I listened to a podcast where an author spoke about using agentic AI to
make a writing "team". He had beta reader/critics responding from
specific perspectives, another grammar LLM, and an editor LLM - he'd
pass drafts back and forth between them like they were people as part of
the process. I have a couple of paid LLMs - Microsoft CoPilot for 365 as
part of a subscription, and Perplexity (I have a 3 month free trial
program). I started playing with them to see how they could benefit.

I was trying to research a science-fiction book I'd read as a teenager,
I only knew the name of one of the races in book, and tried Gemini,
Perplexity, ChatGPT and CoPilot. Only ChatGPT pulled up the book title
and author, along with a summary - and this was the free ChatGPT
version.

I assume that's more a function of the training library, not the LLM
itself.

Then, I tried giving them outlines of plot ideas to write - Gemini came
in last, the others were comparable. As a last task, I asked them all to
write a 500-word short story about an astronaut stranded on Mars, with
elements of the story Robinson Crusoe on Mars and The Martian.

ChatGPT felt more nuanced, CoPilot even used the names of the sources in
the story. Perplexity felt like a direct-to-dvd version of "The Martian"
that you'd see one on of those free channels on Roku".

I think I'll use CoPilot when writing, I like the idea of training it on
my own documents and having it easily identify my writing style and body
of work out of the box.



--- MultiMail/Win v0.52
* Origin: realitycheckBBS.org -- information is power. (21:4/122)

Or you can just... you know... write.

---
Ganiman
bbs.madetoraid.com:[2323/2222]

--- Mystic BBS v1.12 A48 (Linux/64)
* Origin: Made to Raid BBS (21:3/174)

Re: Re: Writing with LLMs
By: Ganiman to poindexter FORTRAN on Mon Sep 15 2025 10:16 pm

Or you can just... you know... write.

It would be helpful to quote what you're replying to so that it's easy for people to follow the conversation (especially the person you're repying to).

Nightfox
--- SBBSecho 3.29-Linux
* Origin: Digital Distortion: digdist.synchro.net (21:1/137)

Hello Nightfox!

** On Tuesday 16.09.25 - 10:28, Nightfox wrote to Ganiman:

It would be helpful to quote what you're replying to so that it's easy for people to follow the conversation (especially the person you're repying to).

Your system doesn't support threading?

References of "Writing with LLMs"
1636 14.09.25 poindexter FORTRAN Writing with LLMs
166 15.09.25 ÃÄÄGaniman
395 16.09.25 ³ ÃÄÄNightfox

Your post links to the two previous ones just fine here. ;)



--- OpenXP 5.0.64
* Origin: (} Pointy McPointFace (21:4/106.21)

Re: Writing with LLMs
By: Ogg to Nightfox on Tue Sep 16 2025 06:21 pm

It would be helpful to quote what you're replying to so that it's easy for
people to follow the conversation (especially the person you're repying
to).

Your system doesn't support threading?

Yes, but it's still considered good etiquette to quote what you're responding to.

Nightfox
--- SBBSecho 3.29-Linux
* Origin: Digital Distortion: digdist.synchro.net (21:1/137)

BY: Ganiman (21:3/174)
On Monday,September 15, 2025 at 09:16 PM, Ganiman (21:3/174) wrote:

Or you can just... you know... write.

Exactly. I am not buying any book from any author that is AI written slop. And most publishers will ban you immediately when it is found out that you use AI to write.

People pay artists and writers to create art -- not to use AI to create junk.

-dmxrob


--- WWIV 5.9.03748[Linux 6.5.0-1026]
* Origin: Off the Wall - St. Peters, Missouri - #VoteBlue (21:4/142)

Re: Re: Writing with LLMs
By: Dmxrob to Ganiman on Wed Sep 17 2025 01:23:23

Hi, Rob.

Exactly. I am not buying any book from any author that is AI written slop. And most publishers will ban you immediately when it is found out that you use AI to write.

It worries me how much I see people depending on ChatGPT - people in work, people I sit next to on the train, university professors... I feel like my brain is rusting up enough from age without speeding it up by outsourcing my thought to a machine as well.

I was recently involved in a tender to be allowed to sell into certain types of public sector organisations - the instructions for applicants stated in bold type that using ChatGPT / LLMs for *any* part of your response would result in immediate rejection. You still get people saying "oh well let's get ChatGPT to write an answer then we can just edit it". Give me strength!

BobW
--- SBBSecho 3.29-Linux
* Origin: >>> Magnum BBS <<< - magnumbbs.net (21:1/205)

Re: Re: Writing with LLMs
By: Bob Worm to Dmxrob on Wed Sep 17 2025 08:23 am

It worries me how much I see people depending on ChatGPT - people in work, people I sit next to on the train, university professors... I feel like my brain is rusting up enough from age without speeding it up by outsourcing my thought to a machine as well.

I'm a software engineer, and recently my manager at work asked me if I've tried using Copilot to write any code. Maybe it's just to try to justify the company buying Copilot licenses.. I don't really want to rely on AI to write code though, at least not fully. I've also seen AI-generated code be very wrong, or sometimes it would give the same as something I'd find with a Google search.

Nightfox
--- SBBSecho 3.29-Linux
* Origin: Digital Distortion: digdist.synchro.net (21:1/137)

Re: Re: Writing with LLMs
By: Dmxrob to Ganiman on Wed Sep 17 2025 01:23 am

Exactly. I am not buying any book from any author that is AI written slop. And most publishers will ban you immediately when it is found out that you use AI to write.

People pay artists and writers to create art -- not to use AI to create junk.

Pretty much this. Among other things, because when things are computer generated, it *shows*.

Heck, even computer assisted translations need a lot of reworking in order not to feel you are reading pure computer output.


--
gopher://gopher.richardfalken.com/1/richardfalken
--- SBBSecho 3.29-Linux
* Origin: Palantir * palantirbbs.ddns.net * Pensacola, FL * (21:2/138)

Re: Re: Writing with LLMs
By: Bob Worm to Dmxrob on Wed Sep 17 2025 08:23 am

It worries me how much I see people depending on ChatGPT - people in work, people I sit next to on the train, university professors... I feel like my brain is rusting up enough from age without speeding it up by outsourcing my thought to a machine as well.

I don't see that much people using it actively, other than for entertainment. Some friends of mine use it to generate boilerplate code or unit tests, but they always have to fix it themselves, so it is not like they are letting the computer do all the work blindly.

I was recently involved in a tender to be allowed to sell into certain types of public sector organisations - the instructions for applicants stated in bold type that using ChatGPT / LLMs for *any* part of your response would result in immediate rejection. You still get people saying "oh well let's get ChatGPT to write an answer then we can just edit it". Give me strength!

I don't think it is bad to have a computer generate some ideas for you to work on. Copyediting computer output, however, is outright cheating.


--
gopher://gopher.richardfalken.com/1/richardfalken
--- SBBSecho 3.29-Linux
* Origin: Palantir * palantirbbs.ddns.net * Pensacola, FL * (21:2/138)

On 17 Sep 2025 at 09:02a, Nightfox pondered and said...

Re: Re: Writing with LLMs
By: Bob Worm to Dmxrob on Wed Sep 17 2025 08:23 am

It worries me how much I see people depending on ChatGPT - people in people I sit next to on the train, university professors... I feel li brain is rusting up enough from age without speeding it up by outsour my thought to a machine as well.

I'm a software engineer, and recently my manager at work asked me if
I've tried using Copilot to write any code. Maybe it's just to try to justify the company buying Copilot licenses.. I don't really want to
rely on AI to write code though, at least not fully. I've also seen AI-generated code be very wrong, or sometimes it would give the same as something I'd find with a Google search.

I've been playing with it recently (well, Claude, not ChatGPT).
I've found it about 2/3 miss and 1/3 hit.

Recent tasks include using it as a glorified search engine,
asking for specific points about the use of a particular
library I wanted to program against, for example. It was...
ok for that. I asked it to generate an example program, and
it was able to do so, but that program was deficient in some
ways. When I asked it to explain the rationale behind some
of the choices, it hallucinated a bunch of nonsense.

Another time, I asked it to explain what, I'm pretty sure, is
a bug in the GNU assembler; specifically, I made it do a bunch
of analysis that I could have done myself, but which would
have been tedious and time consuming. It did an adequate job
of pinpointing the section of code I wanted it to, but
alternated confidently between asserting that the observed
behavior was intended or a bug.

In both cases, my conclusion was that, had I not already had
decent sense of what was going on in both cases, I'd have spent
a _lot_ of time chasing wild geese.

Automated generation of boilerplate code is all well and good,
but even for that, I'm not really seeing the value-add. My
real desire is to use these tools as a research assistant, and
for that, they still fall well short of the mark. Like a bad
intern that speaks with the confidence of a pretty big ego on
one hand, but the sycophantic obsequiousness of an ass-kisser
waiting to stab you in the back.

--- Mystic BBS v1.12 A48 (Linux/64)
* Origin: Agency BBS | Dunedin, New Zealand | agency.bbs.nz (21:1/101)

I've also seen AI-generated code be very wrong, or
sometimes it would give the same as something I'd find with a Google search.

Chances are that AI "stole" that answer from a web search.


* SLMR 2.1a * Four snack groups: frozen, crunchies, cakes and sweets.
--- SBBSecho 3.28-Linux
* Origin: capitolcityonline.net * Telnet/SSH:2022/HTTP (21:1/175)

Ganiman wrote to poindexter FORTRAN <=-

Or you can just... you know... write.

Preach, brother. While I'm intrigued by a team full of LLMs, I write on
a 45 cent spiral notebook with a Bic crystal pen. Blue, not black.

I appreciate reducing the art to the lowest common denominator and
writing words on paper instead of typing.



--- MultiMail/Win v0.52
* Origin: realitycheckBBS.org -- information is power. (21:4/122)

Re: Re: Writing with LLMs
By: Dumas Walker to NIGHTFOX on Thu Sep 18 2025 09:45 am

I've also seen AI-generated code be very wrong, or
sometimes it would give the same as something I'd find with a Google
search.

Chances are that AI "stole" that answer from a web search.

Yeah, I think that's basically how they work.

Nightfox
--- SBBSecho 3.29-Linux
* Origin: Digital Distortion: digdist.synchro.net (21:1/137)

Re: Re: Writing with LLMs
By: tenser to Nightfox on Thu Sep 18 2025 23:49:07

Hi, tenser.

alternated confidently between asserting that the observed
behavior was intended or a bug.

I heard that this is a result of the way the models are trained. Essentially the model gets no "points" for saying that it doesn't know the answer to something, whereas a confidently stated answer with some grain of truth in it (something you'd call bluffing if a person did it) would score at least some points. So you get what's incentivised.

BobW
--- SBBSecho 3.29-Linux
* Origin: >>> Magnum BBS <<< - magnumbbs.net (21:1/205)

BY: Bob Worm (21:1/205)
On Wednesday,September 17, 2025 at 07:23 AM, Bob Worm (21:1/205) wrote:

Hi, Rob.

Hey Bob!

It worries me how much I see people depending on ChatGPT - people in
work, people I sit next to on the train, university professors... I feel like my brain is rusting up enough from age without speeding it up by outsourcing my thought to a machine as well.

It worries me too. And not because I think AI will "take over all the jobs" or any of that nonsense. But exactly as you said, people are letting their brains rust. It was bad enough to start with, but now people are getting to the point where they can't even solve basic problems or they don't know how to distiguish between right and wrong.

Then you have AI increasingly getting things wrong, injecting false information or hallucinating and it is a disaster. There are fewer and fewer "smart people" left to solve the actual problems, and those who are using AI for every little thing don't know what to do when something goes wrong that they can't ask ChatGPT for help with -- and even when it does help, how do you know it's telling you the truth?

It's scary and I think it's going to make the brain drain problem in business much worse. People who can actually THINK, REASON and SOLVE will be able to write their own paychecks.

-dmxrob


--- WWIV 5.9.03748[Linux 6.5.0-1026]
* Origin: Off the Wall - St. Peters, Missouri - #VoteBlue (21:4/142)

BY: Nightfox (21:1/137)
On Wednesday,September 17, 2025 at 08:02 AM, Nightfox (21:1/137) wrote:

I'm a software engineer, and recently my manager at work asked me if
I've tried using Copilot to write any code. Maybe it's just to try to justify the company buying Copilot licenses.. I don't really want to
rely on AI to write code though, at least not fully. I've also seen AI-generated code be very wrong, or sometimes it would give the same as something I'd find with a Google search.

"I don't use it because I know how to solve the problem and can do it quicker than debugging what ChatGPT told me to do."

That's my standard answer.

-dmxrob


--- WWIV 5.9.03748[Linux 6.5.0-1026]
* Origin: Off the Wall - St. Peters, Missouri - #VoteBlue (21:4/142)

Re: Re: Writing with LLMs
By: Dmxrob to Nightfox on Thu Sep 18 2025 09:22 pm

"I don't use it because I know how to solve the problem and can do it quicker than debugging what ChatGPT told me to do."

That's my standard answer.

Yeah, that's basically how I feel about it right now.

Nightfox
--- SBBSecho 3.29-Linux
* Origin: Digital Distortion: digdist.synchro.net (21:1/137)

BY: Bob Worm (21:1/205)
On Wednesday,September 17, 2025 at 07:23 AM, Bob Worm (21:1/205) wrote:

Hey Bob!

It worries me too. And not because I think AI will "take over all the jobs" or any of that nonsense. But exactly as you said, people are letting their brains rust. It was bad enough to start with, but now people are getting to the point where they can't even solve basic problems or they don't know how to distiguish between right and wrong.

Then you have AI increasingly getting things wrong, injecting false information or hallucinating and it is a disaster. There are fewer and fewer "smart people" left to solve the actual problems, and those who are using AI for every little thing don't know what to do when something goes wrong that they can't ask ChatGPT for help with -- and even when it does help, how do you know it's telling you the truth?

It's scary and I think it's going to make the brain drain problem in business much worse. People who can actually THINK, REASON and SOLVE will be able to write their own paychecks.

-dmxrob

--- WWIV 5.9.03748[Linux 6.5.0-1026]
* Origin: Off the Wall - St. Peters, Missouri - #VoteBlue (21:4/142)

While readint the above made me think how I will enter a URL in DuckDuckGo search box so I can use one of the Results to go to that webpage .
I do that thinking if DuckDuckGo is familiar with the website then it should be O.K. to see what is on their page(s).
Ed
--- SBBSecho 3.28-Linux
* Origin: capitolcityonline.net * Telnet/SSH:2022/HTTP (21:1/175)

On 18 Sep 2025 at 09:08p, Bob Worm pondered and said...

alternated confidently between asserting that the observed
behavior was intended or a bug.

I heard that this is a result of the way the models are trained. Essentially the model gets no "points" for saying that it doesn't know
the answer to something, whereas a confidently stated answer with some grain of truth in it (something you'd call bluffing if a person did it) would score at least some points. So you get what's incentivised.

Yup. That sounds about right. But the LLM isn't fishing for
points on a true/false quiz; as a tool, that kind of behavior
stinks.

--- Mystic BBS v1.12 A48 (Linux/64)
* Origin: Agency BBS | Dunedin, New Zealand | agency.bbs.nz (21:1/101)

Dumas Walker wrote to NIGHTFOX <=-

Chances are that AI "stole" that answer from a web search.

I would hate to have a business run on Google ad revenue, they're
scraping and providing AI answers, then sponsored links, then the actual
link you were looking for.



--- MultiMail/Win v0.52
* Origin: realitycheckBBS.org -- information is power. (21:4/122)

Dmxrob wrote to Bob Worm <=-

Then you have AI increasingly getting things wrong, injecting false information or hallucinating and it is a disaster. There are fewer and fewer "smart people" left to solve the actual problems

I heard that one of the problems we'll face is that AI needs to be
trained on real-world, vetted human thought. More and more content on
the web is AI-generated, meaning LLMs are training on output from other
LLMs which taints the model.

It's scary and I think it's going to make the brain drain problem in business much worse. People who can actually THINK, REASON and SOLVE
will be able to write their own paychecks.

That was the case long before AI. Other countries can turn out
knowledge workers, administrators and coders, but the real opportunity
is lateral thinking and leadership. Always has been. It's just much
more apparent now.






--- MultiMail/Win v0.52
* Origin: realitycheckBBS.org -- information is power. (21:4/122)

Dmxrob wrote to Nightfox <=-

"I don't use it because I know how to solve the problem and can do it quicker than debugging what ChatGPT told me to do."

I used ChatGPT to look at some php code that didn't run after a PHP
upgrade and suggest upgrades to the code to remove deprecated commands
and syntax. It worked. I quit while I was ahead. :)



... "We can't stop here, this is bat country."
--- MultiMail/Win v0.52
* Origin: realitycheckBBS.org -- information is power. (21:4/122)

Chances are that AI "stole" that answer from a web search.

I would hate to have a business run on Google ad revenue, they're
scraping and providing AI answers, then sponsored links, then the actual
link you were looking for.

Yeah, there is a joke meme out there somewhere that came out after they
first started doing that. Something about the first several answers being
AI things that could get you killed, with the actual info showing up several answers down. Any more, I do skim the first few answers, but I always keep going until I (hopefully) find the actual links that the AI answers were scraped from.

IIRC, there is/was a way to tell it not to give you an AI answer when you search, but I haven't used it enough to remember the exact format.


* SLMR 2.1a * Acid absorbs 10 times its weight in excess reality.
--- SBBSecho 3.28-Linux
* Origin: capitolcityonline.net * Telnet/SSH:2022/HTTP (21:1/175)

Re: Re: Writing with LLMs
By: Dumas Walker to POINDEXTER FORTRAN on Sun Sep 21 2025 10:09:33

Hi, Dumas.

IIRC, there is/was a way to tell it not to give you an AI answer when you search, but I haven't used it enough to remember the exact format.

There is a proper way to do that (something obscure like h=14?) which I can never remember, however it's very easy to remember the alternative method which is to just swear in your query.

I didn't believe it at first but, sure enough, it works - you get your f-ing search results with no sh-y AI. Supposedly they didn't fancy having news articles written about how Google's AI helped a kid find smut or whatever so if your query is a bit risque then it turns off the AI.

I'm sure if someone reviewed my search history they would think "wow, this guy's angry", shortly followed by "wow, this guy's boring" :)

Now I need to find a way to get rid of the ads / sponsored listings...

BobW
--- SBBSecho 3.29-Linux
* Origin: >>> Magnum BBS <<< - magnumbbs.net (21:1/205)

There is a proper way to do that (something obscure like h=14?) which I can never remember, however it's very easy to remember the alternative method whic
is to just swear in your query.

LOL that is something I'd have never thought of doing!

Now I need to find a way to get rid of the ads / sponsored listings...

I bet that is a lot more difficult than just swearing in your query. ;)


* SLMR 2.1a * 9 out of 10 men who try camels prefer women.
--- SBBSecho 3.28-Linux
* Origin: capitolcityonline.net * Telnet/SSH:2022/HTTP (21:1/175)