Which AI services do you subscribe to?

Forums › Artificial Intelligence (AI, LLM, ML) › Which AI services do you subscribe to?

1 | 2 | 3

442 posts

Ultimate Geek
+1 received by user: 290

#3476368 1-Apr-2026 02:15

Wow - well the LG website photos would have fooled me pretty completely (and I'm a mech eng) !

Those photos showing symmetrical bolts on the LHS & RHS would shout out - unscrew hinge & door-catch and swap them.

And, as you say, photos 1 & 3 show the door swapped. Quite clearly.

But, sadly, LG's using photos from previous models - and the manual for the LG DVH9-10B omits the door-swap instructions that were there in the manual for the previous several models (eg: LG DVH9-09B).

So yes - that's a masterly example of AI getting it completely (and very professionally & in great detail) wrong.

Scott3

4177 posts

Uber Geek
+1 received by user: 2990

Trusted

Lifetime subscriber

#3476370 1-Apr-2026 03:16

pdh:

Wow - well the LG website photos would have fooled me pretty completely (and I'm a mech eng) !

Those photos showing symmetrical bolts on the LHS & RHS would shout out - unscrew hinge & door-catch and swap them.

And, as you say, photos 1 & 3 show the door swapped. Quite clearly.

But, sadly, LG's using photos from previous models - and the manual for the LG DVH9-10B omits the door-swap instructions that were there in the manual for the previous several models (eg: LG DVH9-09B).

So yes - that's a masterly example of AI getting it completely (and very professionally & in great detail) wrong.

In person the screw patterns are different visibly too wide on the right hand side.

Wonder if LG did something funny like chuck the 10kg condensate tray in the 9kg unit so they didn't need to move a an entire dryer in for photography?

9kg does list a reversable door as a feature. (But oddly the screw pattern throughout the photoset is inconsistent.

https://www.lg.com/au/washer-dryers/dryers/dvh9-09w/

cddt

1967 posts

Uber Geek
+1 received by user: 1904

#3476390 1-Apr-2026 08:15

pdh:

Genuinely interested in an example of a time-wasting instance.
Can you recollect one and rough out how it either misled / fooled you ?

A family member uses google search for "restaurant xyz opening hours" and Google AI gives an answer. We drive to the restaurant, the restaurant is closed. The website had the correct opening hours, Google AI's answer was BS.

MikeAqua

8024 posts

Uber Geek
+1 received by user: 3818

#3476421 1-Apr-2026 09:39

Earlier this year, I was putting together a playlist for a 10K run. I wanted something a bit feel good for the six km mark, where it all gets a bit hard ... I asked google AI search who sang What's Up. Its answer "Three Non-Blondes". From that answer I remembered that it was Four Non-Blondes, not three. I got there, but where on earth did it get three from?

Mike

alasta

6889 posts

Uber Geek
+1 received by user: 3363

Trusted

Subscriber

#3476426 1-Apr-2026 09:48

MikeAqua:

Earlier this year, I was putting together a playlist for a 10K run. I wanted something a bit feel good for the six km mark, where it all gets a bit hard ... I asked google AI search who sang What's Up. Its answer "Three Non-Blondes". From that answer I remembered that it was Four Non-Blondes, not three. I got there, but where on earth did it get three from?

At work I fed the AI tool some data with statistics by region. I wanted it to provide some commentary on which regions were performing well, or not.

In its commentary, it misspelled 'Southland' as 'Soutland'. I went and checked my source data and no such misspelling existed.

I have no idea where it got that misspelled text from!

networkn

Networkn
32864 posts

Uber Geek
+1 received by user: 15454

ID Verified

Trusted

Lifetime subscriber

#3476430 1-Apr-2026 10:01

ChatGPT was the first one. I had it humming then something happened over the xmas break and all the specific guardrails I put around the responses started being ignored, and when queried, it acknowledged it was skipping it. Unsubscribed in favour of Gemini which is better but still not perfect, and have been on Claude paid for around 6 weeks and so far it's awesome.

Samsung

Shop now on Samsung phones, tablets, TVs and more (affiliate link).

pdh

442 posts

Ultimate Geek
+1 received by user: 290

#3476445 1-Apr-2026 10:30

Thanks guys - those are awesome examples.

I have my own 'AI got it completely wrong' stories - but they all date from about a year ago, and since then I haven't caught it out. So I reckon it's improving - or I'm asking it questions that tax it less.

The most egregious was a simple engineering calculation that I was too lazy to look up the formula for (beam deflection under load with different end-points - right out of 2nd year mechanics ;-)

As I had phrased the question in a way that let the AI engine decide I wanted a 'Yes it will work' answer, it bent the math to make it work. My old slide-rule era sniff test called BS - the AI-calculated deflection was just too small. So I did the calc carefully myself - then double-checked it, because I'm getting old - and then called BS to the AI session.

It blithely agreed that if it gave me the correct answer I couldn't achieve the outcome I wanted, It seemed to preen itself that it had bent the numbers to give me a 'happy' answer. I said it needed to give me the correct answer - and it apologised and gave me a more accurate - but still wrong - answer.

So I gave up, having learned something. Perhaps the training algorithms of large-language-model AIs are too determined by people in the 'soft' sciences - where today it seems quite acceptable to start with the conclusion and cherry-pick the 'science' that supports the desired outcome. And I'm pointing the finger at some quite large areas of 'science'.

Unfortunately, that approach doesn't work in mechanical engineering ;-)

pdh

442 posts

Ultimate Geek
+1 received by user: 290

#3476449 1-Apr-2026 10:36

> "Three Non-Blondes"... I got there, but where on earth did it get three from?

In the bad old days, this would have engendered a bunch of un-PC jokes about the AI-engine having access to photos that revealed things not publicly seen or known...

roobarb

705 posts

Ultimate Geek
+1 received by user: 656

Trusted

#3476556 1-Apr-2026 12:46

MikeAqua:

"Three Non-Blondes".

That is the shortened radio play version. You were thinking of the extended remix or the album version.

Cybnate

76 posts

Master Geek
+1 received by user: 28

#3476557 1-Apr-2026 12:48

OpenAI api for Nextcloud use and some experimentation. Works for me and you can really manage your costs (chose the right model for the job).
Recently signed up with free account for Claude. Just to see the difference.

allio

895 posts

Ultimate Geek
+1 received by user: 529

#3476594 1-Apr-2026 14:23

I bought $10 of credits on an OpenRouter account. That allows you to use pretty much any paid model you want (obviously until you use up your credits), as well as giving 1000 daily requests for a pretty generous number of free models. Costs for paid models varies but even the expensive ones will take a fair bit of use to burn through $10. Not a bad option if (like me) you are just dipping your toes in and trying to get a sense of the differences between the various models.

I'm genuinely struggling to find a way in which it improves my life to be honest. I can see that it would be pretty seductive if you were a student needing to churn out essays, but it can't do my work and due to the hallucination issues described above, I don't really want to rely on it for any real world decisions.

I have found it quite good at explaining concepts in multiple different ways. Learning maths in school was a huge struggle for me. I always got to a point in the textbook where I didn't "get it". Maybe a chat with the teacher would be able to get me past it for a while, but more often than not I just got stuck, and I didn't have access to any private tutoring. I think it would have been very useful to be able to repeatedly ask an AI tutor to explain the concept I was stuck on in different ways, and really hone in on the part I was missing.

AliExpress

Shop now on AliExpress (affiliate link).

cddt

1967 posts

Uber Geek
+1 received by user: 1904

#3480039 11-Apr-2026 15:02

I have signed up for Z.ai for a month... using it with pi.dev harness.

On the first task it started thinking... so I went to the bathroom, came back and it was done with what would otherwise have taken me several hours.

notesgnome

130 posts

Master Geek
+1 received by user: 91

ID Verified

Lifetime subscriber

#3480054 11-Apr-2026 16:52

My weird AI experience ...

Using ChatGPT (paid account). I asked it for information about a prompt I was going to use.

It thought about it. Thought about it.

Then returned ... nothing.

I pointed out it hadn't provided any output. It thought about it. Thought about it.

Then returned ... nothing.

I repeated this a couple more times over a couple of days. Then I noticed it had a speech output option. In a wild hope, I clicked it, and it provided to tell me all about the prompt, the pros, cons, etc. It just wouldn't display it in the chat window!

Kiwifruta

1425 posts

Uber Geek
+1 received by user: 337

ID Verified

#3480069 11-Apr-2026 19:01

cddt:

A family member uses google search for "restaurant xyz opening hours" and Google AI gives an answer. We drive to the restaurant, the restaurant is closed. The website had the correct opening hours, Google AI's answer was BS.

I find Google maps pretty reliable for this. For far too many times, I find AI tools to be confidently wrong.

Reanalyse

398 posts

Ultimate Geek
+1 received by user: 311

#3480089 11-Apr-2026 20:01

I use Claude Code, but divert it to using z.ai as the engine. Reduces the cost to US$27.00 for three months, which is OK for me as a hobby.

Initially got it to practice while learing from the Ed Donner Udemy course on Agentic AI, but use is mainly as a hobby (and to be able to speak intelligently on the subject to my Data Engineer son)

Still managed to define and code a few games, and today after getting very frustrated at manually creating a website on a VPS let Claude code rip into it with great results.

But still just a fun hobby,

1 | 2 | 3