This is probably a dumb, self-inflicted privacy leak, but as an experiment I asked ChatGPT to look at the last 5 months of bank transactions. No major surprises: we spend too much on eating out, cell phone plans, and streaming services.
@manton The question I'd now have is: Did it actually do the maths, or does it just give you the answer most people who ask that question get? Have you done a cross-check?
@uliwitness I glanced through the data and it seemed about right. It actually did the math. So I could ask it about streaming services and it knows to look for Netflix, HBO, and similar keywords in the transactions, then add things together. (This was a reasoning model… I can’t remember which one because I deleted the transcript as a precaution.)
@manton Good to hear. The more people rely on LLMs, the more I worry because often people take wrong answers for gospel and I'm having trouble explaining to them why the answer they got can't be correct.
That some models work more reliably there will be very helpful.
@manton @uliwitness ChatGPT told me last night that Timothy olyphant played Hickok in Deadwood until I called it out. I don’t trust this tool.
@markstoneman Yes, I should’ve used a local LLM. My machine can’t run something at the level of the models in the cloud, but it would be fine for this.
@WiredDifferently I don’t have the transcript anymore, but essentially I asked it things like: where are we spending money that we could trim out, summarize some specific category of expenses like streaming, how much do we spend on groceries, etc. It was pretty good at flagging expenses that were “essential” and separating them as things we don’t want to cancel.
@manton I’ve done this too, it’s great! It can also answer questions regarding retirement planning that are super helpful to have contextualized. Obviously, double check any action plan but still, the utility is very helpful.