VOGONS


AI users?

Topic actions

Reply 20 of 79, by darry

User metadata
Rank l33t++
Rank
l33t++
keenmaster486 wrote on 2024-07-05, 19:46:

AI bots will frequently brazenly lie and then pretend they agreed with you all along when you correct them.

Also not very helpful to tell a bot from a real person.

Narcissistic humans can be, and in my experience often are, manipulative that way too. AI is definitely getting more similar to humans.

Reply 21 of 79, by wbahnassi

User metadata
Rank Oldbie
Rank
Oldbie
Jo22 wrote on 2024-07-05, 20:02:
I'm not sure I understand you fully. […]
Show full quote
wbahnassi wrote on 2024-07-05, 16:17:

AI posts is a waste of my time both reading and answering.

I'm not sure I understand you fully.

wbahnassi wrote on 2024-07-05, 16:17:

AI posts is a waste of my time both reading and answering.

Please elaborate. What makes you think that AI posts are a waste of your time both reading and answering?

I mean a post made by AI has no entropy for me. It's regurgirating past knowledge into a mess and spitting it out. False information, and at best, no new information, and there is no human behind who will feel happy or gracious for any help you provide. All in all, a waste of time to read and to answer to.

Turbo XT 12MHz, 8-bit VGA, Dual 360K drives
Intel 386 DX-33, TSeng ET3000, SB 1.5, 1x CD
Intel 486 DX2-66, CL5428 VLB, SBPro 2, 2x CD
Intel Pentium 90, Matrox Millenium 2, SB16, 4x CD
HP Z400, Xeon 3.46GHz, YMF-744, Voodoo3, RTX2080Ti

Reply 22 of 79, by Namrok

User metadata
Rank Oldbie
Rank
Oldbie

Man, I can't even wrap my mind around why someone would send AI over here. To what end? To see if it can be done? Do they turn around and try to shill something? I get that there is a lot of it on the more "current year" social media outlets attempting to change public opinion or sell goods. But this just isn't that sort of place.

Everything about it leaves me confused.

Win95/DOS 7.1 - P233 MMX (@2.5 x 100 FSB), Diamond Viper V330 AGP, SB16 CT2800
Win98 - K6-2+ 500, GF2 MX, SB AWE 64 CT4500, SBLive CT4780
Win98 - Pentium III 1000, GF2 GTS, SBLive CT4760
WinXP - Athlon 64 3200+, GF 7800 GS, Audigy 2 ZS

Reply 23 of 79, by wbahnassi

User metadata
Rank Oldbie
Rank
Oldbie

My guess is that it's more a playground. An experiment to see how well it works. It could be also an individual having fun with AI, feeding the posts to an AI agent and copy-pasting back the answers.. or worse, fully automating a bot.. which would be terrible if let loose to make as many accounts and posts it likes.

In all cases, the guy behind it probably gives 0 crap about the retro stuff, and only landed on these forums due to activity and wealth of content.

Again, all assuming that these are really posts made by AI.. I still can't tell for sure.

Turbo XT 12MHz, 8-bit VGA, Dual 360K drives
Intel 386 DX-33, TSeng ET3000, SB 1.5, 1x CD
Intel 486 DX2-66, CL5428 VLB, SBPro 2, 2x CD
Intel Pentium 90, Matrox Millenium 2, SB16, 4x CD
HP Z400, Xeon 3.46GHz, YMF-744, Voodoo3, RTX2080Ti

Reply 24 of 79, by leileilol

User metadata
Rank l33t++
Rank
l33t++
Namrok wrote on 2024-07-05, 21:59:

Man, I can't even wrap my mind around why someone would send AI over here. To what end?

It's to inconspicuously assimilate while trying to sell product on a post they hope no one catches to rank up on the search algos (...if there are even any working search algos anymore as it's all slopgen algos to appease investors for the time being). Spambots think they're getting better at the turing test with the help of the corporate plagiarizing markov chain machine (and sadly, certain responses here shows they are effective)

apsosig.png
long live PCem

Reply 25 of 79, by vvbee

User metadata
Rank Oldbie
Rank
Oldbie

Some conversations are high quality and some are low quality, what else is new. Beyond spam, what's it matter if AI takes part. For sure you can't successfully detect it by the seat of your pants, and thinking people are bots if they don't follow your recommendations is a nonstarter.

Reply 26 of 79, by the3dfxdude

User metadata
Rank Oldbie
Rank
Oldbie
darry wrote on 2024-07-05, 20:02:
keenmaster486 wrote on 2024-07-05, 19:46:

AI bots will frequently brazenly lie and then pretend they agreed with you all along when you correct them.

Also not very helpful to tell a bot from a real person.

Narcissistic humans can be, and in my experience often are, manipulative that way too. AI is definitely getting more similar to humans.

AI software is written by humans! This behavior of sounding convincing is from being programmed by humans who studied language on how words trigger emotion. This is done intentionally.

But I am pessimistic about this software. Natural sounding? Sure. Intelligent? No. Not going to happen.

I wonder what I just said means about humans.

Reply 27 of 79, by progman.exe

User metadata
Rank Member
Rank
Member
vvbee wrote on 2024-07-06, 15:04:

Beyond spam, what's it matter if AI takes part.

Because they cannot add anything.

They can do a statistical approximation of what help reads like, but they are not actually being helpful.

This article has some dramatic wording, its not aimed at anyone in particular, but maybe a plain bollocking is what crypto AI-hype echoers need now?
https://ludic.mataroa.blog/blog/i-will-fuckin … ntion-ai-again/

Reply 28 of 79, by wbahnassi

User metadata
Rank Oldbie
Rank
Oldbie
vvbee wrote on 2024-07-06, 15:04:

Some conversations are high quality and some are low quality, what else is new. Beyond spam, what's it matter if AI takes part. For sure you can't successfully detect it by the seat of your pants, and thinking people are bots if they don't follow your recommendations is a nonstarter.

Yeah, and I'm not claiming that my recommendations are the right thing either. But as I said, it's a combo of multiple indicators.. and in the end.. it can still be someone with a special personality.

But AI or not, I like being here because of the quality. If AI bots take over, quality will degrade, and it will be harder to get the time from the experts to pick their brains and have them tell us all about their nerdy stuff 🙂

Turbo XT 12MHz, 8-bit VGA, Dual 360K drives
Intel 386 DX-33, TSeng ET3000, SB 1.5, 1x CD
Intel 486 DX2-66, CL5428 VLB, SBPro 2, 2x CD
Intel Pentium 90, Matrox Millenium 2, SB16, 4x CD
HP Z400, Xeon 3.46GHz, YMF-744, Voodoo3, RTX2080Ti

Reply 29 of 79, by vvbee

User metadata
Rank Oldbie
Rank
Oldbie
progman.exe wrote on 2024-07-06, 16:40:
Because they cannot add anything. […]
Show full quote
vvbee wrote on 2024-07-06, 15:04:

Beyond spam, what's it matter if AI takes part.

Because they cannot add anything.

They can do a statistical approximation of what help reads like, but they are not actually being helpful.

This article has some dramatic wording, its not aimed at anyone in particular, but maybe a plain bollocking is what crypto AI-hype echoers need now?
https://ludic.mataroa.blog/blog/i-will-fuckin … ntion-ai-again/

The blog post looks to be ranting about practices in IT. Can't blame it, but it's in the weeds.

I assume your underlying point about AI not adding anything is that AI isn't human so it can't share the human experience. In any case, if it can play the role well enough then you can learn from it. Approximations are everywhere anyway, can't function without them.

Reply 30 of 79, by progman.exe

User metadata
Rank Member
Rank
Member
vvbee wrote on 2024-07-07, 01:58:

I assume your underlying point about AI not adding anything is that AI isn't human so it can't share the human experience. In any case, if it can play the role well enough then you can learn from it. Approximations are everywhere anyway, can't function without them.

There is no AI, lets get that straight. There is a marketing term, a grandiose title, but large language model is better.

Your assumption is wrong because you are putting it in the kind of woolly terms AI fanbois use.

A person cannot truly learn from someone that does not understand. A maths teacher can tell you times tables so you can learn by rote, but passing on the understanding is a better use of their time, when a piece of paper can convey the times tables.

A piece of paper can never understand, but it can still teach.

The statistical approximation I am talking about it not the kind of escape hatch you are seeking.

Try digesting this as how LLMs have no intelligence, but look like it to dumb people (which is why the hype flies so well)
https://awful.systems/post/1769506

Basically, phrase something so it sounds like puzzle or paradox, but do not present that, and the LLM will give nonsense. It will think it has seen a paradox, "answer" it because it is just copying, but it will be wrong because all they do is a statistical approximation: and that is just not good enough, things are not true because the answer sounds like it goes with the question.

Reply 31 of 79, by ratfink

User metadata
Rank Oldbie
Rank
Oldbie

I thought chatGPT (and by extension LLMs in general) should not be used to generate answers for anything mission critical. Old games, Dosbox and old hardware is mission critical to some of us.

But to be honest I have followed advice from a well respected moron on here and borked what is now an expensive card, so in all cases I think "buyer beware" and don't slavishly follow any advice if the outcome could be catastrophic - keep your own brain in gear.

However my pet hate is the hype around AI and how things that used to just be statistical methods are now touted as AI by new graduates and the media - and the way that multiple different things are all grouped under the same umbrella "AI". I suppose at least "machine learning", "data mining" and "data science" are being overtaken by the latest catchphrase.

Reply 32 of 79, by Jo22

User metadata
Rank l33t++
Rank
l33t++

This reminds me of F&A. It allowed using databases through natural language.

https://en.wikipedia.org/wiki/Q%26A_(Symantec)

"Time, it seems, doesn't flow. For some it's fast, for some it's slow.
In what to one race is no time at all, another race can rise and fall..." - The Minstrel

//My video channel//

Reply 33 of 79, by vvbee

User metadata
Rank Oldbie
Rank
Oldbie
progman.exe wrote on 2024-07-07, 10:27:
There is no AI, lets get that straight. There is a marketing term, a grandiose title, but large language model is better. […]
Show full quote
vvbee wrote on 2024-07-07, 01:58:

I assume your underlying point about AI not adding anything is that AI isn't human so it can't share the human experience. In any case, if it can play the role well enough then you can learn from it. Approximations are everywhere anyway, can't function without them.

There is no AI, lets get that straight. There is a marketing term, a grandiose title, but large language model is better.

Your assumption is wrong because you are putting it in the kind of woolly terms AI fanbois use.

A person cannot truly learn from someone that does not understand. A maths teacher can tell you times tables so you can learn by rote, but passing on the understanding is a better use of their time, when a piece of paper can convey the times tables.

A piece of paper can never understand, but it can still teach.

The statistical approximation I am talking about it not the kind of escape hatch you are seeking.

Try digesting this as how LLMs have no intelligence, but look like it to dumb people (which is why the hype flies so well)
https://awful.systems/post/1769506

Basically, phrase something so it sounds like puzzle or paradox, but do not present that, and the LLM will give nonsense. It will think it has seen a paradox, "answer" it because it is just copying, but it will be wrong because all they do is a statistical approximation: and that is just not good enough, things are not true because the answer sounds like it goes with the question.

AI can write posts like this, it's a moot point. It's also a dark argument. Ideas are the basis for discourse on open forums, not the poster's gender, race, religion, whatever.

Reply 34 of 79, by Jo22

User metadata
Rank l33t++
Rank
l33t++

What I think shouldn't be forgotten is that LLMs are being based on human input.
Something like ChatGP "learns" from the masses of information it's being fed with.
So it's not superior to human mind, but merely an mirror image of it.

That's why it has some undesired outcome, often.
The answers an LLM provides might be true but considered political incorrect same time.
Which in turn causes maintainers of LLMs to insert filter rules to "bend" LLMs in a way that conflicts with the source material.

The gender or racial topics are a good (or bad) example here.
If you ask LLMs to imagine a picture of, say, Caesar, the Roman emperor, it will draw anyone but an European.

In some way, this is both perverted (in quotation marks) and fascinating at same time.
It's essentially a field of science that focuses on driving the machine into insanity.
It's akin to have it have a fever dream.

Considering this, I'm kind of glad that my country is so backwards* in terms of digitalization, still.
We still have stationary phones on the desktop, fax machines (dedicated units or modern multi purpose devices), signed contracts on paper..

Especially the telephone is important, I think. Faking a human conversation on the phone is hard to do.
Especially if you're in an angry discussion with the person at the tax office. 😁

Edit: *Visiting my home country is a bit like time travel to foreigners, I assume.
It's like going back in time by 20/30/40 years. 😂

DSL in my street is 5 MBit, we have satellite dishes and primary method of payment is cash.
(Just an exaggeration. There's blazing fast 60 MBit internet via cable, too.)

Credit cards are being known, but seldomly accepted (Taxis do, hotels do etc).
Debit cards are fine, but sometimes only for 10€ payments and up (was standard before Corona).
Or, there's no lower limit but it costs extra money.

Visits at the doctors are still normal, telemedicine (video phones, online receipts)
had been explored experimentally in the 1980s/1990s, but it didn't become common.

It wasn't until January 2024 that we got electronic receipts.
Before this, coloured paper forms filled out with matrix printers had been standard.
Private receipts are still using old method. Which isn't bad, per se.

A photocopy of a paper receipt can be put into a fax machine,
for example, so the apothecary of your choice can quickly see what you want.

It's easier than submitting all the PZN numbers via phone
(yes, it's not uncommon to call your apothecary on the phone and make an order).

Edit: Formatting fixed (on PC).

"Time, it seems, doesn't flow. For some it's fast, for some it's slow.
In what to one race is no time at all, another race can rise and fall..." - The Minstrel

//My video channel//

Reply 35 of 79, by theelf

User metadata
Rank Oldbie
Rank
Oldbie

Maybe are post writen by people not english natives...

Reply 36 of 79, by keenmaster486

User metadata
Rank l33t
Rank
l33t
theelf wrote on 2024-07-07, 23:50:

Maybe are post writen by people not english natives...

That's actually a good way to identify a real person. AI nowadays has a good command of the English language. The tells are all in the content, what it's actually saying.

World's foremost 486 enjoyer.

Reply 38 of 79, by darry

User metadata
Rank l33t++
Rank
l33t++
keenmaster486 wrote on 2024-07-08, 17:49:
theelf wrote on 2024-07-07, 23:50:

Maybe are post writen by people not english natives...

That's actually a good way to identify a real person. AI nowadays has a good command of the English language. The tells are all in the content, what it's actually saying.

For now.
Wait until they train AI models to simulate poor English. It will probably be tunable and localisable ( $nationality person writing in $level English).