Bad uses / over-promised of AI

Forums › Artificial Intelligence (AI, LLM, ML) › Bad uses / over-promised of AI

1 | ... | 7 | 8 | 9 | 10 | 11 | 12 | 13 | 14 | 15 | 16 | 17

altered-ego
2474 posts

Uber Geek
+1 received by user: 870

Trusted

Lifetime subscriber

#3488587 8-May-2026 01:11

Journeyman: The keyboard is dirty because it's a shared work keyboard and doesn't get cleaned. I just asked AI to replace the words 😄

Ahhh, AI manipulated, not AI created.

Please keep this GZ community vibrant by contributing in a constructive & respectful manner.

gzt

19162 posts

Uber Geek
+1 received by user: 8286

Lifetime subscriber

#3488762 8-May-2026 15:35

gzt: Richard Dawkins the don of delusion predictably falls madly in love with Claude's ai sycophancy, refers to it as Claudia and declares it is conscious.

The syndicated Telegraph article is based on/or publicity for an original piece by Richard Dawkins.

The original may be more subtle, or not. It is possible Dawkins is saying Claude passed the Turing Test defined as being mistaken for a human so it's not surprising that people including himself feel that way or Dawkins may have been completely fooled. It will cost you $3 to access the original piece to find out.

Monash academics review Dawkins' essay:

https://theconversation.com/is-richard-dawkins-right-about-claude-no-but-its-not-surprising-ai-chatbots-feel-conscious-to-us-282151

gzt

19162 posts

Uber Geek
+1 received by user: 8286

Lifetime subscriber

#3490074 12-May-2026 10:27

An MSN News article - curated by Copilot. This is an interesting development:

https://www.msn.com/en-ca/news/insight/trump-hints-at-third-term-while-unveiling-expanded-election-monitoring/gm-GM3AB31C21

I didn't see any obvious issues but I read only one source article.

solaybro

633 posts

Ultimate Geek
+1 received by user: 228

Trusted

#3490968 14-May-2026 18:19

I was using Copilot at work to vibe code together a small tool that pulls data from PDFs and exports it to Excel. It was working great. Then I asked Copilot to move the export button because I didn’t like where it was. It moved the button but the PDF extraction stopped working.

I told Copilot it wasn’t extracting anything anymore, and it basically said, “Yeah, that’s because I deleted that part of the code.”

What???

kingdragonfly

12160 posts

Uber Geek
+1 received by user: 13179

Subscriber

#3490970 14-May-2026 18:31

Summary don't trust "curated by CoPilot", as AI still hallucinates, and AI is just polluting the news with slop.

It is a rehash of human written (hopefully) original news item

See

Microsoft Publishes Garbled AI Article Calling Tragically Deceased NBA Player “Useless”: Futurism

Former NBA player Brandon Hunter passed away unexpectedly at the young age of 42 this week, a tragedy that rattled fans of his 2000s career with the Boston Celtics and Orlando Magic.

But in an unhinged twist on what was otherwise a somber news story, Microsoft’s MSN news portal published a garbled, seemingly AI-generated article that derided Hunter as “useless” in its headline.

“Brandon Hunter useless at 42,” read the article, which was quickly called out on social media. The rest of the brief report is even more incomprehensible, informing readers that Hunter “handed away” after achieving “vital success as a ahead [sic] for the Bobcats” and “performed in 67 video games.”
...

Behodar

11244 posts

Uber Geek
+1 received by user: 6274

Trusted

Lifetime subscriber

#3491146 15-May-2026 08:45

I was running a (non-AI) code analysis tool on some of my code and it flagged a potential performance issue where I was iterating over an object twice. I couldn't see any way of working around that, but figured I'd at least try asking AI.

It cheerily gave me a two-line replacement that does indeed only iterate once. However, it seemed wrong. After looking at it and working it through, I realised that the original code would return false if given an empty set, but the new code would return true.

"Doesn't this give the wrong answer if existingIds is empty?" I ask. "Yes, the simplified version changes the logic," replies the AI. It then offers a "corrected single-pass version". This "single-pass" version iterates twice. In fact, it's identical to the code I asked it to rewrite in the first place, except that it's renamed one of the variables.

Dell

Shop now for Dell laptops and other devices (affiliate link).

frankv

5714 posts

Uber Geek
+1 received by user: 3679

Lifetime subscriber

#3491151 15-May-2026 09:04

solaybro:

I was using Copilot at work to vibe code together a small tool that pulls data from PDFs and exports it to Excel.

OT, but I have a Java library designed to strip tables from PDFs. It uses the pdfbox library. https://github.com/frankvdh/PdfLinedTables

Tinkerisk

4963 posts

Uber Geek
+1 received by user: 3910

#3491215 15-May-2026 13:46

No one should delude themselves about AI.

Qui nihil scit, omnia credere debet. - He who knows nothing must believe everything.
Firewalls do NOT stop dragons. Really not!
I avoid Big Tech. They try hard to dictate technology and „culture“ across borders.
In effect we have everything to hide from someone, and no idea who „someone“ is.

roobarb

786 posts

Ultimate Geek
+1 received by user: 836

Trusted

#3494829 22-May-2026 20:30

... and so the enshittification of government services begins AI sackings reach New Zealand, which will use it to eject 14 percent of government staff

Replace those who know how to interpret legislation into IT services with dudes who claim to know all about AI and will just vibe code the whole thing.

Who cares about security, GDPR or data sovereignty when all you need to do is ask the AI.

Move along here, nothing to see, what could possibly go wrong?

I expect the new IRD help will advise people how to apply for Sovereign Ciitizenship

Tinkerisk

4963 posts

Uber Geek
+1 received by user: 3910

#3494881 23-May-2026 14:31

... and the „enforced“ global TDM2.0 rollout says hello! 😁

Qui nihil scit, omnia credere debet. - He who knows nothing must believe everything.
Firewalls do NOT stop dragons. Really not!
I avoid Big Tech. They try hard to dictate technology and „culture“ across borders.
In effect we have everything to hide from someone, and no idea who „someone“ is.

Rikkitic

Awrrr
19206 posts

Uber Geek
+1 received by user: 16553

Lifetime subscriber

#3495206 25-May-2026 16:48

Okay we all know AI, especially Chat GPT, makes mistakes and is not to be trusted. But isn't it at least supposed to be good at assembling known facts from documented sources?

Recently I had a couple of interesting sessions with ChatGPT. In the first, it absolutely insisted that the Australian Liberals had never had a female leader! When I pointed out Sussan Ley, it seemed to confuse her with Julia Gillard. As our discussion progressed, it would correctly state in one response that Ley had been a Liberal leader, then again forcefully insist in the next response that no woman had ever headed the Liberal party. It sent back and forth responses like this several times before I finally shut it down. It seemed to be in some kind of weird loop where it would state the correct information, then deny it, then repeat it.

Then today the same thing happened again. I asked about Robert Redford’s death and it insisted rather disdainfully that he was still alive and well. Again we went back and forth a couple times before it finally accepted that Redford was no longer with us, though it was not nearly as bad as the situation with Ley.

The conclusion of this has to be that ChatGPT is only useful for information you already know the correct answer to and can verify. Otherwise it is all just hallucination.

Plesse igmore amd axxept applogies in adbance fir anu typos

Dyson

Shop now for Dyson appliances (affiliate link).

kingdragonfly

12160 posts

Uber Geek
+1 received by user: 13179

Subscriber

#3495422 26-May-2026 09:08

Two AI's. Same questions. Different definitive answers. Both wrong.

one AI overengineers the solution. another invents accessory details. the first doubles down, then retracts.

I'm trying to set up the audio in a glass-walled conference room, with the least amount of money. Even a $10 savings would be great. It's for a 4 meter long conference table, glass-walled room (lots of echoes). Currently there's the old cheapest USB microphone sitting in the middle of the table, which is terrible.

So I buy an extremely cheap old-school auto mixer that supports 4 old-style wired XLR microphones (gating/attenuation). Automixer has a USB out.

It works as advertised, seamlessly switching to strongest source. A real old-school professional setup that originally cost thousands.

Old microphones have 3-pin XLR connections, which you may have seen if you ever had a rock 'n' roll band.

I know Shure conference microphones use mini-XLR proprietary connections, and wiring these is a pain.

So I tell Copilot: I found 5 used microphones for a good price, no cables, no pictures of outputs. To Copilot: "No cables supplied. Does it have a full-sized XLR audio output?"

It then tells me I need to rebuild the room, and should use gooseneck microphones. Meaning every person rotates the closest microphone to them, which they ain't goin' to do.

It says the microphones are indeed full size.

Knowing you can't trust AI, I go to ChatGPT, ask the same question. It says "yes. It comes with a breakout cable". I don't know what a breakout cable is, so I ask. And it says "it's a supplied cable that converts mini-XLR to normal full-sized XLR".

I return to Copilot, and basically say "what the f." I just asked you if it uses a mini connection. It says "Copilot is correct, ChatGPT is wrong. It says here's my proof", sending me a photo that doesn't show the connection, and is fully in Chinese. Not an English letter or Arabic number on it.

I ask "are you sure? Give me another URL". It says there is no other URL, but it's sure.

I say that ChatGPT mentioned a supplied breakout cable, which they are not selling.

It then says "whoops, my bad".

Aaaaargh!

roobarb

786 posts

Ultimate Geek
+1 received by user: 836

Trusted

#3495483 26-May-2026 12:22

If only there was some recent example of a country using an immature, unreliable and misunderstood technology to make savings in government workforce. I suggest that NZ Gov looks at how successful DOGE was, using teenage script kiddies to identify surplus workforce, fire them, then find the predicted cost saving didn't materialise, government services suffered and they had to rehire many of the redundant workers.

At least South Atlantic Penguins are now subject to the tariffs they had been avoiding.

Kookoo

891 posts

Ultimate Geek
+1 received by user: 432

Trusted

#3495526 26-May-2026 15:36

Mo Bitar shared an interesting thought on his channel. He was commenting on the fact that LLM founders take turns announcing that "AI will replace white collar workers by 2028" and so on. He basically asked - who were the saying it for? And the answer he proposed was that it was a runaway sales pitch. It's aimed at the corporations and government organisations who're buying into the cost savings proposition. "Just wait, we're almost there, give it another year or two and you'll be able to reduce your labour costs to zero."

And the reason they keep saying it is simple. AI doesn't stack up as a VAS - the underlying economics don't support adopting AI as a tool for humans to use. So, it has to be sold as a replacement of human labour, otherwise it won't sell at all. The fact that it's not fit for that purpose (let alone the broader economic impact of such replacement) seems irrelevant.

Hello, Ground!

Tinkerisk

4963 posts

Uber Geek
+1 received by user: 3910

#3495530 26-May-2026 15:50

AI: White color or white collar workers? 😁

Qui nihil scit, omnia credere debet. - He who knows nothing must believe everything.
Firewalls do NOT stop dragons. Really not!
I avoid Big Tech. They try hard to dictate technology and „culture“ across borders.
In effect we have everything to hide from someone, and no idea who „someone“ is.

1 | ... | 7 | 8 | 9 | 10 | 11 | 12 | 13 | 14 | 15 | 16 | 17