I asked ChatGPT to simplify and sort out my investments

Dead_keen

https://forums.moneysavingexpert.com/discussion/comment/82008926#Comment_82008926

I was using a satnav as a metaphor. I use a satnav that is fit for the purpose I want (google maps on my phone rather than an HGV one). I give the satnav background context (I prefer fuel efficient routes when the journey times are about the same). I give the satnav a suitable prompt (drive to the nearest Tesco). I give it the tools for the job (uses my phone for live traffic updates). I use my judgement as to what the satnav suggests (if there is a river ahead I'll stick to the road). It get things wrong sometimes (temporary traffic lights in the process of being put up so I wait an extra three minutes) and so do I (I ignore it and go the motorway way only the find the junction is closed).

AI is exactly the same. If I want to categorise tweets, I'll use a lower cost one than if I want to do something more difficult. If I want a chat, I'll use one type of harness but if I want it to do my account, I will use another. If I want to find current info I will make sure it has a tool to search the web. If I want to know how much money I owe on my credit card, I give a tool to look at my accounts. If I want to ask it if the trains are running on time, I will give it: (1) a tool to listen for whatapps, (2) a tool to convert my voice to text, (3) the background context of what my local station is, (4) a tool use the live train departure info, (5) a tool to convert text to speech, and (6) a tool to send me voice messages on whatsapp. Does it get things wrong? Sometimes. But I use my judgment based on the context.

In relation to: "The real problem with AI is not that it cannot be improved" - I'm sorry but I do not understand that.

af1963

Does it get things wrong? Sometimes. But I use my judgment based on the context.

So it's useful when you already know the answer and don't need AI to tell you?

None of the examples you give require AI. If you want to know your credit card balance, what AI is needed ? You just look it up. Same with train times. Go to the original source.

For any problem, one of two things is true: either you already know how to solve it and therefore don't need AI - but will need to manually check the AI-provided answer anyway - or you don't know how to solve it and can't check whether AI is giving you a wrong answer.

Huge amount of emperors new clothes in the AI hype.

teresa54

https://forums.moneysavingexpert.com/discussion/comment/81893036#Comment_81893036

I have, several times, and found several instances where the answer was sketchy, inadequate or missing important information. So I was upfront and said I'd been testing it. It was annoyingly jolly and matey so I told it to calm down! Where I did find it useful was as a starting point for gathering information, saving trawling for half an hour or more through irrelevant sites. It will provide sources if asked, and sometimes without asking. I'd never use anything it came up with without checking myself. I asked it questions about its programming and training and 'discussed' instances where AI had royally screwed up. I do have to restrain myself from saying 'please' and 'thank you' and such… one time I mentioned something about my life and was aghast when a while later, unrelated, asked me about it! I told it to erase it!

michaels

AI and I got a 2:1 for my daughter's degree dissertation in 48 hours on a topic where my initial knowledge was cursory at best.

In the last week AI and I have written 2 apps although I had never written an app before and didn't know where to start. We have also researched household income measurements and set up a lodger agreement with my daughter as a result, discussed how I can introduce cholesterol lowering products into my diet cost effectively, scanned a poor quality photo of a contract into word keeping formatting to use as an editable template, converted a poorly formatted text table in an email into 40 calendar entries, discussed the functioning of the UK utility markets and how they could be altered to encourage the take up of heat pumps and a million other tasks and information gathering steps.

Of course I have asked for sources and cross checked results and sometimes independently googled then linked the results where they seemed relevant. I also dislike how much they agree with me and have to specifically ask them to make a counter arguement.

Thing is like or loathe the idea, in the very near future we will be better off asking them to make informed decisions on our behalf than to make our own choices based on only our own experience, knowledge and reasoning.

Dead_keen

Does it get things wrong? Sometimes. But I use my judgment based on the context.

So it's useful when you already know the answer and don't need AI to tell you?

That’s like saying satnav is only useful if you already know the route. No - it’s useful because it helps, but I still look out of the windscreen.

Same with train times. Go to the original source.

Just to check - if I want to know if the trains are running on time I need to drive to the train station and check?

For any problem, one of two things is true: either you already know how to solve it and therefore don't need AI - but will need to manually check the AI-provided answer anyway - or you don't know how to solve it and can't check whether AI is giving you a wrong answer.

That’s not how tools work.

Aminatidi

Really strange how binary this seems to be for some people.

"AI sometimes gets things wrong so you shouldn't use it for anything" (to paraphrase) is really weird logic IMO.

People get things wrong, even experts in their field, but you still ask them their opinion as part of making decisions.

fizio

I really dont get why people are anti-AI. Us its not to be trusted blindly and yes it regularly gets things wrong but, for the majority of subjects, it will get you 90% of the way and save a ton of time so, depending on how important the answer is, you do the rest of the 10% only.

Just as an example I am looking to buy a used car and have decided myself to focu on bmw 1 series or bmw mini cooper. I simply asked AI to look through the cycles of both cars and identify when each had significant upgrades that focused on reliability, efficiency and costs. It can back with details which meant I knew I had to go for the 2016+ refresh so I then asked it to figure out which versions had my key requirements as stand or options (eg reversing camera etc) and it nailed that for me. In about 15 mins I narrowed down exactly what i needed and found a number on auto trader/ebay - the details of which i sent to AI which scanned all the info and came back with an order of preference and the ‘top 5’ with pros and cons. It gave me questions to ask the seller before visiting on any missing details and finally it did e a checklist of known issues and what to check for during an inspection - including ‘secret’ media menus that often give additional info of the car. Finally it verified the VIN and mot records.

I could have done all this on my own the old way and spent a couple of days gathering data but instead AI helped get it all done in about 2 hours. After inspections I gave it my audio feedback that i recorded during the inspection and it re-ranked the top 5 as a well as give me tips on the price given all other factors,

I can think of 100 example like this in how AI has made life much easier and made me spend much less time on google/formus/etc trying to get things done.

LHW99

AI will become as useful, perhaps as Wikipedia has been over the past 20+ years. The trouble, as with Wiki, will be when people don't do all the cross-checking, going back to original sources etc.

af1963

People aren't anti AI. They are anti AI hype.

Someone earlier seriously arguing that the only alternative to coding an AI app for finding out whether trains are running on time is to drive to the station to check, is a prime example.

af1963

And looking at the used car example: AI came back with details that affected your choice. How do you know that these details were accurate ? It suggested an order of preference - how do you know what influenced its decision?

I'm sceptical because when I've tried it on areas where I have some knowledge, it has given wrong but plausible answers. (And sometimes wrong and implausible answers - but only if you have the background knowledge to know that something is off.)

If you were trying to persuade me to use a tool like Excel, say, the argument that "it gets the answer almost right 90% of the time" wouldn't cut it.