CHATGPT GOOD CALCULATING? Randy: Not even close

[Katomino]

Hey, guys,
I would like to confirm whether what chatgpt wrote to me is correct, can you take a look at it?|

My question was:



"I have MnCl₂ + 420 manganese (10 grams of powder dissolved in 1000 ml of water). I would like to raise the manganese level in my aquarium, which has 2500 liters of water, by 0.0021 mg/l. How much of my solution do I need to add to the aquarium?"

My Answer:​

To calculate how much of your MnCl₂ + 420 manganese solution you need to add to your aquarium, we need to go through a few steps. Let's start with the basic calculations:

1. Calculate the amount of manganese in the solution:​

You have 10 grams of manganese powder (MnCl₂ + 420) dissolved in 1000 ml of water. I assume that the entire mass is dissolved in 1 liter of water. So we need to calculate how much manganese is in 1 ml of this solution:

• 10 grams of manganese powder = 10 grams dissolved in 1000 ml of solution.
• This means that in 1 ml of the solution, there is 0.01 grams (10 mg) of manganese.

2. Calculate how much manganese you want to add to the aquarium:​

You want to raise the manganese level in the aquarium by 0.0021 mg/l. Since your aquarium has 2500 liters of water, we can calculate how much manganese needs to be added:

• Amount of manganese = 0.0021 mg/l × 2500 l = 5.25 mg of manganese.

3. Calculate the volume of solution needed:​

We know that 1 ml of the solution contains 10 µg (0.01 mg) of manganese, so to add 5.25 mg of manganese, we need to calculate how many ml of the solution are required:

• 5.25 mg of manganese = 5250 µg of manganese.
• Volume of solution = 5250 µg / 10 µg/ml = 525 ml.

4. Conclusion:​

To raise the manganese level by 0.0021 mg/l in your 2500-liter aquarium, you would need to add 525 ml of your MnCl₂ + 420 solution.

 

[Randy Holmes-Farley]

My experience with ChatGPT and chemistry has been poor. I do not suggest folks ever rely on it. In this case, it is WAY, WAY off. (I edited the title to reflect it)

I suspect a typo in the question, I presume this

I have MnCl₂ + 420 manganese (10 grams of powder dissolved in 1000 ml of water).

really means

I have MnCl₂ + 4H20 (10 grams of powder dissolved in 1000 ml of water).

Assuming so, then the product has a molecular weight of 198 grams per mole, and manganese is 55 g/mole, so this solid powder is 55/198 x 100% = 27.8% manganese by weight.

The solution you made contains 2.78 g/L, or 2,780 mg/l = 2,780 ppm. That means 2.78 mg/mL.

You want to add 0.0021 mg/L in 2500 L, or 5.25 mg.

1.89 mL of a solution containing 2.78 mg/L contains 5.25 mg.

Thus, you need to dose 1.89 mL.

let's explore what CHatGPT would have you do:

Adding more than half of your 1 L dosing solution adds more than a gram of manganese.

Put that in 2500 L, and you get a concentration > 1 g/2500 L = > 0.0004 g/L = 0.4 mg/L = 0.4 ppm. That's 190 times as much as you wanted.

 

[Randy Holmes-Farley]

Let's see if we can dissect the AI brain to figure out its mistake.

Step 1 mistake: it assumed the material was 100% manganese. Must be some other error too , since that would cause underdosing, not the actual overdosing.

Step 2 calculation is correct.

Step 3 CHATGPT BLUNDER EXTREME.
It believes that solution contains 10 mg/mL from step 1.
It carries that incorrectly to step 3 and it becomes "We know that 1 ml of the solution contains 10 µg (0.01 mg) of manganese"
That carrying of data from step 1 to step 3 made a huge error of 10 mg/ml becoming 10 ug/mL. Off by a factor of 1,000!

So CHatGPT made two errors, one small one and one huge one.

 

[Randy Holmes-Farley]

Just noticed this is your first post. Welcome to Reef2Reef!

 

[KrisReef]

Let's see if we can dissect the AI brain to figure out its mistake.

Step 1 mistake: it assumed the material was 100% manganese. Must be some other error too , since that would cause underdosing, not the actual overdosing.

Step 2 calculation is correct.

Step 3 CHATGPT BLUNDER EXTREME.
It believes that solution contains 10 mg/mL from step 1.
It carries that incorrectly to step 3 and it becomes "We know that 1 ml of the solution contains 10 µg (0.01 mg) of manganese"
That carrying of data from step 1 to step 3 made a huge error of 10 mg/ml becoming 10 ug/mL. Off by a factor of 1,000!

So CHatGPT made two errors, one small one and one huge one.

The situation for the future of knowledge is bleak, imo.

The proper text books and much of the "OLD" scientific literature has been removed from Libraries and replaced with "Pop Science" books, imo, ime! This leaves the public with a difficult task of figuring out how to do math and chemistry without proper guidance and oversite and most of us recall in our early education career we had teachers who could help us learn how to do these things.

Now computers and used as the interface for learning and as this practice becomes more ubiquitos and people are taught to rely upon the interwebs for how to knowledge it is only a matter of time until an experimenter with uranium get together a critical mass before understanding the results of their accumulation.

Thanks RHF for all the effort you provide to help provide a path to the correct answers, and often times the answers themselves.

 

[saltcats]

Hey, guys,
I would like to confirm whether what chatgpt wrote to me is correct, can you take a look at it?|

My question was:



"I have MnCl₂ + 420 manganese (10 grams of powder dissolved in 1000 ml of water). I would like to raise the manganese level in my aquarium, which has 2500 liters of water, by 0.0021 mg/l. How much of my solution do I need to add to the aquarium?"

My Answer:​

To calculate how much of your MnCl₂ + 420 manganese solution you need to add to your aquarium, we need to go through a few steps. Let's start with the basic calculations:

1. Calculate the amount of manganese in the solution:​

You have 10 grams of manganese powder (MnCl₂ + 420) dissolved in 1000 ml of water. I assume that the entire mass is dissolved in 1 liter of water. So we need to calculate how much manganese is in 1 ml of this solution:

• 10 grams of manganese powder = 10 grams dissolved in 1000 ml of solution.
• This means that in 1 ml of the solution, there is 0.01 grams (10 mg) of manganese.

2. Calculate how much manganese you want to add to the aquarium:​

You want to raise the manganese level in the aquarium by 0.0021 mg/l. Since your aquarium has 2500 liters of water, we can calculate how much manganese needs to be added:

• Amount of manganese = 0.0021 mg/l × 2500 l = 5.25 mg of manganese.

3. Calculate the volume of solution needed:​

We know that 1 ml of the solution contains 10 µg (0.01 mg) of manganese, so to add 5.25 mg of manganese, we need to calculate how many ml of the solution are required:

• 5.25 mg of manganese = 5250 µg of manganese.
• Volume of solution = 5250 µg / 10 µg/ml = 525 ml.

4. Conclusion:​

To raise the manganese level by 0.0021 mg/l in your 2500-liter aquarium, you would need to add 525 ml of your MnCl₂ + 420 solution.

The thing to remember about chatGPT is it is literally just a fancy predictive text. It's not doing any math, it's not calculating anything, it isn't thinking or parsing information. It's a pretty impressive predictive text in terms of producing responses that scan and sound natural, but that's all it is.

It's just saying "statistically, this word is frequently followed by this other word, so add that one next. That word is usually followed by this other word, so add that one next."

If you wouldn't just press your phone's suggested next word repeatedly to get the answer to whatever you've asked, don't use chatGPT either.

I have a background in machine learning and specifically in natural language processing.

 

[rt456]

The thing to remember about chatGPT is it is literally just a fancy predictive text. It's not doing any math, it's not calculating anything, it isn't thinking or parsing information. It's a pretty impressive predictive text in terms of producing responses that scan and sound natural, but that's all it is.

It's just saying "statistically, this word is frequently followed by this other word, so add that one next. That word is usually followed by this other word, so add that one next."

If you wouldn't just press your phone's suggested next word repeatedly to get the answer to whatever you've asked, don't use chatGPT either.

I have a background in machine learning and specifically in natural language processing.

I’ve been experimenting with using AI tools for reef tank dosing and using it for statistical data for work (published a couple of papers on AI in education). The key is to give it super-clear instructions—it can handle the math, but you have to guide it really closely. If you provide it with a fully worked example, it can often follow the pattern and apply it to a new situation. But honestly, you really have to know what you’re expecting to see and double-check the answers, or you’ll end up with some weird results.

For example, with dosing, I’ve had it work out magnesium and calcium corrections using the Red Sea instructions, took a picture of them and gave it that, then I asked it to verify input the stock concentrations, tank volume, current levels, and targets, and it actually did pretty well—factored in safe changes over 24 hours too. Still, it might not save much time if you can just use a calculator yourself.

It gets trickier with stats. I do a lot of statistical work and tested it against other tools; it tends to make stuff up if you’re not careful. You have to be super precise with what numbers you give it and in what format, otherwise, it goes off track. If you’re not sure what info it needs, just ask, and it’ll usually list out what’s required for the calculation.

One interesting thing is using something like Claude to build a custom calculator app. You can set it up to take the values you input (like concentration, volume, etc.) and it’ll do the math for you. It’s basically an on-demand calculator for specific tasks.

But yeah, if you’re relying on it without knowing what the end result should look like, especially with complex stuff like calculations, chemistry and really most things, you’re asking for trouble. Seen a few students submit AI-derived work that’s complete nonsense because they didn’t double-check.

 

[SliceGolfer]

Or another way to look at it - an inaccurately informed prompt will provide an incorrect response

 

[rt456]

Or another way to look at it - an inaccurately informed prompt will provide an incorrect response

Yes, this is where students struggle as they don’t have the knowledge to know if the stuff it’s spits out is actually correct. If they had done the assessment they would then know what the AI should be outputting for you to confirm before using it for the given task. So in this case you have to know how to dose, how to calculate the concentrations and have a general idea what you would expect the values to be and then it can do the math for you much faster than you could (for most people at least) some are math wizards…

The analogy most students seem to make sense of for this is that you need to know how to drive a car from A to B to get there, using an AI is like driving a faster car so you can get from A to B quicker but you still need to know the route, and how to tell it to get there or else you end up at C or D instead.

 

[Nonya]

Hmm. Could the OP have actually meant magnesium dosing, not manganese?

 

[Randy Holmes-Farley]

Hmm. Could the OP have actually meant magnesium dosing, not manganese?

The ai clearly made two mistakes that are independent of the material choice.

Folks should not ever trust ai chemistry math.

 

[exnisstech]

Interesting that this is the only post by the OP and they have not been here since 3 days after this thread was started. Wonder what thier intent was

 

[Rolliad]

Folks should not ever trust ai chemistry math.

Folks should not ever trust non chemist math for doing chemistry, including AI. <---my version of your statement. Haha!

 

[Rolliad]

Interesting that this is the only post by the OP and they have not been here since 3 days after this thread was started. Wonder what thier intent was

Could just be a result of the holiday. They may not be home.

 

[exnisstech]

Could just be a result of the holiday. They may not be home.

That's a long holiday since this thread was started Nov 7th but possible I suppose. It just seems strange to me to ask this type of question then not follow up

 

[Rolliad]

100% fair!

 

[PotatoPig]

The ai clearly made two mistakes that are independent of the material choice.

Folks should not ever trust ai chemistry math.

Just adding some backup to this: Recently ChatGPT couldn’t count the number of “r’s” in Strawberry.

It doesn’t know math. It has math containing stuff in its database somewhere, and the algorithm reaches into that, but it seems it does so from a language pattern standpoint rather than any real mathematical software coded in. IOW - it returns “math-like outputs”

I wouldn’t trust it for pretty much anything technical or put any faith in any factual statements it churns out.