AI for writing documentation

my man, we have a stack of RFCs printed out that we reference on a regular basis, you are barking up the wrong tree entirely. maybe you should spend time engaging with the actual material instead of delegating understanding to the slop machine.
 
Why exactly do you think I ask questions about RFC documents for? (and of course I ask such questions only after reading the complete thing, duh.) Ok, I'm really tired reading you. I'm here for only a few weeks, and I think I haven't seen a single message written by you that wasn't obvious bad faith. You're really not interesting.
 
huge thing that makes us not want to deal with any of this garbage is that most of its purveyors can't take a "no", and have to give us this pushy hard sell that sounds like it's coming from a wild-eyed zealot trying to get us hooked on his supply. it's creepy.
 
Last edited by a moderator:
You're the only zealot here, dumping your hate on every thread discussing AI, usually with low effort dismissive one-liners (which is rich, coming from someone complaining about slop - AI slop from genAI is indeed a problem, but not as big a problem than *your* slop). As all haters, whatever the object of their hate, you're completely defined by it, because hate is an obsession. You can't help pursuing it.

Now, don't worry, you won't have to suffer ever again seeing me disturbing your hate, because I'm adding you to my ignore list. I'm not interested in reading anything from you. I didn't leave social networks just to be exposed yet again to a constant stream of hate in a technical forum. I'm letting you know so that you don't find yourself ever waiting for a reply from me.
 
the extra cool part is that there's no way to prove if any of the code came from, say, GPL or proprietary sources, which, i'm sure, is going to result in 0 additional problems besides all of the other problems with slop code. :)
 
atax1a said:
the extra cool part is that there's no way to prove if any of the code came from, say, GPL or proprietary sources, which, i'm sure, is going to result in 0 additional problems besides all of the other problems with slop code. :)
Click to expand...
Besides, does an AI have an interest in anything at all? All the licenses/rules/permissions/laws are made by humans who have an interest in profit. I mean, a reasonable human mind can possibly conclude that an AI might be interested in RAM, electrical power, and related stuff - but that's only the case when there are humans around that are interested in stuff. On its own, an AI is nothing more than a pile of poisonous rare earths and metal, and can't possibly want or care for anything... not even GPL code.
 
What's the point to ask an AI to write something that you'll need te re-read and correct ? I still don't understand the hype around this thing, just losing time and giving money to tech merchants. Writing docs IS part of the coding/operating a software, we just lost it. Take any 80's software, any hardware came with a solid manual, so were the games too. There was no "AI" doing this, the manuals were excellent and the translations were perfect.
I document almost each line of the code I write, by hand, just like dinosaurs did. But if I write a variable is a integer, IT IS an integer ; I don't give a robot the risk to mistakenly write it's a string. Kind of very small errors that could break everything. "Real" human professionnal translators do know it, if you translate a big company contract to another language and you use only a small wrong word, that's a big hole that could ruin the company. An IA will not care about this.
 
Update:
I have been updating my md2mdoc(7) for bug-fixes and added more syntax handling. I say "my", here, because my repo is now 15+ (read: I stopped counting) ahead of the github version but that's not the point I'm really trying to make (if the consensus from most seems to be "Ai is easier", md2mdoc(7) will/should go back to being private because: "Do nothing which is of no use."). -e.g. In combination with mandoc(1) I can now convert simple (readable) markdown to html, ascii, ps, pdf, etc. and in batch. -i.e. a directory of markdown files can now be converted with a makefile and/or script.

I believe I have achieved at least a few major goals over the use of Ai (my programming abilities aside--at least I'm trying and if md2mdoc(7) goes/stays private this does not matter):

1. Lowering the technical debt (allowing users that do not have experiences with mandoc(1) macros the ability to write manpages). Which possibly ("arguable" at best) closes the workflow gap for documentation bug fixes and/or new documentation additions.

2. Opened the possibility for more/easier collaboration in "technical writing" (-e.g. markdown files--that can be later converted to manpages or other formats--can be viewed/edited on VCS web type interfaces). Which would allow more experienced writers the ability to guide in documentation writing.

While neither of the above goals, require md2mdoc(7) to be public -i.e. I can, at minimum, achieve these goals on a personal/small scale. However, I'd be interested in hearing from more experienced developers if these goals are "worthwhile" on larger scale and/or how are technical documents generated now (I understand professionals are now being "pressured/forced to use Ai", which is a different topic)? I can certainly see how an experienced manpage writer can/may easily write/review the manpage themselves but is there any value in offering the possibility for more collaboration in docs?

REF:
- This thread is essentially about choosing form over function and the "validity" of.
- I have been told my project will be referenced/used in one (1) other project so I'm also looking for advice/opinions (I'm feeling a little out of my depth/nervous). I do not know to what extent yet, though.
 
Talking myself off the ledge:
I guess the worst that could happen is that my program fails to produce a document (the end-user's machine or document is not harmed or modified) so I should not feel 'nervous'.
 
Here is an excellent article by Stefano Marinelli (the BSD Cafe guy) about problems he's having with customers who use AI to suggest fixes.
It doesn't grow, doesn't listen, doesn't update its mental model based on what you write back - and above all, it doesn't know what it doesn't know.
Click to expand...
I love that last part. Also,
That's why I'd like companies to consider that AI systems are stochastic machines, not experts. They can solve some problems, but there's a limit. There will always be a limit, at least with current technology, and we can't afford to ignore it. The damage risks far outweighing the "savings" generated.
Click to expand...
And here's a thought:
So I find myself wondering: if they're so convinced that AI is better than senior professionals, why don't they replace the bosses with AI? I'm fairly confident the decisions would be considerably better - and humans would end up exactly where they should be.
Click to expand...
 
drhowarddrfine said:
Here is an excellent article by Stefano Marinelli (the BSD Cafe guy) about problems he's having with customers who use AI to suggest fixes.

I love that last part. Also,

And here's a thought:
Click to expand...
I fear some people misunderstands my stance on AI. For instance, I agree with all the excerpts you have highlighted. Anyhow, I'd rather not elaborate because I always end up being bashed in my online head when I post about this topic and it's unpleasant.
 
drhowarddrfine
Thank you for the link, it is actually a good article indeed.

Personally I like this part:
The enormous problem with my work these days is the extreme confidence that certain companies project, replacing humans - even senior ones - with AI, with no right of appeal.
Click to expand...
I would just switch the word companies to people, because this a major problem for the world in general.

The main issue with "AI" technology is its name, artificial intelligence is obviously a straight lie chosen for marketing reason which should not be allowed.
Now we have people who really believe dealing with an intelligent tool.
 
> it doesn't know what it doesn't know.
Neither does anyone.

> is obviously a straight lie chosen for marketing reason which should not be allowed.
'allowed' by whom?
The main attraction, distraction
Got ya number than number than numb
Empty ya pockets, son, they got you thinking that
What ya need is what they selling
Make you think that buying is rebelling

From the theaters to malls on every shore
The thin line between entertainment and war
The frontline is everywhere
There be no shelter here
~ Rage Against the Machine
Click to expand...

---

"--'And even Stigand, the patriotic archbishop of Canterbury, found it
advisable'--"

"Found _what_?" said the Duck.

"Found _it_," the Mouse replied rather crossly; "of course, you know
what 'it' means."

"I know what 'it' means well enough, when _I_ find a thing," said the
Duck; "it's generally a frog or a worm. The question is, what did the
archbishop find?"
~ Alice in Wonderland.
Click to expand...
 
JohnK said:
Neither does anyone.
Click to expand...
Well, now I want to add something, because this is the first thing I thought also.

Currently, an AI doesn't know EITHER what it knows. It doesn't know ANYTHING. They don't think; they calculate. And they don't "evolve" or "learn" automatically; a new version has to be trained to improve the previous one. THEY ARE NOT THINKING MACHINES, they simulate thought. And they are very useful for certain things, but they are not what the marketing says. They don't think; they aren't your friends; they don't care if you continue the conversation or not. In fact, the whole "chat" is processed every time from the beginning, so there's nothing similar to "continuity" in them.

But for creating sh scripts to manipulate images in bulk, for example, they are very convenient if you have priorities other than writing the scripts yourself.
 
You must log in or register to reply here.
Back
Top