SIMON BROWN: Cutting through the bot-chat

It is 15 months since ChatGPT arrived on the scene and amazed everybody with how clever it was. That amazement has faded as we’ve seen the shortcomings — the most notable being that it is not so smart and its “knowledge” isn’t current.

I pay for an OpenAI subscription because I have found a bunch of uses for it and want to detail some of them this week.

First, which one do you use? I have GPT-4 with my subscription. It’s also available via Microsoft Bing, though that version does have some limitations. The free OpenAI version is 3.5 and Google has Bard (being renamed Gemini), which is about the same as GPT-4. Overall I find the paid version to be the best and, for me, it’s worth the cost.

In my investing it wasn’t very useful at first. It still struggles with simple compound annual growth rate equations. And having a data set that is not current means it doesn’t know about recent results or developments; GPT-4 Turbo runs only up to April 2023.

That said, I have found ways to make it really useful and it certainly saves me time — but verify everything it says. Chatbots are not above lying (or hallucinating, as it’s called).

One key trick is to first tell the chatbot who it is, as this helps refine the process.

So, for example, I was recently digging into uranium as a possible investment. I began by telling the chatbot it was an investment analyst with expert knowledge in mining and energy. I then asked about uranium supply and demand, and it returned good answers with references that I was able to check before reading further.

Asking about companies involved in either uranium mining or nuclear energy was a bit more of a hit and miss — and it missed a bunch. But when you get one answer for a uranium miner, tell the chatbot it works for that company and then ask it to do an analysis on the competition. This returned some great insights and a wider list of stocks (though not all in the sector are listed).

I have also created my own chatbot and inputted all my columns from the past 15 years. Now I can query things I have written and it operates as a memory bank for myself.

You can even ask it what other investors (specific or generally) think about stocks or sectors, which again leads to useful insights.

The important point is to keep refining and checking. It is a long process, but I find it markedly better than using Google search and extremely helpful for building research.

Last, where it really excels is in the kitchen. Tell the chatbot it is a Michelin star chef and tell it what you have in the fridge and cupboard. Then ask it for a recipe that’ll take no more than 30 minutes to cook — you’ll get some awesome results.