It happens, I once ask a question about a spinning wheel and the rpm required to have 1G. chatGPT started a couple of calculus and in the end answered that to have 1G, the diameter should be 7 times the radius. I answered back “this does not make sense, a diameter is by definition 2 times the radius”, it apologized and redid the right calculus :)
This is because LLMs do not inherently understand math. They stick characters together that are likely to go together based on the content they were trained on. They’re literally just glorified autocorrect.
If you want a tool that can actually do math from natural language input, try WolframAlpha.
I am so glad to see someone pushing Wolfram Alpha. It’s frustrating that more people don’t know about it, when it’s been around so long now.
I paid like $5 for the Android app (now WolframAlpha Classic) like 10 years ago and it’s been worth every penny. I use it for anything that needs complicated unit conversions.
The Android app is incredible. WolframAlpha has a premium subscription, but I don’t get why anyone would pay for it when the app includes all the same features.
Wolfram was completely indispensable for me during college. I always meant to get into Mathematica as well, but never got around to figuring it out.
Calling the inputs to use wolfram alpha effectively natural language is very generous. I could never get the hang of it
That’s why I can’t ever trust what it says. No conviction. I’m like stick to your guns chatGPT.
This is what people don’t understand about char gpt. It’s not a tool for accuracy, even the company who made it says that. Then idiots come in and say “see it does math wrong! And it can’t get a fact right! Only a moron would say this is the wave of the future!” And don’t get me wrong. Google added it into their search engine because of the hype and low and behold, the LLM was inaccurate.
What is chat gpt food for? Creativity and abstracting. An LLM model is really good when you need a list of 10 terrible names for a dog food company to get your brain thinking. It’s good at helping someone outline a story they want to write or make their email sound more professional. It’s useful as an aid to help someone plan a schedule. The best use of chatgpt is to work with it, not try to get it to do a task without you. An LLM works best with piecemeal feedback and someone knowledgeable in the subject that can vet the answers it’s giving.
I highly recommend GNU units for stuff like this. It’s insanely powerful. See the following comment for its immense versatility
Thank you
Edit: looks like I’m switching to GNU Units
Use Wolfram Alpha, which is a mathematics engine first and text parser second (and it shows: the math is flawless but it wouldn’t understand the query; both need to be asked separately: 1/2). ChatGPT performs similarly this time but I wouldn’t trust it to expand a polynomial because there is very high chance that it would hallucinate some terms.Of course, any calculator will do for this, it’s easy to verify that 2÷3×14 = 14÷1.5, no need to have a server run a billion times more complex calculation.
isn’t it funny how humans will end up using a trillion times more energy so that the calculator can be less reliable, but acts like a person
