code generated by ChatGPT

[ara]

Here is then answer ChatGPT gave when I asked for a window with the text "hello world" in red and when the user clicks on the text, it should play a sine wave sound with 200kHz (yes, that was my mistake). Fascinating how it invents new OS functions

#include <stdio.h>
#include <proto/intuition.h>
#include <proto/graphics.h>
#include <proto/exec.h>
#include <proto/dos.h>

#define WIDTH 300
#define HEIGHT 200

struct Window *win;

void cleanup() {
  if (win) CloseWindow(win);
  CloseLibrary(IntuitionBase);
  CloseLibrary(GfxBase);
  CloseLibrary(DOSBase);
}

void handle_click() {
  // create a sine-wave audio sample with 200KHz frequency
  int samples = 44100;
  int16_t data[samples];
  for (int i = 0; i < samples; i++) {
    data[i] = (int16_t)(32767.0 * sin(2.0 * M_PI * i * 200000.0 / 44100.0));
  }

  // open the audio device
  struct IOAudio *audioIO = (struct IOAudio*)AudioIOInit();
  if (!audioIO) return;

  // set the audio format
  audioIO->ioa_Request.io_Command = ADCMD_ALLOCATE;
  audioIO->ioa_DataFormat = ADFMT_S16_NE;
  audioIO->ioa_Frequency = 44100;
  audioIO->ioa_Volume = 0x10000;
  audioIO->ioa_Channels = 1;
  audioIO->ioa_MixFreq = 44100;
  if (OpenDevice(AUDIONAME, 0, (struct IORequest*)audioIO, 0) != 0) {
    AudioIOFree(audioIO);
    return;
  }

  // play the audio sample
  audioIO->ioa_Request.io_Command = CMD_WRITE;
  audioIO->ioa_Request.io_Data = data;
  audioIO->ioa_Request.io_Length = samples * 2;
  audioIO->ioa_Request.io_Offset = 0;
  AO_PlaySamples(audioIO);

  // wait for the audio sample to be played
  while (audioIO->ioa_Request.io_Message.mn_Node.ln_Type != NT_MESSAGE) {
    Wait(1 << audioIO->ioa_Request.io_Message.mn_ReplyPort->mp_SigBit);
  }

  // close the audio device
  CloseDevice((struct IORequest*)audioIO);
  AudioIOFree(audioIO);
}


int main() {
  if (!(IntuitionBase = OpenLibrary("intuition.library", 39))) return 1;
  if (!(GfxBase = OpenLibrary("graphics.library", 39))) return 1;
  if (!(DOSBase = OpenLibrary("dos.library", 39))) return 1;
  atexit(cleanup);

  win = OpenWindowTags(NULL,
    WA_Left, 0, WA_Top, 0,
    WA_Width, WIDTH, WA_Height, HEIGHT,
    WA_Title, (ULONG)"Hello world",
    WA_IDCMP, IDCMP_CLOSEPATTERN | IDCMP_MOUSEBUTTONS,
    WA_Flags, WFLG_CLOSEGADGET | WFLG_DRAGBAR | WFLG_DEPTHGADGET | WFLG_ACTIVATE,
    WA_Gadgets, NULL,
    TAG_DONE);
  if (!win) return 1;

  SetRast(win->RPort, 0x000000FF); // set the text color to red
  Move(win->RPort, 20, 20);
  Text(win->RPort, "Hello world", strlen("Hello world"));

  struct IntuiMessage *msg;
  ULONG msg_class;
  while (1) {
    WaitPort(win->UserPort);
    while (msg = (struct IntuiMessage*)GetMsg(win->UserPort)) {
      msg_class = msg->Class;
      ReplyMsg((struct Message*)msg);
      if (msg_class == IDCMP_MOUSEBUTTONS) {
        // handle click on the text
        handle_click();
      } else if (msg_class == IDCMP_CLOSEPATTERN) {
        break;
      }
    }
    if (msg_class == IDCMP_CLOSEPATTERN) break;
  }

  return 0;
}

[Hedeon]

Wow!

What was the exact question you asked?

[ara]

I don't remember anymore the exact words, but it was similar to this: "Write a program in C for AmigaOS to open a window that shows the text "Hello World" in red and when you click on the text it plays a sine wave audio sample with 220KHz [sic]".

I first did a query without the audio sample part and the produced code even checked whether the mouse click was inside the bounds of the "Hello world" text.

It is very good at producing things that look correct at first glance. I am wondering where it got the non-existing stuff like IDCMP_CLOSEPATTERN from.

[Octopus66]

Looks quite amazing!

[jotd]

Amazing but non functional and fake.

ChatGPT please make me some chinese 060 cpus please

[BastyCDGS]

Originally Posted by ara

It is very good at producing things that look correct at first glance. I am wondering where it got the non-existing stuff like IDCMP_CLOSEPATTERN from.

Well, the AI is just, like their human counterparts, learning Amiga development. It's actually "funny" that the pattern of errors is very similar to human behavior.

Was the name of the damn function AllocMem or AllocMemory you know...

Also don't forget that ChatGPT doesn't try to run WinUAE, compile the code and actually test it and therefore can't start the continuous development process involving fixing bugs over time.

Considering those limitations, this result is really amazing!

EDIT:

Did you try what happens, when you reply to ChatGPT something like this:
"Your example is pretty good, however IDCMP_CLOSEPATTERN does not exist. Maybe you could try again?"

[pandy71]

This show that "AI" is not aware of the concept Nyquist frequency and Nyquist rate i.e. there is no I behind A except capability of mimicry behind words... This remains me modern media where people pretending they are technical experts when mixing pentagram with pantograph or modern politicians...

[tygre]

Hi!

Interesting experiment

Also, the bot is not learning: "teach" it something in one session, for example that "AmiModRadio allows downloading modules from Web sites to play them on your local Amiga", create a new session, and ask it about AmiModRadio...

Definitely no "I"

Cheers!

[Olaf Barthel]

Originally Posted by ara

Here is then answer ChatGPT gave when I asked for a window with the text "hello world" in red and when the user clicks on the text, it should play a sine wave sound with 200kHz (yes, that was my mistake). Fascinating how it invents new OS functions

This is pretty good, actually, except for the problem of reproducing functions which do not exist in the code. I have seen other synthesized Amiga code which did not bother about cleaning up after the resources allocated were no longer used. The code also looked a lot like it was written for a Unix-like system, with "return 1;" used in place of a more appropriate "return RETURN_ERROR;", etc.

That said, the function calls which are supposed to play back the sound and free the audio resources likely came from a program which contains these in a separate file. ChatGPT combines information to produce sequences of plausible items in context. Hence, ChatGPT might not have picked them up and reproduced them because they were not sufficiently plausible.

[malko]

Originally Posted by BastyCDGS

[...] Also don't forget that ChatGPT doesn't try to run WinUAE, compile the code and actually test it and therefore can't start the continuous development process involving fixing bugs over time.

Considering those limitations, this result is really amazing! [...]

You have a strange way of looking at things. Following your reasoning, everybody not compiling and testing his code is doing "amazing result".
ChatGPT should better reply that it doesn't know instead of writing such things (until it knows).

[tygre]

Hi all!

I was actually impressed by the code generated in answer to "Write a C program for AmigaOS to list all devices and volumes." over the week-end, it had generate a nice piece of code.

The problem is that it generates crappy code today. It's not consistent!

Cheers!

[giantclam]

Interesting watch -- [ Show youtube player ]

[BastyCDGS]

Hi malko!

Originally Posted by malko

You have a strange way of looking at things. Following your reasoning, everybody not compiling and testing his code is doing "amazing result".
ChatGPT should better reply that it doesn't know instead of writing such things (until it knows).

I meant this relative to what AI had been capable so far.

The truly amazing thing about ChatGPT is the language model, i.e. that you can talk to it almost like to another human (including complex speech) and still understanding what you expect from the AI.

However, in detail, the results often contain mistakes, so blind trust is never recommended.

Also remember that Amiga stuff is a niche and therefore is not that much training data for the AI available.

[gimbal]

Originally Posted by BastyCDGS

so blind trust is never recommended.

This is true in general. Faith maybe, blind faith no.

[BastyCDGS]

For all those, who don't have a ChatGPT account, you can test it out for free and without registration here (using GPT-3.5-Turbo engine):
https://seoschmiede.at/chatgpt

Please note, that while the page is in german, you can still ask ChatGPT english and other language questions.

If you want to test without context, you can input your question to ChatGPT at: "Was ist deine Frage?" and click the green button right to it "Frage stellen ->" in order to submit your query.

If you want to try with context, click on:
Neuer Chatbot (mit Verlauf) ->

Then type your query in the left field where the button "Los ->" is and click "Los ->" to submit the query.

Please note that the free version is limited to processing 1000 tokens at max per query.

Also note that using ChatGPT is NOT free for the site owner, so please do not ask way too many questions.

Have fun!

Example:

Question from me:
Simulate creation of a virtual AD&D character, using your PRNG to simulate the dice rolls, choosing a name, gender, race and class yourself! Show the virtual character sheet! Now begin simulating a virtual combat against a Beholder while explaining the AD&D combat rules step by step , also showing virtual PRNG dice results. Show a summary of all involved entities at begin of each combat round!

Reply from ChatGPT:
Sure, I can assist you with that. Let's start by creating a virtual AD&D character.

Name: Thorgar
Gender: Male
Race: Dwarf
Class: Fighter

Now, let's use my PRNG to simulate the dice rolls for Thorgar's abilities. Here are the results:

Strength: 10
Dexterity: 16
Constitution: 12
Intelligence: 9
Wisdom: 13
Charisma: 8

Based on these results, Thorgar's sheet looks like this:

Character sheet:
------------------
Name: Thorgar
Gender: Male
Race: Dwarf
Class: Fighter
------------------
Ability scores:
Strength: 10
Dexterity: 16
Constitution: 12
Intelligence: 9
Wisdom: 13
Charisma: 8
------------------
Hit points: 10

Now, let's simulate a virtual combat against a Beholder. The Beholder is a fearsome creature with multiple eye-stalks, each capable of powerful magical attacks. Thorgar will need to be careful.

Here's a summary of all the entities involved:

Thorgar, Dwarf Fighter, Hp: 10
Beholder, Monster, Hp: 45

Round 1:
Thorgar rolls initiative: 15 (10 + 5 from dexterity)
Beholder rolls initiative: 10

Thorgar goes first. As a fighter, Thorgar has two attacks per round. He rolls to attack with his warhammer: 13 (7 + 3 from strength + 3 from proficiency). The Beholder's armor class is 18, so Thorgar misses.

The Beholder retaliates by using its Eye Rays. It randomly chooses three eye rays to shoot at Thorgar. I'll use my PRNG to determine which eye rays are used.

Eye ray 1: Disintegration Ray (2d6 damage if it hits)
Eye ray 2: Sleep Ray (Thorgar must make a Wisdom saving throw or fall asleep)
Eye ray 3: Paralyzing Ray (Thorgar must make a Dexterity saving throw or be paralyzed)

I'll use my PRNG to simulate the accuracy and effects of each ray.

Eye ray 1: 17 (against Thorgar's armor class of 15), hits. Damage: 7 (2d6).
Thorgar takes 7 damage and his hit points are reduced to 3.

Eye ray 2: 12 (against Thorgar's Wisdom saving throw DC of 13), misses.
Thorgar is unaffected by the Sleep Ray.

Eye ray 3: 21 (against Thorgar's Dexterity saving throw DC of 14), hits.
Thorgar fails the saving throw and is paralyzed for one round.

At the end of round 1, here's the summary of entities:

Thorgar, Dwarf Fighter, Hp: 3 (Paralyzed)
Beholder, Monster, Hp: 45

Round 2:

Thorgar is paralyzed and cannot act this round. He'll need to make a saving throw at the end of his next turn to see if he can break free.

The Beholder takes advantage of Thorgar's vulnerability and uses three Eye Rays again.

Eye ray 1: 11, misses.
Eye ray 2: 16, hits. Thorgar must make a Wisdom saving throw or fall asleep.
Eye ray 3: 19, hits. Thorgar must make a Dexterity saving throw or be paralyzed.

Using my PRNG to determine effect:

Eye ray 2: 9, Thorgar fails his saving throw and falls asleep.
Eye ray 3: 23, Thorgar fails his saving throw and is paralyzed for another round.

At the end of round 2, here's the summary of entities:

Thorgar, Dwarf Fighter, Hp: 3 (Asleep and Paralyzed)
Beholder, Monster, Hp: 45

This combat looks pretty tough for Thorgar. He's been hit by two different Eye Rays and his hit points are getting low. It's possible that he could fall asleep soon, making him an easy target for the Beholder.

I hope this helps to explain the AD&D combat rules and how to use a PRNG to simulate dice rolls. If you have any other questions, feel free to ask!

Second try, almost same question:
Simulate creation of a virtual AD&D level 17 character, using your PRNG to simulate the dice rolls, choosing a name, gender, race and class yourself! Show the virtual character sheet! Now begin simulating a virtual combat against a Beholder while explaining the AD&D combat rules step by step , also showing virtual PRNG dice results. Show a summary of all involved entities at begin of each combat round!

Reply from ChatGPT:
Sure, I can definitely help you with that. Let's create a level 17 AD&D character using my PRNG to simulate the dice rolls.

Name: Arinelle
Gender: Female
Race: Half-Elf
Class: Cleric

Now, let's use the PRNG to simulate the ability scores for Arinelle. Here are the results:

Strength: 15
Dexterity: 13
Constitution: 16
Intelligence: 12
Wisdom: 18
Charisma: 10

Based on these results, Arinelle's character sheet looks like this:

Character sheet:
------------------
Name: Arinelle
Gender: Female
Race: Half-Elf
Class: Cleric
------------------
Ability scores:
Strength: 15
Dexterity: 13
Constitution: 16
Intelligence: 12
Wisdom: 18
Charisma: 10
------------------
Hit points: 116

Now, let's simulate a virtual combat against a Beholder. This Beholder will be even tougher than the one in the previous simulation, so Arinelle will need to be at her best.

Here's a summary of all the entities involved:

Arinelle, Half-Elf Cleric, Hp: 116
Beholder, Monster, Hp: 80

Round 1:
Arinelle rolls initiative: 18 (13 + 5 from dexterity)
Beholder rolls initiative: 12

Arinelle goes first. As a cleric of level 17, Arinelle has three attacks per round. She rolls to attack with her mace: 33 (18 + 5 from strength + 10 from proficiency). The Beholder's armor class is 23, so Arinelle misses.

The Beholder retaliates by using its Eye Rays. It randomly chooses three eye rays to shoot at Arinelle. I'll use my PRNG to determine which eye rays are used.

Eye ray 1: Petrification Ray (Arinelle must make a Constitution saving throw or turn to stone)
Eye ray 2: Fear Ray (Arinelle must make a Wisdom saving throw or be frightened)
Eye ray 3: Enervation Ray (2d8 damage if it hits)

I'll use my PRNG to simulate the accuracy and effects of each ray.

Eye ray 1: 15 (against Arinelle's Constitution saving throw DC of 19), misses.
Arinelle is unaffected by the Petrification Ray.

Eye ray 2: 20 (against Arinelle's Wisdom saving throw DC of 18), hits.
Arinelle is frightened until the end of her next turn.

Eye ray 3: 27 (against Arinelle's armor class of 18), hits. Damage: 10 (2d8).
Arinelle takes 10 damage and her hit points are reduced to 106.

At the end of round 1, here's the summary of entities:

Arinelle, Half-Elf Cleric, Hp: 106 (Frightened)
Beholder, Monster, Hp: 80

Round 2:

Arinelle is frightened and must use her action to move as far away as possible from the Beholder.

The Beholder decides to move closer to Arinelle and use three Eye Rays.

Eye ray 1: Telekinesis Ray (Arinelle must make a Strength saving throw or be moved 30 feet in a random direction)
Eye ray 2: Paralyzing Ray (Arinelle must make a Dexterity saving throw or be paralyzed)
Eye ray 3: Death Ray (Arinelle must make a Dexterity saving throw or take 10d10 damage)

I'll use my PRNG to simulate the accuracy and effects of each ray.

Eye ray 1: 22 (against Arinelle's Strength saving throw DC of 18), hits.
Arinelle fails her saving throw and is moved 30 feet in a random direction.

Eye ray 2: 11 (against Arinelle's Dexterity saving throw DC of 19), misses.
Arinelle is unaffected by the Paralyzing Ray.

Eye ray 3: 29 (against Arinelle's Dexterity saving throw DC of 19), hits.
Arinelle fails her saving throw and takes 10d10 damage.
I'll use my PRNG to simulate the damage from the Death Ray: 54.

Arinelle takes 54 damage and her hit points are reduced to 52.

At the end of round 2, here's the summary of entities:

Arinelle, Half-Elf Cleric, Hp: 52 (Moved 30 feet, Frightened)
Beholder, Monster, Hp: 80

This combat is even tougher than the previous one, and it's clear that Arinelle will need some luck if she wants to win. She's been hit by two different Eye Rays and her hit points are getting low. She'll need to use her spells wisely if she wants to stay alive.

I hope this helps to explain the AD&D combat rules and how to use a PRNG to simulate dice rolls. Let me know if you have any other questions.

[lesta_smsc]

ChatGPT

Please crack x game protection, thank you kindly. This will put @Stingray out of a job lol.

[meynaf]

Thanks for the link, i could finally test it !

Now i know what they mean by "confidently wrong". I asked it if it knew places where it could be accessed without having to create an account and it only listed social media and openAI's website where it's obviously not...

[BastyCDGS]

Hi!

Originally Posted by meynaf

Thanks for the link, i could finally test it !

Now i know what they mean by "confidently wrong". I asked it if it knew places where it could be accessed without having to create an account and it only listed social media and openAI's website where it's obviously not...

Don't forget that ChatGPT only has training data until somewhere in 2021, so everything happened after 2021 is unknown to it.

[meynaf]

Originally Posted by BastyCDGS

Don't forget that ChatGPT only has training data until somewhere in 2021, so everything happened after 2021 is unknown to it.

I asked it if it had access to internet to update itself and it replied yes

[BippyM]

The problem with the term Artificial Intelligence is misleading. To be intelligent it needs to be able to learn from input and mistakes. It does not so therefore it isn't intelligent. If it could learn from mistakes and input, and the input was shit... The intelligence would not work and with so many users giving input there would definitelyt be a lot that would intentionally give it incorrect information just to see what happens. It also doesn't understand right from wrong and there is no common sense.