Claude (genai) is smarter than me.

[dern]

I buy a battery for my vfr, it’s the one listed so should fit, bought a Yuasa so it should be good. Doesn’t even vaguely fit, it’s way too tall for the seat to fit, loose in the holder and the terminals are the wrong way round. I google it on owners forum expecting to find loads of owners complaining that the specified battery doesn’t fit and nothing. I fire up Claude and it tells me the battery should fit, it’s the right size and specification. I refute this and it gives me the dimensions and they match the battery but it doesn’t fit, not even close. It asks me if I’m sure I have the right battery ytz14s and I check and it is and it tells me that it should fit. It then says can I take a photo of the battery in place and it’ll take a look. I do that and upload it and tells me that I’ve mistakenly installed it with the top of the battery upwards rather than install it on its side as it should be. I try that and it fits perfectly and the terminals line up.

I’ll blame the heat.

[Taff]

In work there is a big push to embrace AI and we were getting a briefing on Claude a couple of weeks ago.

The in house version of chat gpt was described as a switched on high school kid where Claude can be described as the PhD student, and those that know how to use AI properly really like it.

[dern]

I think it’s absolutely unbelievable. Genuinely useful both in work, holiday planning and now in making me look a fool.

[iansoady]

I've used it a few times and it can be both brilliant and notably stupid. Brilliant: I have a compressor that was misbehaving and after detailing the symptoms it correctly identified the fault. It was also complimentary when I fixed the faulty valve rather than buying a new one!

Stupid: I needed to compress the shock absorber spring on the engine mainshaft of an elderly BSA. It told me to use a threaded bar up the middle of the spring. Yes, but there happens to be a crankshaft in the way........

[dern]

It’s not infallible. I just correct it when it’s clearly wrong and off we go again. Another simple use case I just used it for was to identify a bulb by showing it a picture. Huge time saver, especially in this case where it’s a non standard headlight.

[McNab]

It’s not infallible. I just correct it when it’s clearly wrong and off we go again. Another simple use case I just used it for was to identify a bulb by showing it a picture. Huge time saver, especially in this case where it’s a non standard headlight.

But what about when it's not clearly wrong, just wrong and you don't know?

[Cousin Jack]

Just tried it for finding someone to sort out my decking. Says it doesn't have live access and sends me of to Checkatrade et al. Not helpful at all!

Few weeks ago my daughter use ChatGPT to find a place to park in Newquay. The recommended nearest car park doesn't exist!

I am not convinced by AI.

[dern]

It’s not infallible. I just correct it when it’s clearly wrong and off we go again. Another simple use case I just used it for was to identify a bulb by showing it a picture. Huge time saver, especially in this case where it’s a non standard headlight.

But what about when it's not clearly wrong, just wrong and you don't know?

I guess that you'll soon find out. I wouldn't use it for medical advice

[dern]

Just tried it for finding someone to sort out my decking. Says it doesn't have live access and sends me of to Checkatrade et al. Not helpful at all!

Few weeks ago my daughter use ChatGPT to find a place to park in Newquay. The recommended nearest car park doesn't exist!

I am not convinced by AI.

A lot of it comes down to how good the model is that you're using and the free version of chatgpt is shit. I use a paid version of Claude and asking for recommendations for people to fix decking gives me this...

Screenshot 2026-07-13 at 11.45.33.png (1.18 MiB) Viewed 151 times

I wouldn't say AI in general is the problem behind your experiences.

[Taff]

It’s not infallible. I just correct it when it’s clearly wrong and off we go again. Another simple use case I just used it for was to identify a bulb by showing it a picture. Huge time saver, especially in this case where it’s a non standard headlight.

But what about when it's not clearly wrong, just wrong and you don't know?

This is know as AI hallucinations, and they are a big problem so it's the operators responsibility to check the ai output. Couple that with using AI to do specific tasks like code writing or creating an app rather than what should be general Google searches - I'd expect Google maps to have up to date information on where car parks are.

[dern]

Even google ai mode can provide decent answers and that's free (location is different as I didn't tell claude where I was)...

Screenshot 2026-07-13 at 11.53.38.png (471.14 KiB) Viewed 139 times

No idea what AI you used that couldn't even take a stab at it.

[Taff]

I'm not suggesting that all ai tools couldn't take a stab at it, but different tools have their strengths, Google AI is effectively a Google search and that's exactly where is be putting questions like that.
Co pilot is there place where I'd be asking how to use excel or word features because it's tagged onto M365, and so the answers will usually be right with multiple options.

Using an AI tool that's been developed for coding to ask general questions is like using a hammer to drive a screw home, it'll do the job and look right on the surface.

I'm not suggesting all instances of Claude is specifically for coding.

[Cousin Jack]

I used the free version of both Claude and GPT, so no real surprise it is shite.

The problem here is that often there isn't anyone local, or local to them means SW England but not extending as far as Cornwall. Google search is worse, used to be OK but the last few years sees it throwing up paid-for shite even if you search for something specific by name

[Horse]

Even google ai mode can provide decent answers and that's free

... 'ish'.

I know you meant the decking search. But, as a reasonable guide, as a minimum it's a good idea to check the linked references.

I saw the term 'tourniquet conversion'. Knew what a tourniquet is, but not the 'conversion' bit.

Google AI gave a fair description, that it's when bleeding is controlled (eg by direct pressure) so the tourniquet can be removed.

It used the term 'reversible injuries', apparently such as 'amputation'. For geckos, perhaps, but not for most of us!

[Rockburner]

Even google ai mode can provide decent answers and that's free

... 'ish'.

I know you meant the decking search. But, as a reasonable guide, as a minimum it's a good idea to check the linked references.

I saw the term 'tourniquet conversion'. Knew what a tourniquet is, but not the 'conversion' bit.

Google AI gave a fair description, that it's when bleeding is controlled (eg by direct pressure) so the tourniquet can be removed.

It used the term 'reversible injuries', apparently such as 'amputation'. For geckos, perhaps, but not for most of us!

I'd imagine it got the "reversible injuries" bit because it's word association system often finds "reversible" and "injuries" close to, or even, next to each other.

Hoever it's totally unable to understand that when these two words in close assocation also have the word "non" in front of them, then the meaning is completely reversed.

Because it doesn't see "non" , "reversible", or "injury" as works - all it sees is "78d" "98760had7fh", and "oiyiouyhijh" or some other non-sensical gibberish which has no meaning when used in conjunction with each other, it just sees that these pieces of gibberish are often seen close to or even next to each other, and in turn they are close to other pieces of gibberish which it translated from the words you used in your search (ie, in the same sources) so they must be relevant, and therefore should be translated back out of gibberish into whatever your language is in it's answer.

[Horse]

Even google ai mode can provide decent answers and that's free

... 'ish'.

I know you meant the decking search. But, as a reasonable guide, as a minimum it's a good idea to check the linked references.

I saw the term 'tourniquet conversion'. Knew what a tourniquet is, but not the 'conversion' bit.

Google AI gave a fair description, that it's when bleeding is controlled (eg by direct pressure) so the tourniquet can be removed.

It used the term 'reversible injuries', apparently such as 'amputation'. For geckos, perhaps, but not for most of us!

I'd imagine it got the "reversible injuries" bit because it's word association system often finds "reversible" and "injuries" close to, or even, next to each other.

Hoever it's totally unable to understand that when these two words in close assocation also have the word "non" in front of them, then the meaning is completely reversed.

It sort of made sense if you think solely about the blood loss aspect and consider 'reversing' as changing the situation from catastrophic, 'soon to be fatal' to 'blood not flowing'. eg As a binary bleeding yes/no?

[dern]

Even google ai mode can provide decent answers and that's free

... 'ish'.

I know you meant the decking search. But, as a reasonable guide, as a minimum it's a good idea to check the linked references.

I saw the term 'tourniquet conversion'. Knew what a tourniquet is, but not the 'conversion' bit.

Google AI gave a fair description, that it's when bleeding is controlled (eg by direct pressure) so the tourniquet can be removed.

It used the term 'reversible injuries', apparently such as 'amputation'. For geckos, perhaps, but not for most of us!

Right but at the end of the day it's a language model that's a useful research partner, not something to be obeyed. Unless you're the sort of dumbarse that drives in to a lake because their tomtom told them to, you can work out that you can't stick your head back on

[Rockburner]

Even google ai mode can provide decent answers and that's free

... 'ish'.

I know you meant the decking search. But, as a reasonable guide, as a minimum it's a good idea to check the linked references.

I saw the term 'tourniquet conversion'. Knew what a tourniquet is, but not the 'conversion' bit.

Google AI gave a fair description, that it's when bleeding is controlled (eg by direct pressure) so the tourniquet can be removed.

It used the term 'reversible injuries', apparently such as 'amputation'. For geckos, perhaps, but not for most of us!

Right but at the end of the day it's a language model that's a useful research partner, not something to be obeyed. Unless you're the sort of dumbarse that drives in to a lake because their tomtom told them to, you can work out that you can't stick your head back on

I may be being pessimistic, but id say roughly 75% of the population wouldn't disobey their sat nav...

[Cousin Jack]

I NEVER totally trust a twat nav. I have far to many experiences of being sent via totally unsuitable tracks to achieve a 300 yard shorter route.

[Horse]

Right but at the end of the day it's a language model that's a useful research partner, not something to be obeyed. Unless you're the sort of dumbarse that drives in to a lake because their tomtom told them to, you can work out that you can't stick your head back on

Hence ...

, as a reasonable guide, as a minimum it's a good idea to check the linked references.