Currently ChatGPT is giving wrong financial advice / education, because that wrong understanding is pretty much ALL the prose found in the media. So Chat is doing correctly what it is trained to do - reflecting the common prose understanding. Yet it is provably wrong with basic math.
IMO there needs to be a way to notify OpenAI when this is happening, so that a pop-up can warn users that Chat’s response may be wrong.
I’m sure mine is not the only situation where prose trumps math in Chat. I found this error when asking Chat what the net benefits of the tax-shelter accounts are (Trad vs Roth in the USA, and RRSP vs TFSA in Canada). When explaining their benefits, almost without exception, ‘experts’ claim the RRSP/Trad contribution’s tax reduction is a benefit and a reason to contribute. They claim that taxes on profits and growth are only deferred but taxed at withdrawal.
In Chat sessions I have repeatedly convinced it of its errors and taught it the understanding proven by math: a) The contribution’s tax reduction is never a benefit. It is more like the government funding the account with its own money, for you to invest for them alongside your own after-tax savings. b) Profits are never taxed, not the profits earned by your after-tax savings or by the government’s funding. c) The withdrawal tax at the same %rate as at contribution, is always exactly funded by the government’ own funding at the start (plus its profits).
d) Which leaves intact your after-tax savings plus its un-taxed profits. That results in a net benefit always exactly equal to the same benefit as from a Roth/TFSA. e) The RRSP/Trad has a possible bonus/penalty equal to the multiple of the ($drawn) *(difference in tax %rates between contribution and withdrawal).
But Chat does not access SSRN papers, or Wikipedia, or my website (recently closed) or Quara. Chat has enough math that I can prove, and get Chat to accept that it is wrong, over and over again, but it never ‘learns’. Or my voice is over powered by the full-throttled advisor industry. There needs to be a warning that Chat reflects popularity, not fact.