David Busoli, director of SMSF Alliance, said a lot has been written on the business process possibilities of using ChatGPT and many advisers and SMSF members may have already experimented with it, but that it should be approached with caution.
“ChatGPT can be used for anything, and with an SMSF, although you can’t administer a fund with it, it can still do various functions,” Mr Busoli said.
“You can put your bank accounts into it and ask it to do calculations or produce a particular report.
“If you are doing analysis on bank statements, you obviously don’t put in the accountant numbers or names of clients, or anything that you don’t want to become a permanent record.
“You can put in a computer code and ask it to a write a formula to do something, or to analyse the computer code and tell you where the errors are.”
Mr Busoli said he has experimented with artificial intelligence and said although it can make things like writing reports easier, it should never be relied on.
“I did a search on an obscure SMSF item of future service benefit, and the ChatGPT version I was using quoted myself back,” he said.
“People, especially advisers, should appreciate that what it finds and presents in a very convincing manner, may in fact be wrong.
“If you are using it to write reports you need to give it very specific instructions and information and ensure you check what comes out very thoroughly.”
Mr Busoli said although the technology can be used as a tool to get information, he believes it is dangerous to do so and said it can get trustees in trouble.
“But for advisers to get the correct wording and modify reports for different situations, it can be a more useful tool,” he said.
“They should always know the answer to what they’ve asked before using it, but it can be used to get some beautiful wording.”
Mr Busoli said although AI technology is gaining popularity in many professions, in the SMSF space it is always best to follow the know-your-client rule, which only comes with face-to-face encounters and conversations.
“ChatGPT learns from, and is trained on, all information inputs. In short, it memorises everything you enter into the system so don’t provide confidential details like tax file and bank account numbers,” he said.
“The tool doesn’t differentiate between private and general information.
“The application seems incredibly smart, but it can’t do everything. Ultimately, you are responsible for your own work. If there’s a mistake, no computer is going to step in and take the blame. That’s on you, so check everything carefully.”
Mr Busoli also cautioned that the free version of ChatGPT is only trained on information predating September 2021.
“This means any updates to legislation or tax codes since then will not be reflected in its outputs but even if you’re using the paid version it can hallucinate,” he said.
“That means it can produce bold, confident answers that are totally incorrect, so don’t trust it – you will need to use your judgement.”
The base data may be from 2021 but every-time someone uses, it potentially stores new knowledge etc. David being given his answer back is funny, but really shoots home the fact that the more you are in the airwaves, the more likely it will be your content being consumed. Whilst it is great if it’s David’s content, the point about the potential for a garbage answer is real. The worst of it is that you get the sense (some) news articles are being written by AI and a human doesn’t even check for spelling! So we are letting this take hold and giving it more credit than, for now, it deserves.