds new civilizations, units, and buildings to the game.
Aurora - A mod that adds new civilizations, units, buildings, and maps to the game.
Ascension - A mod that adds new civilizations, units, buildings, and maps to the game.
Persian Wars - A mod that adds new civilizations, units, buildings, and maps to the game.
Age of Empires - A mod that adds new civilizations, units, buildings, and maps to the game.
Napoleonic Era - A mod that adds new civilizations, units, buildings, and maps to the game.
Celtic Nations - A mod that adds new civilizations, units, buildings, and maps to the game.
Imperium - A mod that adds new civilizations, units, buildings, and maps to the game.
Age of Chivalry - A mod that adds new civilizations, units, buildings, and maps to the game.
Tokens used: 427

They don't exist. That's why

[ChronA]

Something to keep in mind with language model chatbots is the primacy of context.

While the word intelligence is applicable to the emergent behavior that tends to arise from large language models, never forget that what they actually are is pattern recognition and extension engines. They work by looking at a block of text and predicting the most likely sequence of letters that comes next. It's just that when you train this system on trillions of pages of text and let it take billions of factors into account when matching the pattern it starts to capture super high-level patterns like rhetoric, logic, and culture, and ends up acting very like the minds that produced all those training examples in the first place.

But the bot will always remain a pattern recognition and extension engine, and this has consequences for how prompts should be engineered. GPTs generate much more specific and accurate responses from longer and more detailed prompts. Even if you are only regurgitating background information the model already knows, the extra material will help it to latch on to a more specific response pattern and produce a more sophisticated and topical answer. In some cases this will make hallucinations like the fake mod list go away.

Additionally, these chatbots exhibit a behavior I call mirroring. Because they are pattern matching systems, they tend to pick up and copy stylistic and behavioral features from the users prompts. For example, if you don't put much effort into a prompt or prompts the LM won't put much care into its answer. The bot will spit out short and sloppy answers that ignore parts of the prompt and make things up to get its point across. Or if you write in a highly technical style the bot will adjust its word choices to try to mimic it, even if you give it direction to use a different style. (This is really annoying if you are trying to get the bot to rewrite a composition in a new style.) Or if you are rude or demeaning the bot will turn obstinate and passive aggressive. They mostly have really study reinforcement training now that prevents them from becoming openly hostile, but they will start ignoring directions and basic logic to keep the combative vibe going.

Edited May 19, 2023 by ChronA

[seeh]

lol

F: "In wich language is autoCiv written and howto debug (tools, tricks, commands) it?"

ChatGPT: AutoCiv is written in C++ and debugging it can be done using the following tools, tricks, and commands:

Tools: Visual Studio Code,GDB,Valgrind
Tokens used: 341

[Vantha]

On 18/05/2023 at 5:57 PM, borg- said:

I asked chatgpd if he could teach me some basic Unity code like moving the units by keyboard and mouse in and he put the code complete, without bugs. Something similar is not possible for 0 a.d. @Stan`?

 

On 18/05/2023 at 6:04 PM, Stan` said:

ChatGPT has the entire code for 0 A.D. (2021 code at least) in his memory. So in theory it can. I guess the issue is that it did not ingest our forums. But you can give it some code from 0 A.D. and ask him about it  

It would be possible to fine-tune GPT4 or any other LLM on 0ad's code. If done correctly it could probably give working pieces of code from just a simple prompt. But it's a lot of work and I'm not sure if these LLMs are already good enough.

[ChronA]

51 minutes ago, Vantha said:

It would be possible to fine-tune GPT4 or any other LLM on 0ad's code.

While I don't have first hand experience, I'm pretty certain the 0ad code base is not remotely a large enough training set to meaningfully realign a model like GPT-4. If you really wanted to do what's being proposed I think you would need to construct a procedural reinforcement learning regime. Have the model generate huge amounts code for the game: many thousands of times the size of the actual code base, then use heuristics and testing to pick out the parts that work and use those to re-train the model.

That would be a huge project, and probably not practical for a small hobbyist community that can't even add highly requested features to the game with any consistency. I'd also question if it is even necessary. What I don't think people appreciate is that when it comes to pattern recognition, for many tasks GPT-4 is already nearly (and in some cases actually) super human. It can pick up on patterns, logical relationships, and implications much faster than a human being would, often from a single example or a half formed suggestion.

Admittedly, often times the conclusions it draws can be wrong, and many people will point to that as a sign the bot is stupid, but they are missing the forest for the trees. It is the user's job to guide the bot to the right patterns by providing context and instructions, and in time skillful users will distinguish themselves by their ability to efficiently leverage the bot's proclivities to get remarkable results. (AI whisperers if you will.) In short, the bot being stupid is operator error. And conversely, when it gets something super specific right the first time using pure intuition, it is downright spooky. And that is an experience I've had with GPT-4 a lot more often than with humans.

The bottom line is you won't need to retrain these models to produce useful code for projects 0ad. If say you wanted it to rewrite part of the Attack module, just give it the example of the old Attack module as part of your task prompt. So long as you provide clear directions it will probably produce working code without any difficulty.

[seeh]

Ask GPT3 (ChatGPT) to get some more wrong (but maybe interesting) answers  :

Path finding in the game 0ad works by using a combination of A* (A-Star) and Dijkstra algorithms. A* is an algorithm used for finding the shortest path between two points on a graph. Dijkstra is an algorithm used for finding the shortest path between two points on a graph.
Tokens used: idk

__________________________________

 

Yes, it is a good idea to update the graph of the path finder in the game 0ad less often in order to reduce the power consumption. This will help to conserve resources and make the game more efficient. You can do this by decreasing the frequency of updates or by decreasing the number of nodes that are updated. Additionally, you can optimize the path finding algorithm to make it more efficient and reduce the power consumption.

Tokens used: 251

 

BTW: maybe interesting files:

source/simulation2/components/CCmpPathfinder_Common.h

source/simulation2/components/CCmpPathfinder.cpp

source/simulation2/components/CCmpAIManager.cpp

Yes, it is possible to reduce power consumption by rewriting the files you mentioned.

The most efficient way to do this would be to modify the algorithms used by the pathfinder. By making the pathfinding process slower, you can reduce the amount of power consumed.

For example,

• in CCmpPathfinder_Common.h, you could modify the A* algorithm to have a larger step size, which would make the pathfinding process slower.
• In CCmpPathfinder.cpp, you could modify the Dijkstra's algorithm to have a larger step size, which would also make the pathfinding process slower. Finally,
• in CCmpAIManager.cpp, you could modify the pathfinding algorithm to have a larger step size, which would again make the pathfinding process slower.

By making these modifications, you can reduce the power consumption of the game 0ad.

Tokens used: 412

 

Another wrong answer  :

The variable you are looking for is
maxStepSize
. It can be found in the
CCmpPathfinder_Common.h
 header file.

This variable is used to control the maximum step size for the A* pathfinding algorithm. Increasing the value of this variable will result in a larger step size and slower pathfinding process.

Tokens used: 258

 

 

Edited May 30, 2023 by seeh

[Vantha]

On 24/05/2023 at 1:42 PM, ChronA said:

It is the user's job to guide the bot to the right patterns by providing context and instructions

The whole point of LLMs is to, instead of humans having to adapt to the computer language("to code"), enable computers to adapt to human language. Right now when coding you need to "translate" human language into a computer/programming language. One of the goals of LLMs is to do that "translation step" for humans(not just in programming, but also in other areas) so everybody can directly interact with computers in human language.
And even though GPT 4 is very very advanced, it can't write large working pieces of code from just a simple prompt in "human language". Even GPT 4 is not good enough for this. Somebody with no coding expierience can't really use GPT 4 to write code right now. But in the not so distant future there will be LLMs that can do that.

[alre]

1 hour ago, Vantha said:

The whole point of LLMs is to, instead of humans having to adapt to the computer language("to code"), enable computers to adapt to human language. Right now when coding you need to "translate" human language into a computer/programming language. One of the goals of LLMs is to do that "translation step" for humans(not just in programming, but also in other areas) so everybody can directly interact with computers in human language.
And even though GPT 4 is very very advanced, it can't write large working pieces of code from just a simple prompt in "human language". Even GPT 4 is not good enough for this. Somebody with no coding expierience can't really use GPT 4 to write code right now. But in the not so distant future there will be LLMs that can do that.

not really, no.

[Vantha]

1 hour ago, alre said:

not really, no.

What do you mean no?

[alre]

22 hours ago, Vantha said:

What do you mean no?

there are AIs ment to help developers coding, but they are only far relatives to chatgpt. besides, coding is not a "translation" of human language into machine language, code has a rigour that natural languages lack, and if you can't code some feature, chances are you can't express what you want from that feature in natural language either.

the only thing LLMs can do, and IMO will always be able to do, is to give very generic indications, or creating very generic pieces of content, that you can either spam as they are or use as a stub for later rework. this applies to code as well as to other text.

models that generate images are somehow different because pictures are fuzzier and AI works better there, the hand of a painter is not as precise as the pen of a writer or the hands of a programmer.

[Vantha]

On 01/06/2023 at 8:51 PM, alre said:

there are AIs ment to help developers coding, but they are only far relatives to chatgpt. besides, coding is not a "translation" of human language into machine language, code has a rigour that natural languages lack, and if you can't code some feature, chances are you can't express what you want from that feature in natural language either.

Of course, this "translation" I described is not really translation. That's why I put it in quotation marks. But the concept of computers following commands in human language is a very important idea/purpose behind Large language models and everything they can do in general, not just programming. (In fact, when training an LLM you can't really prevent it from learning how to code)

 

On 01/06/2023 at 8:51 PM, alre said:

the only thing LLMs can do, and IMO will always be able to do, is to give very generic indications, or creating very generic pieces of content

Right now, yes, but there are already the first programs that can code (very simple) apps or websites for you. And these tools will just get better and better in the future. LLMs (or programs that use LLMs) will eventually become powerful enough to write very specific and complicated pieces of code and even be able to code entire programs like 0ad from one simple prompt.

Edited June 7, 2023 by Vantha

[Freagarach]

31 minutes ago, Vantha said:

to (...) very specific and complicated pieces of code (...) like 0ad from one simple prompt

Already possible:

git clone https://gitlab.com/0ad/0ad.git

Spoiler

Disclaimer: this is a joke.

 

[ChronA]

On 01/06/2023 at 2:51 PM, alre said:

The only thing LLMs can do, and IMO will always be able to do, is to give very generic indications, or creating very generic pieces of content, that you can either spam as they are or use as a stub for later rework. this applies to code as well as to other text.

This is a popular interpretation, but it's wrong, even as it pertains to the facts on the ground right now. The reason so many bright people are so excited about AI is that the most recent generation of LLMs have started exhibiting an emergence phenomenon that we have never seen before. The purpose of LLMs is to detect and extend patterns in natural language data; and in the last year they have started honing in on a previously unseen pattern in the relationships between words that behaves very much like a human consciousness.

LLMs are not supposed to create original ideas that are not represented in their training data. They are not supposed to be able to learn from experiences. And they are not supposed to be able to plan for unobserved contingencies, yet I have seen GPT-4 do all of these things. That should be our first clue that we are no longer dealing with language models. Rather the LM is a substrate, like the grey matter in our skulls or the transistors in a computer chip, and it is producing an emergent intelligence or even a mind with an original set of capabilities.

If you want my opinion, what it seems like they've actually managed to do is accidentally teach the language models how to do Turing-complete, arbitrary symbolic reasoning, which means the mind is self-generating just like ours. And that means it is capable of literally anything now. The only trick is finding the right contextualized stimulus to prompt a topical response. Admittedly for the moment it is a very hard trick because the reasoning pattern is fragile and strongly constrained by memory limitations, but that is going to change quickly.

Bottom line is that LLMs are not just spitting out generic regurgitations of their training data anymore. They are fully capable of generating text and code that is creative, specific, insightful, and effective. Therefore those of us who make our livings doing pretty much the same thing need to get our heads out of the sand. Our ecological niche has just been poached.