|
Given the recent thread on Financial Planning using ChatGPT, I'm curious what parameters and questions one inputs? I use three different online financial calculators, but I would love to drill down a bit deeper, and be more specific.
TIA |
|
Just don’t is my advice. Actual math is one of the things that LLMs are worst at. You can end up with general advice that might be helpful but for any specific advice you need to know enough to spot the mistakes and by that time what is the point.
You are better off asking bogleheads |
LLMs don't do math. They use calculators like any financial advisor would. The anti-AI crowd operates from a very old knowledge base, meaning take their advice (not the AI's) with a serious grain of salt. |
| I use Gemini. |
|
OP, here's what I did with ChatGBT.
1. I never named where we have our accounts or any other identifying info. To come up with retirement withdrawal strategies, I provided the current balances in all categories (401k, IRA, Roth, HSA, brokerage). I said that we're maxing out 401Ks and HSAs and that we want to max out mega backdoor Roths. I named our income levels in the last few years of retirement. 2. I said current ages, dates when we'll stop working, dates when we'll take social security and the amounts. I named the amount that we'll get in pensions. 3. I named our county and state so that it could figure tax rates. 4. I named our goal to have $x in monthly income during our first 10 years of retirement (when we'll be travelling). And then said to drol it to $x thereafter. 5. I asked it to assume long term returns of 3%, 4%, 5%, and 6% and run the different scenarios to name our ending balances at ages 90 and 95. As it was giving me back information (always with explanations), I continued to tweak and ask for different models or revisions as they occurred to me. It continued to prompt me to see if I wanted to take all this to the next steps. For example, year by year asset balances, which assets to draw from first before RMDs, when to draw from the brokerage in order to avoid Medicare surcharges. It gave me worksheets specific to our situation and numbers for future calulations and an annual check-list. At the end, I cut and paste everything into different documents. Then I told it to delete the entire thing and not to share or store it. I learned so much by doing this. I recommend it. |
On its face it’s just implausible that an LLM will discover all of the various interactions of different rules (like IRMAA surcharges, ACA credits, taxability of SS, Roth conversions, anticipated rates of return of different asset classes, etc etc). There is purpose-built sophisticated software that has problems doing all of that— I just don’t believe an LLM will get it right and in fact whenever I’ve seen someone do this it has made obvious errors. I mean the other day I asked chat to compare 2 raincoats I was considering and it suggested one and I asked why and it suggested the other. That’s the kind of thing that should be in its wheelhouse and it can barely do it. |
The next step would be to copy/paste your outputs into a model like Gemini to validate and look for issues/errors. All foundation models have the ability to research now, but I find it particularly good with Google/Gemini's deep research capabilities. It's like having a council of advisors working with you. |
| I put in my account balances and ask what they will grow to in 10, 15, and 20 years, assuming a 4%, 5%, 6% or 7% real rate of return. Then I ask how much they will grow if I add specific contributions each year, like I use $25k, $50k, $75k, and $100k. Then I ask what I can withdraw annually using a 4% withdrawal rate starting in 10 years or 15 years under each scenario. |
This is really easy to do in Excel and you can check your math more easily |
| I tend to be more specific with questions. Explain basically what I am doing and ask for refinements. It has been really helpful. |
It's easier using an AI assistant. |
What if the real rate of return is zero? |
What if aliens come from outer space? |
OP here. Thanks! I will work on this tomorrow. |
Interesting. Thanks! |