Two days ago, I asked Google AI a simple math question that it blew. Google now answers the exact same question correctly.
I was stunned to find people defending a blatant math error by Google. It has since rectified its error.
Playing Around with Google’s AI Shows Serious Flaws
On October 25, I asked Google a simple question: “Divide millions by what to get billions?”
For discussion, please see Playing Around with Google’s AI Shows Serious Flaws
Google’s response was idiotic. Google responded “You have to divide billions by 1,000 to get millions. To convert billions to millions: divide the number of billions by 1,000.”
This was idiotic because to convert billions to millions you MULTIPLY by 1,000 not divide.
I was more flabbergasted by the number of math and logic illiterates defending Google’s response than the fact at AI was fooled by a trick question.
Google Corrects Its Error
I asked Google the same question today, plus a similar one regarding pounds and ounces.
Today Google responded correctly, as noted by the lead chart.
Google Two Days Ago vs Today
- Two Days Ago: Google said Divide billions by 1,000 to get millions.
- Today: Google Google correctly said Divide millions by 1,000 to get billions.
Some people claimed I asked an invalid question? What the hell is an invalid question?
Another person said I wanted a conversion but my question did not ask for a conversion.
Excuse me, but my question logically implied a conversion (millions to billions or in today’s new question pounds to ounces).
And please note that Google understood the question was a conversion. To repeat: Google said “To convert billions to millions: divide the number of billions by 1,000.”
No! To convert billions to millions you MULTIPLY the number of billions by 1,000.
How many people want to defend that idiotic statement two days ago now that Google fixed it?
Regardless, I am please to report that Google fixed it’s error, as I stated all along.
I believe someone at Google saw my post and made a correction. If not, that implies an even worse problem at Google: Randomly bad math answers.
I am still flabbergasted by the number of people who flunked this easy test.
“Your first prompt was not clear if you were doing unit conversion. ”
Sorry!
Please note Google used the word “convert”
What about this idiotic response do you fail to understand: ” To convert billions to millions: divide the number of billions by 1,000.”
1: How the flying F can anyone defend that idiotic sentence.
2: Does AI require perfectly worded question?
3: What the hell was even ambiguous given Google understood I was asking for a conversion?
There *is* a degree of ambiguity in the original question, since it was not clear whether you were asking how to convert between “units”, or how to interpret a fixed quantity using two different kinds of “units”.
But it can give a pretty smart-sounding answer if you ask it to derive the equations for quantum chromodynamics.
There are two legitimate interpretations of your original prompt. As much as you deny it, it is not as simple and obvious as you are arguing. I have a Ph.D. in physics and I misunderstood the meaning of your question. That doesn’t make me smarter than anyone else here, but I think it gives evidence that I am not an idiot or idiotic. Considering the number of comments that have argued against your simple conclusion, I think you might admit that legitimately intelligent people can have the same misunderstanding of your intent.
OK so you don’t like the way I phrased my question.
But even if you thought my post was not explicit, it is irrational to say, as Google AI did, that “To convert billions to millions: divide the number of billions by 1,000.”
That is nothing short of idiotic. PERIOD.
And I went over that in comments many time, with people attempting to defend sheer idiocy.
“I believe someone at Google saw my post and made a correction.”
Are you sure it was someone and not a self-correcting algorithm? serious question, answer has implications to the current AI capabilities.
It’s a confusing and incomplete instruction without any prior context, and missing any prompt direction on how it should proceed to clarify any ambiguity.
A: “Hey I just won a million dollars!”
B: “A million is nothing. Talk to me when you have billions.”
A: “Hmm, Multiply millions by what to get billions?”
B: “Multiply millions by 1000 to get billions.
Its great that Google AI is learning. But I’m wondering whatever happened to multiplication? I learned in public school that one multiplies to get larger numbers and divides to get smaller ones. I’d hate to think my public ed is somehow better than the colossus known as Google. *shudder!*
But it might come down to this large language model simply being stuck on a quirk of the English language. Someone further down noted the phrase “express millions in billions”. Likely the Google AI is caught up in trying to rectify that simple little word using what seems like contradictory arithmetic.
Just for kicks, I asked a couple of AIs on what day of the week my birthday fell this year. One said that it was a Tuesday. The other said that, given leap years, there’s a formula, yada, yada, yada, and therefore, because February can only have up to 29 days, that date doesn’t exist. I looked up on the calendar and it was on a Wednesday.
No doubt, Google is absorbing your blog and using it for its AI engine. It learned from you!
Your question was very badly posed.
No it wasn’t, but so what if it was?!
How the flying F can anyone justify this idiotic response:
I would argue that AI should have asked you to clarify instead of assuming. It seems as though we baked in human ego, tendency to assume and narcissism into AI. The irony.
This is getting sad.
Instead of getting defensive perhaps take some time to carefully read the many comments explaining your misunderstanding.
Yes or no question: Do you agree that one billion divided by 1000 is one million?
Yes but how is it relevant?
And in context – if you bothered to read or listen that was the second half of an idiotic explanation by Google.
I am annoyed because people are still trying to refuse to admit Google’s response was idiotic.
So yes, this is sad.
It’s relevant because most commenters thought that’s what you were asking, based on the way you posed the question.
If humans don’t understand your question, why do you expect AI to?
Ask confusing questions, get confusing answers.
I now see it but the question is undoubtably ambiguous to express millions in billions you divide by 1000 . This should have made it clear
To get millions of billions you do need multiply by 1000 but to make it unambiguous ask to convert billions to trillions
Am I missing something.To turn millions to billions you can either multiply millions by 1000 or divide by .001??
The basic confusion is the difference between Millions when used as a unit of measure and millions when you use it as a number of things. For unit conversion 1 Billion = 1000 Million and everyone agrees one has to multiply Billions to get Millions, but if you are DISPERSING a billions of somethings over a million somebodies you do DIVIDE BY 1000.
Your first prompt was not clear if you were doing unit conversion. The terms millions and billions were used more as numbers of things and google tried to answer how to SPLIT billions over millions
AI is at this stage imperfect and to use it you must be very precise in your prompts and Case counts. As does all previous context of prompts enter in the session, and if you are a paid account, all the previous prompts and topics discussed may affect how the AI interprets your words.
It makes mistakes, no question about that but in this instance the issue was more the ambiguity of the prompt.
“Your first prompt was not clear if you were doing unit conversion. ”
Sorry!
Please note Google used the word “convert”
What about this idiotic response do you fail to understand: ” To convert billions to millions: divide the number of billions by 1,000.”
1: How the flying F can anyone defend that idiotic sentence.
2: Does AI require perfectly worded question?
3: What the hell was even ambiguous given Google understood I was asking for a conversion?
Dan W. was the first person in your last post to point out what the error actually was. As soon as I read that, I understood.
Your question to AI was confusing as some readers (including me) didn’t see it as a conversion question at first glance.
I was thinking in terms of how dividing a number by a fraction (like 0.001) would result in a larger number. So I initially didn’t interpret the problem as an equation (with both sides being equal).
What else puzzled me about your post was when you said AI “implies 2 billion and 2 million are the same”. If that part was left out and you pointed out what the error was, that whole post wouldn’t have caused such confusion.
yes that was exactly my interpretation.
Similar to you, I was also analyzing the same way that if you divide by a fraction or a decimal you get a larger number. But of course if you are converting how many millions are in a billion, then there are a thousand millions in a billion.
P.S. I have a bachelors and masters in Electrical Engineering so either I cheated my way through six years of school or else I am mathematically literate. I’ll give myself the benefit of the doubt and assert the latter.
Hopefully my second question today regarding pounds and ounces helped others.
Somehow I suspect pounds and ounces may be easier to see.
But either Google fixed the problem (hopefully), or it can still be tricked by awkward wording, randomly.
Divide millions by what to get billions?
My implied question is this: What do you divide millions by to convert to billions?
You cannot let Google off the hook because it understood I was asking for a conversion!
Google replied ”To convert billions to millions: divide the number of billions by 1,000.” And that’s not a typo because Google gave examples on that basis.
It is not clear, based on the above answer, if Google would have got the answer right had I asked the implied question, because it used the word convert.
Today, google got things correct on the EXACT same wording as I used two days ago.
That said, My explanation two days ago in my post could have been better.
The disappointing thing to me was even after I spelled the error out in comments, people still said I was wrong.
So, was Google wrong two days ago or today?
The answer of course was two days ago.
Actually, Google’s first mistake 2 days ago was presuming you intended to ask the question the other way around.
The question was short and to the point.
The answer was anything but.
based upon the national math scores in schools recently, you should not be surprised by the responses you received.
Maff is hard
Good point. But I suspect the culprit here is English.
Some said my question was ambiguous. No, it wasn’t.
Logically, It was crystal clear.
I believe someone at Google saw my post and made a correction. If not, that implies another problem at Google: Randomly bad answers.
Perhaps a case can be made that my question was “awkward”. But it was not ambiguous, nor illogical, nor improperly worded.
My biggest disappointment was when Papa Dave said I wanting a conversion but not ask for one.
Google fully understood I was asking for a conversion because it used “convert” in a reply, getting its example ass backward.
“ My biggest disappointment was when Papa Dave said I wanting a conversion but not ask for one.”
Lol! We will just have to agree to disagree then. It simply is not worth wasting any more of our precious time.
Cheers Mish!
Not all problems in the world are communication problems, but most of them are.
SAT scores on math and verbal are embarrassing for most people.
https://kdwalmsley.substack.com/p/top-venture-capitalists-tour-china
I should paste some of it, in case people aren’t familiar with Walmsley or doubt it matters enough to click. Here is an excerpt
####
A group of Venture Capitalists visited China and saw firsthand, for themselves, China’s dominance in clean tech. And again it doesn’t matter what industry we put there, shipbuilding or electronics or robotics or cars. China’s advances everywhere make competition difficult, and these VC’s decided it’s pointless to try. They’re shutting down new investments in companies unless the projects are somehow in partnership with firms here. This is a regular feature too, that China has prioritized energy security—along with food security and technological security and transportation and medicine. Their economic system is ruthless and failure rates are high, and China is localizing all the supply chains for everything possible. Western companies can’t catch up.
Now it’s safe to say out loud what these VC’s privately feared for a long time. The investment thesis for their entire industry segment is gone. Western firms cannot compete against the Chinese. That was their conclusion after their trip to China.
My own 2 cents (not Walmsley’s):
No matter how much we might regret our oligarchs voluntarily shipping jobs and know-how to China, the foregoing report suggests China has been forging ahead on its own R&D.
It’s US+EU businesses that might start demanding IP sharing agreements, to benefit from China’s R&D leadership.
Or, maybe US oligarchs expect they’ll just ask their AI to invent whatever they need. IMO that’s possible and would explain some things. (Doesn’t make it true. But it would explain some behaviour, including the talk of reduction in foreign students, H-1’s, and general immigration.)
Stop blaming the oligarchs. Blame yourself. You bought the low-priced Nike shoes from China, and put the US manufacturer out of business.
More: This looks more like a fee market to me than “The Shining Bailout City on the Hill” we got going on over in D.C. and the Marriner Eccles building…
#######
… This section here speaks to another problem we all have, when trying to describe the Chinese economic system. It’s definitely not anything like the Socialism or Communism that we learned about growing up. But China isn’t capitalist either, at least not in any way that these Venture Capitalists are hoping for. The business model that made them rich in New York or London or Berlin won’t work here, at all. In China, scale comes before profits. The state benefits, the consumers benefit, but investors not nearly as much. Beijing is relaxed about hundreds of companies going under, because the handful of companies that do survive here dominate the world. Corporate Darwinism is the rule here, and even successful VC’s are shocked by the extent of it. …
…here again we see the difference in business models. These companies already have proof-of-concept. They don’t need VC money to fund the research and development, that’s already done. Their products are already in the market, generating cash flow. That cash flow gets reinvested to produce at scale, not distributed to private investors in dividends or capital gains, or investment banking fees.
A lot of it had to do with cost of capital prior to Fed f*(kery. Japan’s was low. The US’ was high. Japan could have long time horizons. The US had short time horizons.
The nail in the economic coffin came when Wall Street rewarded speculation more than productivity.
@ Mish, please ask Deepseek. It discusses the math a bit. It shows the formula and solves for it.
Sometimes I just want a fish. But generally, I prefer being taught or reminded how to fish.
Google has the math correct today, with examples.
Grok was correct as was I two days ago.
I would have been surprised if DeepSeek got this wrong.
Ask the exact question and post the answer.
Deepseek displays a couple of math formulas during the steps to “solve for x”. Posting it here won’t do it justice, because of the difference in markup language support.