Has anyone else found Google's AI overview to be oddly error prone?
I've been quite impressed by Google's AI overviews. This past week, though, I was interested in what I thought was a fairly simple question - to calculate compound interest.
Specifically, I was curious about how Harvard's endowment has grown from its initial £780 in 1638, so I asked Google to calculate compound interest for me. A variety of searches all yield a reasonable formula which is then calculated to be quite wrong. For example: {calculate the present value of $100 compounded annually for 386 years at 3% interest} yields $0.736. {how much would a 100 dollar investment in 1638 be worth in 2025 if invested} yields $3,903.46. {100 dollars compounded annually for 386 years at 3 percent} yields "The future value of the investment after 386 years is approximately $70,389." And my favorite: {100 dollars compounded since 1638} tells me a variety of outcomes for different interest rates: "A = 100 * (1 + 0.06)^387 A ≈ 8,090,950.14 A = 100 * (1 + 0.05)^387 A ≈ 10,822,768.28 A = 100 * (1 + 0.04)^387 A ≈ 14,422,758.11"
How can we be so reasonable and yet so bad!?
Just invent a 'common' saying and add 'explanation' at the end.
It's terrible. Gemini 2.5 Pro is great, but the AI overviews must be using a smaller model. I hate it when I look up something niche and it smugly tells me that I must be mistaken because there is no such thing. Also it gives annoyingly family-friendly responses to questions that it would be better off not responding to. The other day I was trying to find a Sopranos quote about two kinds of businesses being recession-proof, one of which being "certain aspects of entertainment" (i.e. prostitution) and it was telling me the certain aspects were filmmaking and music because they make people happy.
Why wouldn't they use 2.5 flash first, and then if an identical query is made by lots of people rerun it with 2.5 pro? Sometimes it seems much more error prone than 2.5 pro or even 2.0 even on common searches.
It seems expectedly error prone.
Aside from the general limitations of this technology, Google needs this to be quite cheap if it runs for every request.
There is not a lot of revenue for a single search, and right now the AI results are actually pushing the links people are paying Google to display further down the page.
There was this Reddit post yesterday where it completely makes up being able to plant flowers in Elden Ring https://www.reddit.com/r/Eldenring/comments/1k6hupy/thanks_g...
The AI overview is worse than useless. It either hallucinates things or it treats shitposts as equally valid information-wise as anything else.
Our robot overlords are _terrible_ at anything financial. I'm on a financial forum where, lately, people are _constantly_ posting stuff that they got from The Oracle and asking what they're doing wrong because they don't understand the result, and the answer is inevitably that ChatGPT or whatever fed them plausible-looking rubbish (this is the _real_ AI safety problem; laypeople tend to take the output as infallible, even though it's usually rubbish.)
Though also as a sidenote Harvard's endowment probably wasn't put in a bank account with a flat 3% interest rate for a few hundred years...
They are awful often for me. Examples - recommending installation of packages and software that doesn't exist, or settings changes that don't exist I In applications, etc. They fill the page but it's sadly noise so it cheapens the whole experience when I would have just preferred a link to a page from a person that knows what the hell they are talking about.
> recommending installation of packages and software that doesn't exist
"slopsquatting" is the term coined for this.
Essentially, bad actors are registering these packages and uploading malware. If you happen to just blindly follow the AI, there's a chance your system gets infected.
It's not the AI doing math wrong as a lot of people are commenting. It's the way it's parsing your sentence. When it reads 'the present value of $100' it thinks that today's value of the investment is $100, and it needs to determine what the investment was worth 386 years ago (assuming a 3% interest rate).
Almost like there's a reason we use highly specific languages when telling a cpu what to do
I saw a report via Simon Willison that if you make up a phrase and add "meaning" to the end of your Google search, it'll invent a meaning for it.
His example was "A swan won't prevent a hurricane meaning"
https://simonwillison.net/2025/Apr/23/meaning-slop/
Why would you use an LLM for this? A simple spreadsheet can do this sort of calculation easily and deterministically.
Also, the assumption of '3% interest' is wrong. There are records of stretches achieving 15% returns for several years and reaching 23% in 2007, for example.
https://www.bloomberg.com/news/articles/2005-01-11/harvard-l...
https://www.wsj.com/articles/SB118771455093604172
This was 2 minutes of old school search, no LLM needed.
I don't think they intended to use AI, but tried to search instead and was presented with Googles AI summary / time wasted before the actual search results.
Long term interest rates over hundreds of years are a lot closer to 3% than 15%. You can't extrapolate a few good years like that.
You can't compound 760 1600s pounds for 400 years and get to a dollar amount either. The whole exercise is spurious. That is beside the point.
What I am saying is that asking an LLM to do interest calculation is absurd in itself, let well alone the absurd setting of trying to calculate interest rates across 4 centuries and different denominations.
It would be much more rational, in seeking to understand the growth of the Harvard endowment, to search for factual information about its modern history is my point. And if you wanna do abstract financial modelling exercises just use spreadsheets. Either way LLMs are a hilariously bad fit.
780 compounded by 3% per year for ~400 years is about 100 million by the way. So ignoring all else, off by at least two orders of magnitude.
Of course! I was originally imagining google would give me a website with an embedded calculator. I was most surprised by how everything was beautifully accurate up until the end when the number felt suspicious. (Much less suspicious than the examples I posted, actually.)
You can just search for "interest calculator" there many such sites.
Also, I don't quite get it:
> everything was beautifully accurate up until the end when the number felt suspicious
The LLM generated text about compounding interest over 400 years from early modern british pounds to modern dollars was accurate? How is it possible to be accurate about an absurd operation?
I recently searched for a person and it concatenated the lives of several different people with the same name together like “X is a senator… He is also a professional baseball player…”
I think it is because it is using a "mini" model with the search results as a RAG source so they can afford to use it on every single query. Thus it doesn't know very much and doesn't have much context to work with.
I think I’d call these examples “predictable” failures instead of “odd”.
I recently used Gemini and Google search (with overview) to confirm whether a snack i bought from japan has expired. Used gemini to take a picture of the label written in japanese
One item said 25/7/25 the other one said 25/7/24 as you can imagine I was sure the first one was safe but the second one was confusing.
It told me that it's safe to eat because japanese date format is Year / Month / Date.
I looked up japanese date format in google (with overview) just to confirm. I guess we'll find out. Will report back soon.
LLMs can't do math.
This. People need to manage their expectations.
Why offer a solution then? Seems fairly easy for google to avoid giving the final number?
Expectations can only be managed by someone who has sufficient understanding - in this case Google by not providing the result.
LLMs and tempered expectations, like oil and water
We're giving them calculators though, surely Google could provide a limited set of tools given Search already has a fairly sophisticated calculator.
I've been having my AI stuff successfully do math since early gp3 days with this method— even before "tool calling."
LLMs can’t do math. But that’s a solved problem. ChatGPT has had a built in Python runtime that can do math for years - at least the paid version.
We're all sadly gullible.
We're all in IT. We know what an LLM is. But still we're fooled!?
It's not "AI" - its an LLM that is doing the equivalent of a college freshman padding out a paper for length. Confident, verbose, polished, but ultimately based on little hard reasoning - aka bullshitting. When it's wrong, and if you notice it and call it out, it will happily apologize and "correct" itself with the same well-written prose while making another mistake (or even the same exact one). LLMs certainly have utility, but it's more as generating inputs to some verifying processes rather than as a standalone oracle that competently answers questions.
LLMs are all bad at math. But there are worse ways Google fails.
Like people asked "does Lululemon use <name of some chinese company> to make its products" and Google says "yes", with no source except one tiktok video that falsely claims it to boost sales in face of tariffs. (Ignoring that it's not in the actual supplier list published by lululemon on their site)
Which means basically people would see that tiktok, go to fact check on google if it's true, and google overview will say "yes" (+ paragraphs of text that no one reads) citing that tiktok.
Vicious circle of LLM factchecking. Google used to be immune to it until it started to shove chatbot output to people's faces.