@PortfolioPlus good idea. I was thinking about starting a Forum for handy AI prompts and hacks to aid investing research.

One thing to consider is that some of the LLMs have a date cut off, so are not always up to the minute. I still early in my learning curve on this.

So, for example, this morning I did a web page data scrape of text and then tried to get AI to tabulate the data. Big fail. It tried but failed.

If we start a forum on this, then those StawPeople who don’t want their feeds bombarded with AI ideas can mute the forum.

Just putting it out there as I don’t want to start something that would be considered a nuisance by members.

Personally, I’m finding AI is really boosting my research productivity but you have to invest in verification because they all still make mistakes - sometimes quite serious.

I was having a muck about with ChatGPT 4o (latest and greatest paid version) on the weekend after feeding it the contents of my SMSF. I asked it to calculate the weighted PE ratio of the top 20 holdings. Now granted, there were probably some things in there that weren't shares but the majority of the things it came up with an assumption of the underlying PE i.e. about 17/20. It also included some holdings with a .01% weight and then its eventual answer of 11.13 was way lower than almost everything in the set of 20.

My first lessons:

• Garbage in / Garbage out - It might not have liked the way I gave it the data (a report out of sharesight)
• It makes shit up - i.e. most of the PE ratios of the holdings were assumed
• It can't do maths - no way the weighted calculation is correct (I suspect it took the weighting in the overall portfolio and just applied that as the top 20 weight)

Still ... looking forward to continuing to see where this stuff might be useful and factor it into my research and analysis process. I suspect a better place to start might be on an individual holding and pull together a bunch of recent news and other analysis.

Love this idea @PortfolioPlus, and agree with the comments from @mikebrisy & @RhinoInvestor

I've definitely found maths to be a weak point for AI, but it presents it's results with a lot of confidence so you definitely need to check things.

There are some models that are specific to maths (i found one here) but haven't tested them yet.

My main use case for these models has been as a tool to summarise and search text (see the recent post on Inoviq). It can save a lot of time.

I find i get better responses with detailed prompts and specific requests. ie. short and vague does not work well.

Thnx for the feedback Rhino - perhaps like Edison and his light bulb, we will work out what to do by finding out what NOT to do. The weekend Aussie had an amusing article entitled 'The Secret to Talking to an AI Chatbot' - advice included (1) tell the chatbot it is an expert (2) encourage the chatbot to do better (3) ask the AI to suggest prompts (4) be genial and don't get mad and (5) encourage it to be methodical.

Frankly, I don't know whether this is a 'stitch up' which was cooked up after an extended session at the pub.

But, I'm going to persist and my first try will be to understand the lingo and key words as used in ASX announcements. My end intention is that on the top of each hour it would deliver me a list of company announcements to peruse and the reasons why. But hey, sitting in the dunces corner on this stuff, I do understand why a few propeller heads might be chortling at my expense. I do hope more Straw people might see some advantage in a community attempt at this stuff.

I might add that I have found perplexity.a1 as a much better searching tool than Google - and there is a FREE version.

Agree. I am finding Perplexity really useful, because it cites all its references, and you get 5 free PRO searches per day. Which is more than I am currently using.