AI: The Rise of the Machines... Or Just a Lot of Overhyped Chatbots?

ShakyJake · Monday at 8:08 AM

ShakyJake said:
I'm experimenting with how well current coding agents can handle a demanding homebrew arcade conversion. Specifically, I'm working on a Commodore VIC-20 Pac-Man port that aims for a true one-to-one recreation: gameplay that matches the arcade as closely as possible, along with graphics and audio that stay as faithful as the hardware allows.

The VIC-20 is quite limited -- it has no sprite or bitmap support, so everything must be done with character graphics. To prepare, I ran a deep research report in ChatGPT focused on writing optimized 6502 assembly for the VIC-20 and on known techniques for extracting the most performance from the machine. A few Pac-Man conversions already exist (some of them quite good), so this should make for an interesting side-by-side comparison.

First pass, took about a day. Frustratingly, Claude REFUSED to work on this. Had to use Codex / GPT-6.1 Sol on xHigh. Video below is gameplay up to the first intermission.

I think the results are pretty damn good. It starts from Mark Burkley's C translation of the arcade program, adapted and compiled to native 6502 with LLVM-MOS, and uses a software renderer made from programmable characters (no hardware sprites). The full portrait display fits in 112×216 pixels; white maze walls and pellets keep the shared 8×16 color cells from turning into a patchwork. Original ghost rules, scoring, sounds, and the three intermissions are all retained. It currently holds about 48 FPS in normal play and 54 FPS when the ghosts are frightened, though flicker is still visible and is the next thing to tackle.

Kharzette · Monday at 8:37 AM

I've got some coding stuff that I want to do that I'm not particularly interested in doing myself (most game stuff I'd rather do on my own). Friend suggested Muse. Anyone signed up for that? I've been trying the last two days but it seems like it can't even send a goddamn signup mail.

sleevedraw · Monday at 8:52 AM

ShakyJake · Monday at 8:58 AM

Kharzette said:
I've got some coding stuff that I want to do that I'm not particularly interested in doing myself (most game stuff I'd rather do on my own). Friend suggested Muse. Anyone signed up for that? I've been trying the last two days but it seems like it can't even send a goddamn signup mail.

Not sure if there's a particular use case for Muse. It's pretty mediocre from what I've heard. Claude Opus 5.5 is the de-facto best for coding. ChatGPT Images 2.5 does some banger artwork, though.

TomServo · Monday at 9:37 AM

ShakyJake said:
First pass, took about a day. Frustratingly, Claude REFUSED to work on this. Had to use Codex / GPT-6.1 Sol on xHigh. Video below is gameplay up to the first intermission.

I think the results are pretty damn good. It starts from Mark Burkley's C translation of the arcade program, adapted and compiled to native 6502 with LLVM-MOS, and uses a software renderer made from programmable characters (no hardware sprites). The full portrait display fits in 112×216 pixels; white maze walls and pellets keep the shared 8×16 color cells from turning into a patchwork. Original ghost rules, scoring, sounds, and the three intermissions are all retained. It currently holds about 48 FPS in normal play and 54 FPS when the ghosts are frightened, though flicker is still visible and is the next thing to tackle.

Hey 8 bit guy!

Fucker · Tuesday at 2:03 PM

Fujifilm to boost tape production 5x as AI fuels archiving demand

Low costs and offline security mean magnetic tape still has a job.

arstechnica.com

Kajiimagi · Tuesday at 7:49 PM

Fucker said:
Fujifilm to boost tape production 5x as AI fuels archiving demand

Low costs and offline security mean magnetic tape still has a job.

arstechnica.com

makes sense, I recall reading a book about the piracy craze of music / video games and the only people that made money off that was the university that funded the mp3 research.

It seems you need Secret Clearance to get to your Nook library now but I think this is the book:

How Music Got Free

Stephen Witt.

Leadsalad · 2026-11-04T09:54:31-0500

Lanx · 2026-11-04T21:15:08-0500

Kajiimagi said:
makes sense, I recall reading a book about the piracy craze of music / video games and the only people that made money off that was the university that funded the mp3 research.

It seems you need Secret Clearance to get to your Nook library now but I think this is the book:

How Music Got Free
Stephen Witt.

i'm pretty sure my first mp3 was smells like teen spirit 4mb took 25mins to d/l on 28.8

ShakyJake · 2026-11-05T09:38:56-0500

ShakyJake said:
First pass, took about a day. Frustratingly, Claude REFUSED to work on this. Had to use Codex / GPT-6.1 Sol on xHigh. Video below is gameplay up to the first intermission.

I think the results are pretty damn good. It starts from Mark Burkley's C translation of the arcade program, adapted and compiled to native 6502 with LLVM-MOS, and uses a software renderer made from programmable characters (no hardware sprites). The full portrait display fits in 112×216 pixels; white maze walls and pellets keep the shared 8×16 color cells from turning into a patchwork. Original ghost rules, scoring, sounds, and the three intermissions are all retained. It currently holds about 48 FPS in normal play and 54 FPS when the ghosts are frightened, though flicker is still visible and is the next thing to tackle.

I’ve finished the current optimization pass on my Pac-Man conversion for the VIC-20. GPT 6.1 on xHigh handled the implementation and profiling under my direction, and I play-tested the changes.

A new renderer cuts most of the flicker from erasing and redrawing characters. Fourteen speed improvements raised the average redraw rate from 22 to nearly 34 FPS. I also refined the Pac-Man and frightened ghosts sprites and redrew the fonts.

The video shows a computer-controlled run using PacStrats patterns through nine boards and the third intermission. Gameplay is identical to the arcade!

I had a lot of fun with this one. Some people will say using a coding agent means you didn’t do anything, but guiding it to a result like this still takes real effort. It is not a one-sentence prompt.

Cad · 2026-11-05T11:21:16-0500

ShakyJake said:
I’ve finished the current optimization pass on my Pac-Man conversion for the VIC-20. GPT 6.1 on xHigh handled the implementation and profiling under my direction, and I play-tested the changes.

A new renderer cuts most of the flicker from erasing and redrawing characters. Fourteen speed improvements raised the average redraw rate from 22 to nearly 34 FPS. I also refined the Pac-Man and frightened ghosts sprites and redrew the fonts.

The video shows a computer-controlled run using PacStrats patterns through nine boards and the third intermission. Gameplay is identical to the arcade!

I had a lot of fun with this one. Some people will say using a coding agent means you didn’t do anything, but guiding it to a result like this still takes real effort. It is not a one-sentence prompt.

Do web-based Carrier Command from the Amiga and I'm in. Or mac.

Good Morning Smile GIF by DreamWorks Animation

Noodleface · 2026-11-05T12:11:06-0500

Or,.. hear me out....

EQ... But in a browser

ShakyJake · 2026-11-05T17:15:37-0500

Cad said:
Do web-based Carrier Command from the Amiga and I'm in. Or mac.

I guess the software companies complained loud enough. Claude will absolutely refuse to do anything it detects as reproducing a piece of software or game. I had GPT research the gameplay of Carrier Command, then took that research document and reworked it so it didn't explicitly name Carrier Command. Created a game design document off of that. Handed it to Claude and it immediately said nope, what you're describing is Carrier Command and I ain't doing it.

There has to be a way to jailbreak this. I mean, at what point does having an idea that's sorta close to something else mean AI just refuses to work on it?

I can create a space-themed version of it, however.

Fucker · 2026-11-05T17:29:53-0500

ShakyJake said:
I guess the software companies complained loud enough. Claude will absolutely refuse to do anything it detects as reproducing a piece of software or game. I had GPT research the gameplay of Carrier Command, then took that research document and reworked it so it didn't explicitly name Carrier Command. Created a game design document off of that. Handed it to Claude and it immediately said nope, what you're describing is Carrier Command and I ain't doing it.

There has to be a way to jailbreak this. I mean, at what point does having an idea that's sorta close to something else mean AI just refuses to work on it?

I can create a space-themed version of it, however.

I had Claude make a simple system utility a few weeks ago. I wanted to remake it a few days ago, and Claude wouldn't do it. They are going to put so many nannies and Can't Do's in their product that people will quit using it.

Cad · 2026-11-05T17:45:28-0500

ShakyJake said:
I guess the software companies complained loud enough. Claude will absolutely refuse to do anything it detects as reproducing a piece of software or game. I had GPT research the gameplay of Carrier Command, then took that research document and reworked it so it didn't explicitly name Carrier Command. Created a game design document off of that. Handed it to Claude and it immediately said nope, what you're describing is Carrier Command and I ain't doing it.

There has to be a way to jailbreak this. I mean, at what point does having an idea that's sorta close to something else mean AI just refuses to work on it?

I can create a space-themed version of it, however.

I mean if you've done all that, just make it a space game where you are confined to the single plane (aside from the fighters which can fly, they can go 3d) and have Claude make up some lore reason that the main vehicles act like ships and tanks rather than airplanes. Good to go!

Thanks man!

Noodleface · 2026-11-05T19:52:47-0500

Fucker said:
I had Claude make a simple system utility a few weeks ago. I wanted to remake it a few days ago, and Claude wouldn't do it. They are going to put so many nannies and Can't Do's in their product that people will quit using it.

You have to keep in mind who is paying the bills. Those $100/200 a month unlimited plans ain't making them shit. Those big companies paying for enterprise API usages are what's making them money. It's no surprise they'd side with the big dogs.

That said I wholeheartedly agree. They'll quit using anthropic and move on to the next AI company with better models.

Control · 2026-11-05T20:09:12-0500

ShakyJake said:
I guess the software companies complained loud enough. Claude will absolutely refuse to do anything it detects as reproducing a piece of software or game. I had GPT research the gameplay of Carrier Command, then took that research document and reworked it so it didn't explicitly name Carrier Command. Created a game design document off of that. Handed it to Claude and it immediately said nope, what you're describing is Carrier Command and I ain't doing it.

There has to be a way to jailbreak this. I mean, at what point does having an idea that's sorta close to something else mean AI just refuses to work on it?

I can create a space-themed version of it, however.

Do it incrementally. First, let's make a spaceship fly around, now lets make it shoot, now let's add things to shoot at, etc.

ShakyJake · 2026-11-05T20:18:58-0500

Control said:
Do it incrementally. First, let's make a spaceship fly around, now lets make it shoot, now let's add things to shoot at, etc.

Ugh. that would be painful. But that's not a bad idea.

Fucker · 2026-11-05T20:19:36-0500

Noodleface said:
You have to keep in mind who is paying the bills. Those $100/200 a month unlimited plans ain't making them shit. Those big companies paying for enterprise API usages are what's making them money. It's no surprise they'd side with the big dogs.

That said I wholeheartedly agree. They'll quit using anthropic and move on to the next AI company with better models.

I know who their main users are, but they are playing with fire across the whole spectrum. Imagine working on a big project one day, and Claude decides it is out of bounds the next day. This only has to happen once before a company would move on. Trust is an issue with Anthropic, and it is a growing issue.

Control · 2026-11-05T20:29:08-0500

ShakyJake said:
Ugh. that would be painful. But that's not a bad idea.

Yeah, it's kind of painful, but really, building anything this way and very thoroughly testing at each step is the best way to keep it from making slop. Alternatively, Codex is still happy to explicitly reverse engineer (at least as of last week) and china models probably always will be (at least for non-chinese ip).

Search

AI: The Rise of the Machines... Or Just a Lot of Overhyped Chatbots?

ShakyJake

Kharzette

Watcher of Overs

sleevedraw

Revolver Ocelot

ShakyJake

TomServo

Fucker

Log Wizard

Fujifilm to boost tape production 5x as AI fuels archiving demand

Kajiimagi