AI as an Investor Assistant

[dixonge]

At this point, would not recommend!

So, recently did some Google searching. I was specifically looking into RSP (Invesco equal-weight S&P 500 fund) and comparing it to SPY for current trends. I am considering moving my allocation from RSP to SPY. For some reason, when you search for comparisons on Google you see results like this one high on the first page:

https://www.nasdaq.com/articles/3-reasons-why-rsp-is-the-better-etf-than-spy-now

So I engaged an AI tool (the Leo AI tool in the Brave browser, which uses the Mistral AI models, or the Llama 13B, a model created by Meta, when you run out of Mistral credits) and asked for 20 and 10 and 5-year comparisons. It stated specific percentages and always showed RSP coming out on top. It quoted Yahoo Finance as the source, so I pulled up a 5-year comparison chart and lo and behold the numbers were reversed! Here are the comparisons form Leo and from my Yahoo chart for the last 5 years:

Leo - RSP: 165.51% SPY: 145.34%
Yahoo Finance - RSP: 60.84% SPY: 83.66%

I mean, those numbers just came from outer space or something! I gave this feedback to Leo and asked it to doublecheck and correct. It did so, very apologetically.

I'd say this was definitely in the 'not ready for prime time' category...

 

[TOOLMAN]

It could be "AI hallucinations"

 

[Gotadimple]

It could be that the AI tool you are using hasn't learned enough 'yet' about how to read Yahoo's site, and how to do comparative analysis. The problem with AI, as many may have heard, is that the engine cannot validate truth from supposition or outright misdirection. IMO the developers haven't yet figured out how to assure the info the tool is learning about is accurate.

As a result the responses become corrupted with false info and if it picks up enough false data that becomes the fact or conclusion it presents to you. The other more glaring issue is that there is bias in much of the information the tool learns leading to incorrect analysis.

Your experience is somewhat harmless because you what to understand the answer you received. But this link discusses three different situations where large institutions are using AI as a way to make a decision. Unfortunately that can lead to outcomes that reflect bias in the data, and could cause harm.

https://datatron.com/real-life-examples-of-discriminating-artificial-intelligence/

 

[target2019]

What was the exact wording of the query?

 

[OldShooter]

"Real intelligence" has never been able to reliably predict the future. I don't know why we would expect more from "artificial intelligence." The issue is that the future is random and random can't be predicted.

For things involving only the past, like real estate comps, AI probably has potential.

 

[Lsbcal]

I recently took a look at RSP vs SPY. For this I use Morningstar charts which include dividends (Yahoo charts do not I believe).

Anyway I will stick with SPY (actually VOO).

 

[target2019]

Copilot. Question and answer. https://sl.bing.net/kbhINZDkY7o

 

[jebmke]

A decent AI bot should have told you that historical returns are not predictive and left it at that. Big miss by the bots.

 

[dixonge]

What was the exact wording of the query?

I can't access my chat history, evidently, but just something like "What is the difference in gains between the RSP and SPY over the last 10 years?" - I did not ask for 'which is better' or any opinionated type of question...

 

[dixonge]

"Real intelligence" has never been able to reliably predict the future. I don't know why we would expect more from "artificial intelligence." The issue is that the future is random and random can't be predicted.

For things involving only the past, like real estate comps, AI probably has potential.

Who said anything about the future? I would never ask an AI about the future. That would be stupid! I was, in fact, asking about the past, about quantifiable numbers that should be rather firm.

Copilot. Question and answer. https://sl.bing.net/kbhINZDkY7o

Co-pilot at least came up with solid numbers! Good to know.

A decent AI bot should have told you that historical returns are not predictive and left it at that. Big miss by the bots.

I wasn't asking it for opinion, just numbers. I'd prefer my AI agents to *not* attempt persuasion or to spout moral judgments...

 

[jebmke]

Who said anything about the future? I would never ask an AI about the future. That would be stupid! I was, in fact, asking about the past, about quantifiable numbers that should be rather firm.



Co-pilot at least came up with solid numbers! Good to know.



I wasn't asking it for opinion, just numbers. I'd prefer my AI agents to *not* attempt persuasion or to spout moral judgments...

wouldn't it be faster just to look on the fund website? Unclear to me what "AI" has to add to this process.

 

[marko]

wouldn't it be faster just to look on the fund website? Unclear to me what "AI" has to add to this process.

...or my go-to: M*. Nonetheless, a very interesting exercise! I've yet to delve into it but find it quite interesting. However I haven't really seen much difference between "AI" and a standard Google inquiry.

 

[target2019]

Who said anything about the future? I would never ask an AI about the future. That would be stupid! I was, in fact, asking about the past, about quantifiable numbers that should be rather firm.


Co-pilot at least came up with solid numbers! Good to know.


I wasn't asking it for opinion, just numbers. I'd prefer my AI agents to *not* attempt persuasion or to spout moral judgments...

Yeah. If you look at the way I phrased the question you'll see why I got a decent result with Copilot (MS). I did try Gemini (GOOG) first, but it told me, "Unfortunately, I cannot directly access and analyze financial data. However, I can guide you on how to perform this analysis and provide resources to help you find the data."

It did then go on to explain exactly what to do. It included the dividend reinvestment question, and so on.

I think when you're interested in a popular topic such as AI you'll hear from some about why it's dumb, can't work, shouldn't know, and so on.

I've read a little, and have upped my AI game a bit. At the core this is dealing with human "knowledge." Therein is one problem. Many boundaries are crossed, and there exists a multitude of bad facts. Now that search engines have gone on for so long, and bad facts are successfully dealt with (more or less), it's time for the next step of evolution - AI.

Every financial question I've asked AI always has a statement at the end about past and future results.

 

[jebmke]

...or my go-to: M*. Nonetheless, a very interesting exercise! I've yet to delve into it but find it quite interesting. However I haven't really seen much difference between "AI" and a standard Google inquiry.

the fundamental problem with the consumer stuff (AI) is that it is using the internet as a training database. IMO well over 3/4 of the info on the internet is garbage so the bots have no chance. All they do is give you better organized garbage. I almost never use search any more for the same reason.

Eventually the industry will shake out and some decent products may emerge. There will probably be a lot of capital destroyed in the process not unlike most bubbles. Strong companies like NVidia will probably survive but could easily lose their luster and a lot of market value.