Chat GPT is usually wrong on financial matters
- Thread starter Brendan Burgess
- Start date
Brendan Burgess
Founder
- Messages
- 58,253
The problem is illustrated by the earlier posts in this thread.
Fibernacci got completely wrong answers from Chat GPT.
When two expert practitioners in the area told him that Chat GPT was wrong, he seemed to dismiss them and to rely on Chat GPT rather than on strangers. This is a terrible waste of time. He may as well have insisted that Liam and Ger were wrong because, to quote Ciru, a guy in the pub told him the opposite.
@Duke of Marmalade
You posted a link to a Chat GPT output today without any explanation of the content. It was a difficult technical issue, and people should not be relying on Chat GPT for an answer.
I've had a other look at that one. The main answer to the question asked was correct (but the details accompanying it were completely wrong), but the answer should have pointed out that the question, as asked, would give rise to an answer that would lead the asker to draw the wrong conclusion.
Is growth inside a PRSA tax free when the contributions do not receive tax relief?
Chat GPT answered "yes".
The human answer would have been, "Yes, but the whole fund is taxable on the way out of the PRSA, apart from any tax free lump sum."
Duke of Marmalade
Registered User
- Messages
- 5,578
If I had posted the content without attribution to the bot it would look as if this was my opinion. That would be incorrect as I would need corroboration before I was prepared to stand by the content.
ChatGPT is not a poster herself but I thought that my posting what she thought and attributing it to her did contribute to the thread.
But of course I am fine with whatever course is decided in regard to its use on AAM.
So if I want to ask ChatGPT for some advice, I know how to do that - I would not be impressed if I asked a question here, and someone came back with either "well ChatGPT says..." or worse (much worse!) an unattributed answer from ChatGPT. I think people pose questions on the forum in the hope of receiving advice from people who (hopefully) know what they're talking about, or draw on their own experience if not expertise.
The other part of this - asking ChatGPT a question and getting poor guidance, and then coming here to double check the guidance strikes me as a bit gauche, but it's unlikely we can stop this even with a policy change.
Rather than banning it, I'd advocate discouraging it, with perhaps an indication of escalating responses if someone persists in posting garbage because they're convinced the AI knows better.
Duke of Marmalade
Registered User
- Messages
- 5,578
Very good point. I suppose I was assuming that us ChatGPT folk are still a minority - a bit patronising, my bad. I won't be posting ChatGPT any more.
Fibbernacci
Registered User
- Messages
- 191
Fibernacci got completely wrong answers from Chat GPT.
When two expert practitioners in the area told him that Chat GPT was wrong, he seemed to dismiss them and to rely on Chat GPT rather than on strangers. This is a terrible waste of time. He may as well have insisted that Liam and Ger were wrong because, to quote Ciru, a guy in the pub told him the opposite.
This is completely incorrect. I didn't dismiss anyone and didn't rely on ChatGPT. I found it interesting that AI's reading of the law concluded something against the grain. Laws have loopholes all the time and there is another thread about the government closing a pension loophole for directors in companies etc. I know full well that the pros on AAM know how the system works which is why I asked.
There's nothing wrong with investigating an avenue that's thrown up, it's another thing betting your house keys on something unverified. That's why I also clearly stated you have to verify anything you're given.
You're kind of putting words in my mouth a bit there, probably because written word leaves you to draw your own perceptions.
FWIW, and ban me if you must, but I think you're being quite dramatic and overreacting a tad.
When two expert practitioners in the area told him that Chat GPT was wrong, he seemed to dismiss them and to rely on Chat GPT rather than on strangers. This is a terrible waste of time. He may as well have insisted that Liam and Ger were wrong because, to quote Ciru, a guy in the pub told him the opposite.
This is completely incorrect. I didn't dismiss anyone and didn't rely on ChatGPT. I found it interesting that AI's reading of the law concluded something against the grain. Laws have loopholes all the time and there is another thread about the government closing a pension loophole for directors in companies etc. I know full well that the pros on AAM know how the system works which is why I asked.
There's nothing wrong with investigating an avenue that's thrown up, it's another thing betting your house keys on something unverified. That's why I also clearly stated you have to verify anything you're given.
You're kind of putting words in my mouth a bit there, probably because written word leaves you to draw your own perceptions.
FWIW, and ban me if you must, but I think you're being quite dramatic and overreacting a tad.
This is how you responded to @LDFerguson's correction of the erroneous AI generated info that you posted:
And @LDFerguson isn't exactly an anonymous stranger but an industry professional with a long history of helpful and accurate posting on pension and other matters here on Askaboutmoney.
Fibbernacci
Registered User
- Messages
- 191
Are you telling me that as a layman I should blindly follow advice on AAM without verifying it for myself because some established posters have a lot of respect for some other established posters and know them much better than I do? Really? My assertions are sound - > Don't rely on AI and Don't rely on posters in internet forums, I'll continue to live by that.
I think you and some other folks seem intent on manufacturing a mountain out of a molehill. AAM is great, LDFerguson is great but that will never change the fact that this is an internet forum at its core.
My points have been misinterpreted because you took whatever meaning suited you best at the time, ergo you overreacted. If you need some clarification on what I mean or whether I'm 'dismissing' people you can ask me and pay attention to my response, don't make presumptions.
I think you and some other folks seem intent on manufacturing a mountain out of a molehill. AAM is great, LDFerguson is great but that will never change the fact that this is an internet forum at its core.
My points have been misinterpreted because you took whatever meaning suited you best at the time, ergo you overreacted. If you need some clarification on what I mean or whether I'm 'dismissing' people you can ask me and pay attention to my response, don't make presumptions.
Brendan Burgess
Founder
- Messages
- 58,253
So how do you verify advice?
Check AI? That is ridiculous for an Irish financial product when AI is clearly wrong so often on these issues.
In that case, don't rely on Revenue either because they can and do give out incorrect information and won't allow that as an excuse for taxpayers acting incorrectly on it.
So many seem to see asking chatgpt as being the same as googling something when they are not the same at all
AI is so unreliable if you don't know the true answer to something - see here
www.irishexaminer.com
Basic critical thinking and lateral reading seems to have disappeared, yet have never been more important
AI is so unreliable if you don't know the true answer to something - see here
Terry Prone: Badger-licking test proves AI is not as smart as it’s cracked up to be
'AI has no oil on its feathers, no capacity to laugh at you or itself'
www.irishexaminer.com
Basic critical thinking and lateral reading seems to have disappeared, yet have never been more important
Fibbernacci
Registered User
- Messages
- 191
For me it suffices to come onto an internet forum to gather ideas and inform your own personal strategy. If I was making a major strategic decision, I would do a level of due diligence before committing to something.
For example, I learned here that ARFs are taxed in Ireland when you are resident in Spain. This is a critical point for my plans so I checked the DTA. There you go, trust but verify.
If something were significant enough, with enough wealth behind it. Take professional advice because someone that you can possibly sue won't be too casual about the advice they give you.
You really need to stop with the whole 'Fibbernacci checks advice by consulting AI' line. I didn't suggest that and you shouldn't insinuate it.
For example, I learned here that ARFs are taxed in Ireland when you are resident in Spain. This is a critical point for my plans so I checked the DTA. There you go, trust but verify.
If something were significant enough, with enough wealth behind it. Take professional advice because someone that you can possibly sue won't be too casual about the advice they give you.
You really need to stop with the whole 'Fibbernacci checks advice by consulting AI' line. I didn't suggest that and you shouldn't insinuate it.
Fibbernacci
Registered User
- Messages
- 191
'
I think you're right about this. They didn't respond to me and I thought. 'why would they'.
In that case, don't rely on Revenue either because they can and do give out incorrect information and won't allow that as an excuse for taxpayers acting incorrectly on it.
I think you're right about this. They didn't respond to me and I thought. 'why would they'.
Brendan Burgess
Founder
- Messages
- 58,253
Go back and read what you actually said.
You had a chinwag.
You wondered was it right.
You were told it wasn't by two experts who deal with the issue every working day
But you said you would check with Revenue anyway.
I think you should also get Senior Counsel's opinion on it as well in case everyone other than Chat GPT has been getting it wrong for years.
Fibbernacci
Registered User
- Messages
- 191
Go back and read what you actually said.
You had a chin wag - Irrelevant, dont be cheeky
You wondered was it right - Correct
You were told it wasn't by two experts who deal with the issue every working day - Already dealt with this, I don't know from Adam who different posters are or what their backgrounds are and in any case I wouldn't 100% trust something that was told to me on an internet forum.
But you said you would check with Revenue anyway. - it's a free country, I can check what I want. It's not a crime and there is no reason to drag someone through the mud for it.
I think you should also get Senior Counsel's opinion on it as well in case everyone other than Chat GPT has been getting it wrong for years. - This is a nonsense statement and shows you're not listening to a word I said here.
You had a chin wag - Irrelevant, dont be cheeky
You wondered was it right - Correct
You were told it wasn't by two experts who deal with the issue every working day - Already dealt with this, I don't know from Adam who different posters are or what their backgrounds are and in any case I wouldn't 100% trust something that was told to me on an internet forum.
But you said you would check with Revenue anyway. - it's a free country, I can check what I want. It's not a crime and there is no reason to drag someone through the mud for it.
I think you should also get Senior Counsel's opinion on it as well in case everyone other than Chat GPT has been getting it wrong for years. - This is a nonsense statement and shows you're not listening to a word I said here.
Wandering_Dazed
Registered User
- Messages
- 172
Dare I say the main topic here is AI rather than wading into a certain poster's comments and interpretations of ChatGpt?
Quick thoughts on the main topic, AI is here to stay and there are many AI variants than just ChatGPT. (Gemini, Claude etc etc)
AI is only as good as the data it is fed. I had an argument with ChatGPT 3(iirc) 2 years ago as it insisted I was wrong. That was because it had only been fed data up to 2022.
Questioning and critical thinking is vital more so than ever, whether AI or forum posters. That is super healthy and should be encouraged.
However, it would be really interesting to train a LLM on this forum and see what the results are. There is a wealth of info on here that could help people get the basics right and then more nuanced questions being dealt with by the deep experience and knowledge humans have.
Quick thoughts on the main topic, AI is here to stay and there are many AI variants than just ChatGPT. (Gemini, Claude etc etc)
AI is only as good as the data it is fed. I had an argument with ChatGPT 3(iirc) 2 years ago as it insisted I was wrong. That was because it had only been fed data up to 2022.
Questioning and critical thinking is vital more so than ever, whether AI or forum posters. That is super healthy and should be encouraged.
However, it would be really interesting to train a LLM on this forum and see what the results are. There is a wealth of info on here that could help people get the basics right and then more nuanced questions being dealt with by the deep experience and knowledge humans have.
Fibbernacci
Registered User
- Messages
- 191
Fully agree.
That's an interesting point you make about training an LLM on AAMs knowledge base. How would that happen in practical terms?