AI and my projects

[exxos]

AI and my projects - said once, so I don't have to keep saying it

I've mentioned this on the forum plenty of times over the years, but it's been on my mind for the past few weeks what to do about it all, so it seems worth putting in one place so there's no confusion.

These days pretty much everything I work on uses AI for something. That could be code, firmware, debugging, documentation, test programs, web pages or just untangling a problem I've been stuck on. It's also things like searching for parts and part numbers and checking whether they're actually in stock. That job is incredibly tedious, and after doing it over and over for literally years, I simply despise it. I now use AI for it almost every single time. The design decisions, the hardware, the testing on real machines and the responsibility for the end result are still mine. If something doesn't work, that's on me, not the tools.

Going forward, I'm not going to label every project, post or release with "AI was used here". Just assume it was, because it almost certainly was. If that upsets anyone, then I apologise, but I'm not going to change how I work because of it.

I'm saying this partly because earlier this year a release of mine got picked up on an Atari scene site, and someone insisted it should carry an AI disclaimer. That ended up costing me the best part of a day sorting it out with the site owner, for something I had never hidden in the first place. The next day I asked myself why I even cared. Frankly, I don't anymore. I don't feel I owe anyone an explanation for what tools I use, any more than I'd explain which brand of soldering iron I used.

I'm already doing the work of about 20 people as it is. Trying to keep track of every little thing AI has or hasn't touched, and remembering to declare this or that to keep people happy, is just wasted time in my opinion. I have better things to do with my time.

A good example is GB7. It had effectively been shelved because of a double screen refresh bug, which would have taken an enormous amount of time to track down manually through all the code. Honestly, I couldn't face it. AI found the problem and fixed it. Without that, GB7 would most likely still be sitting on the shelf. But I'm not going to start logging every bug fix, line of code or tweak AI was involved in. That really would be a waste of everyone's time, mine included.

Why AI?

As I've said for years, I have far more work than one person can keep on top of. The store, the forum, the server, hardware development, firmware, support, the wiki and everything in between. I also have RSI, and AI takes a significant amount of the typing and repetitive work off my hands. That alone makes it a big win for me.

The other reason is simply that the community is small. Over the years I've asked for help on a lot of things, sometimes for literally years, and much of the time the answer was silence. That isn't a dig at anyone. Some people do help, and that has always been very much appreciated, and you know who you are. But the reality is that most of my questions and requests for help have gone unanswered. AI fills that gap. Without it, a lot of what I've released recently simply wouldn't exist, or would still be sitting on the bench half finished.

The Atari Music Network articles are a perfect example. I'd been asking for help with those for literally years, and nobody wanted the job. Recently around 40 articles were completely recovered, rebuilt, updated to a proper format and layout, and posted on the forum one a day. No doubt many of you saw them appear. AI did the recovery and the rebuilding, and even wrote the script that posted them on the forum. My part was detailing what needed doing, instructing it, and checking and verifying everything before it went live.

So AI is doing the jobs nobody else wanted to do. If anyone feels the need to complain about that, I'd gently point out that I don't really see why I should have to justify how a job got done when there was nobody else interested in doing it in the first place. I'd much rather spend my time getting things finished and into people's hands than defending how they were made.

Looking at other forums, this whole "was AI used?" debate is only going to get bigger as AI becomes part of everyday tools. Before long it will be in pretty much everything, and demanding a declaration on every project will make about as much sense as asking which compiler or text editor someone used. I'd rather not spend my time going round in circles on it, so this post is my answer, now and in the future.

[dad664npc]

Good on ya Chris! I can't be bothered with one other Atari Forum now as they have gone annal on AI. I echo your sentiment that AI is a tool. I could not have made the PiSTorm JIT advances without it. I also sympathise with your cry for help, I too have been there and feel for you. Good luck with your AI endeavours.

[Darklord]

A logical, no-nonsense approach, IMHO.

[prog99]

Never seen much in your output to offend.
Sadly theres a fair degree of poorly tested shovelware appearing recently which has poked the bear so to speak.

[exxos]

Good on ya Chris! I can't be bothered with one other Atari Forum now as they have gone annal on AI. I echo your sentiment that AI is a tool. I could not have made the PiSTorm JIT advances without it. I also sympathise with your cry for help, I too have been there and feel for you. Good luck with your AI endeavours.

And you're definitely making excellent progress with it as well !

I think possibly if we was a absolutely massive community then my opinion may be slightly different but the Atari scene is just so near non-existent these days anyway.. I don't think we all should start getting into "arguments" about if AI is used or not. I do understand all the pros and cons of it all but I think we all have better things to do at the end of the day.

None of those are exactly getting any younger. If I was in my 20s I would think nothing about spending the next 10 years learning assembly language or something like that.. As I fast approaching 50, that is basically the last thing on my mind these days.

[exxos]

A logical, no-nonsense approach, IMHO.

Time is precious at the end of the day.. Sucked into the latest Internet debate and then the next one after that is just taking away more important things to be done. The Internet can argue about it all they want.

[exxos]

Never seen much in your output to offend.
Sadly theres a fair degree of poorly tested shovelware appearing recently which has poked the bear so to speak.

Yeah that's the problem. How much testing all these software's have at the end of the day and how long they will be supported for and who you report bugs to etc is another huge problem.

I mean if you click on the tools icon in the bar at the top of the forum those are all tools which are created with AI. The server would not even be online if it wasn't for AI. Similar with my web store which AI fixed recently. And said earlier elsewhere, AI is now fixing the wiki. I'm just one person at the end of the day and just do not have the time to work for absolutely everything on my own.

I mean if people complained about using AI to fix the server we have really lost the plot at that point. If it wasn't for AI fixing the server then they would not have a platform to complain on anyway..

[exxos]

A follow up on how I actually use AI for my posts

I have always been open and honest about my use of AI, which is the whole reason this thread exists in the first place. But there seems to be some confusion creeping in, so to clear it up: AI does not write my posts. I write them, and AI rewords, formats and spellchecks what I have said. Yes, the end result looks like AI output, and technically it is, but the content is mine, and the original reply has still likely taken me an hour or more to put together in the first place.

In the past I did post some raw AI output, which landed me in trouble, so I do not do that any more. There may be the odd exception I have forgotten about, but as far as I can remember, I have not posted anything like that recently. These days most of what I post is based on the years worth of documented tests and updates from my own threads, which the AI has in its memory. So when it pulls information together, it is my own work it is drawing on, and it is only relevant to my own projects anyway.

Below are two simple screenshots showing how it works in practice. The first is what I actually typed about Greg's problem, typos and all, and the second is what got posted after the AI tidied it up. Same points, same questions, just readable. Greg's is only a very simple example, of course. A lot of my posts are much longer, and what I type off the top of my head often ends up in a weird order as I go back and forth between things. The AI pulls those thoughts together into something which flows more naturally and is less confusing for everybody.

greg1.PNG (233.25 KiB) Viewed 21 times

greg2.PNG (25.85 KiB) Viewed 21 times

Before anyone says it would be quicker to just type the reply myself than upload screenshots into an AI, that is not how it works. I use the AI as a documented journal for all the technical work which is ongoing, so it holds a technical reference of everything. What I post on the forum is probably only a couple of percent of what has actually happened that day, or over the past few days. All the technical context, the trial and error, the test results and the updates are held there, so I can easily look back when I forget something or need to check something. When I do want to post a reply, it already has the full work history and context to hand. I simply cannot remember thousands of tests in my head, which is why I use AI for this work now, and why technical conversations get uploaded into it and it helps with the replies as well.

I do check what gets posted, and I do have to correct things. But I am only human, and some things will get missed, exactly the same as if I had written the post myself. Bear in mind I am replying to people continuously, all day, every day, often until late at night, and it is exhausting. Mistakes are going to creep in either way, whether by my own hand or by me missing something, and I simply do not have the time or the energy to make every post bullet-proof.

As I have said before, I have RSI. Pulling together what I want to say, formatting it and spellchecking it takes a significant workload off my hands, and frees up time for more constructive things. Right from day one of this forum, part of the moderators' job was correcting the typos in my posts. I wanted as few typos as possible for our non-English-speaking friends, who would struggle to work out what the words were when a post was riddled with them. AI now does that job, which saves at least two people's work, and after 30+ years of correcting my own typos, it is repetitive work I just do not want any more either.

If people would rather not have the better formatting, that makes no odds to me. I can go back to posting big blocks of unformatted text full of typos, like I used to, or just post an update once a month or so. But then a lot of information would end up getting missed, which does not help anyone.

If people are going to get offended by bullet points, bold text and headings, then I think that is pretty sad really. But if people feel strongly enough about it, the replies will just be more boring blocks of text which are harder to read, and I think fewer people would bother reading them than a properly formatted post. Then again, if people will not read a post because they think it is LLM output, it is a bit of a no-win situation either way. I cannot please everyone, unfortunately.

At the end of the day, I am only doing what I have to in order to keep up and reduce my workload. I am writing forum posts, not peer-reviewed technical journals, even though some people seem to think I ought to be.

So if a post looks like it came from an AI, it is still me behind it. The AI just does the bit I used to dread, like having a spell checker that actually works, rather than the one on my old word processor that thought "Atari" was a spelling mistake.

-----------------------------------------------------------------------------------------------------------------------------------------------------------------------

And to prove the point, here is exactly the same post again, but with the headings and bullet points some people seem to object to. Same words, same points. Which one would you honestly rather read?

How I use AI for my posts

The short version

• AI does not write my posts. I write them, and AI rewords, formats and spellchecks them.
• The content is mine, and the original reply has often taken me an hour or more.
• I have always been open about this, which is why this thread exists.

Raw AI output

• I did post some raw AI output in the past, which landed me in trouble, so I do not do that any more.
• There may be the odd exception I have forgotten about, but not recently as far as I can remember.

Where the content comes from

• Years of documented tests and updates from my own threads.
• The AI acts as a technical journal for all my ongoing work, holding the context, trial and error, test results and updates.
• What I post is probably only a couple of percent of what actually happened that day.
• I cannot remember thousands of tests in my head, so this is how I keep track and look things back up.

Why it is not quicker to type it myself

• The AI already has the full work history to hand, so it is not a case of uploading a screenshot and starting from scratch.
• What I type off the top of my head often ends up in a weird order as I jump between things.
• The AI pulls it into something which flows and is less confusing for everybody.
• Greg's example above is a very simple one. Most posts are much longer.

Typos and checking

• I check what gets posted and correct things, but I am only human and some things will get missed, the same as if I wrote it myself.
• I reply to people all day, every day, often until late at night, and it is exhausting.
• I have RSI, so this takes a significant workload off my hands.
• From day one, part of the moderators' job was correcting my typos, so our non-English-speaking friends could actually work out what I meant.
• AI now does that, saving at least two people's work.

The alternatives

• Big blocks of unformatted text full of typos, like I used to post.
• Or one update a month, which means a lot of information gets missed.
• If people will not read a post because it looks like AI, and will not read walls of text either, it is a no-win situation. I cannot please everyone.

Bottom line

• I am only doing what I have to in order to keep up.
• These are forum posts, not peer-reviewed technical journals.
• If it looks like AI, it is still me behind it.