Best AI Tool For Assistance With Physics And Computer Programming?

Click For Summary
ChatGPT is currently being used as a personal tutor for learning Python and solving problems in physics, but it struggles with advanced concepts and accuracy in complex tasks. Alternatives like MathGPT have been found to be less effective, while Wolfram Alpha is mentioned as a potential option, though its effectiveness remains untested by the user. There is frustration with forum dynamics where experienced members prioritize self-sufficiency over direct assistance, leading to inefficient help-seeking experiences. The user emphasizes the need for reliable verification of answers to maintain academic performance, particularly for graduate school aspirations. Overall, there is a call for more supportive and practical guidance in problem-solving discussions.
Ascendant0
Messages
175
Reaction score
38
I use ChatGPT at this point as basically my personal tutor. It has been a huge help in learning Python code, fixing issues when I'd have them, verify it's working as intended, etc. Also helped me with some of my problems in other classes too (it's difficult for me to make it to office hours a lot of the time). Only problem is when it comes to more advanced concepts, sometimes it gets things wrong. I was wondering if there is something better out there that can take advanced physics and math problems?

I have tried MathGPT, and from what I've experienced, it's even worse than ChatGPT when it comes to difficult problems. The only thing I found that might have potential is Wolfram Alpha. But, not sure how decent that one is, and I don't have anything going on to test it out with currently. Anyone have any other suggestions, or experience with Wolfram vs ChatGPT to know how they compare on the more challenging stuff?
 
Computer science news on Phys.org
I think the best you can do is work the problem yourself and give the AI smaller chunks to work on so you can verify if it makes sense.

In the final analysis, whatever you do has your name on it. You have to check everything. Whats nice about code writing is that it's easier to describe an algorithm then look at the result and test on various cases until you're satisfied.

I've used it to generate simpler versions of what I want and then ask it to add a feature until I get what I want or something I can customize further. I'm not using it for work since I'm a retired retiree yipee but when I need something it helps jumpstart my project.
 
  • Like
Likes FactChecker and Ascendant0
jedishrfu said:
I think the best you can do is work the problem yourself and give the AI smaller chunks to work on so you can verify if it makes sense.

In the final analysis, whatever you do has your name on it. You have to check everything. Whats nice about code writing is that it's easier to describe an algorithm then look at the result and test on various cases until you're satisfied.

I've used it to generate simpler versions of what I want and then ask it to add a feature until I get what I want or something I can customize further. I'm not using it for work since I'm a retired retiree yipee but when I need something it helps jumpstart my project.
Thanks. Like I said, I do use it for more as well, as I always like to check my answers after I complete problems. I don't know many students (I'm a non-traditional older student that just came back last semester, so I have hardly any connections other than a few class group chats that people rarely use). So, it's good to have some AI that can verify I did it right, and if I did mess up, where I messed up and why so I can learn and fix it

ChatGPT just can't handle it sometimes. I could try breaking some of those things into smaller chunks, but for some problems, I don't see how it would be possible. But, I'll see what I can do as far as that, thanks
 
Ascendant0 said:
ChatGPT just can't handle it sometimes.
Can you give an explicit example?
 
Remember that the real goal is to learn. Don't get bogged down trying to use ChatGPT to give you the correct answer.
 
robphy said:
Can you give an explicit example?
example.jpg

This right here. The equations themselves were pretty clear-cut imo, and so I knew I had all those right. But with Python, I didn't even know where to start. I wasted about an hour trying to compile it and getting errors before I asked for help. While ChatGPT got the equations correct, it was completely and utterly off with the graphs. Even after I pointed out how they *should* look, what equations are supposed to represent each line being plotted, and explained multiple other things to ChatGPT, it kept getting it wrong over and over, and like WAY off most the time. I tried MathGPT, and it was no better. If anything, it was worse

So, I want something that can verify my work on problems like this. Something that can go through ALL the problem, and solve all questions, so I can verify that every answer of mine is right (and if it isn't, find out exactly where I'm going wrong and fix it). ChatGPT can't do it, and honestly, I don't even feel it's that hard. I just don't know Python much at all yet (only 2 weeks into using it)
 
jedishrfu said:
Remember that the real goal is to learn. Don't get bogged down trying to use ChatGPT to give you the correct answer.
Ok, I'm going to be respectfully completely open and honest here. This website (and those that most frequently assist on it) are WAY WAY too overboard with the whole "you have to do all the work," "don't let anything else do the work for you" "don't check your answers" "show us every single step of work you've done in the last two hours or we won't help you, and don't show us what you wrote out... waste an hour typing all of it in LaTeX or we won't help you" types of things. And yes, I've dealt with every single one of those statements at points in time in the past.

And then when someone does come along that genuinely helps me in figuring out the step that I had been stuck on, one of the higher-ups come along complaining "don't give him any information! don't help him! make him do ALL the work himself!" Then what's the point of this forum if not to help us????

Seriously, it's a waste of our time most the time, and I honestly feel it's a power-trip for the vets. If someone wants to get answers handed to them on here, what should that matter to you? Warn them in a disclosure at the TOP of the channel, not in every single message in every single thread. If they're going to short themselves like that, that's their problem, and they will pay for it when it comes time for an exam, and they fail it, because they never learned how to figure it out themselves.

But, the extreme that many of the vets go to on here gets to the point where they're no longer helpful. I mean I can use our tutor system our college gives us 3hrs/week on (I just don't like waiting in queue on it, and not a huge fan of the interface), I can go to the professor's office hours (not always an option depending on the times), I can go on Chegg, I can use AI, I mean people can use any number of sources for help. And of all those, I can say literally every single one is more helpful for me than asking questions on here (which is why I don't ask questions about homework problems on here anymore - it wastes hours and hours of my time that I don't have doing extra unnecessary work to appease vets). I've had times where I ask questions on here, and all I get is a ton more questions about the problem, with no guidance, which takes up my time. Then, I'll be asked to show every single step of my work, which takes my time. Then, I'll be asked to delve into explicit detail exactly why I'm doing something the way I'm doing it (even if I'm doing it wrong, and it's a waste of my time exploring the wrong direction I'm heading). I've had questions on here that had numerous posts over the course of literally days, with not a single shred of useful information that helped me figure out where I went wrong. And one time, after three days of the weekend with over a dozen replies and no actual help on here for a problem I was trying to figure out, I went to my professor's office, and he made perfect sense of it to me in less than 30 seconds. Literally, 30 seconds, 45 seconds at most. Meanwhile, I put at least 2-3hrs into that thread over that weekend directly (let alone hours outside of that working on the problem itself), all for absolutely nothing.

I got all As last semester. I am in our college's honors society. I am well aware of what help is beneficial to me, and what would be hurtful. I always try to do the problem on my own first, and put a substantial amount of time (when I'm able to) working on them myself. But, if I'm stuck and feel I'm not using my time wisely, that is when I look for assistance to help me through a specific part of the problem I'm stuck on. Then, once I'm helped over that hurdle with an explanation, equation, or whatever it is, I go back to working on it myself.

The problem here a lot of the times is that you guys don't give us direct answers. We're asked tons of questions about information that goes way over our heads, or told to use such and such equation that we've never seen (with no explanation about how the new equation/rule/etc. they just brought up ties into our problem in any way whatsoever, not what step of the problem it is applicable to, etc.), instead of simply telling us the step we're going wrong on in our problem and why. It's impractical, and it's a complete waste of the problem-solvers time, all so that vets on here can feel in control. It's not helpful in most cases. It's extremely inneficient, frustrating, and it leaves the problem solver extremely discouraged that entire time.

Anyway, I know that came off as extremely critical, and it might not be applicable to you directly, and if so I sincerely apologize. I've just been a member on this board for years, and I feel over the years, it got worse and worse, to the extreme now where I don't even bother asking for help on homework problems anymore. Questions on other topics like this, sure, but homework problems, not anymore, because literally any other source I use will help me much faster, and in a far clearer way.

I really feel that board regulation needs to be modified. Again, if people want to cheat themselves out of working on something, that should be their choice, and again, they will pay for it on their exams. The people on here aren't mommy and daddy, they aren't some regulatory college board. It's supposed to be regular, kind, helpful humans on here trying to help others figure things out. But, I feel that when it comes to assistance with problem solving, it's no longer the case here. It's disheartening, as I do love this board for other things and other topics, but for homework, no, it's far, FAR too roundabout and "figure it out yourself" for me

----

Back to my issue personally and my purpose for using it... before handing in my homework, I want to make sure my answers were correct. How else am I supposed to do that if not checking with a resource that has the answers??? If I wait until I turn it in and get feedback from the professor after the fact, that could be after the next exam, so then that hurts me, or they may not give enough information on my homework to help me see where I went wrong, so that doesn't help, etc. Not to mention the markdowns on homework are going to hurt my grades as well, and if I don't maintain at least a 3.7 GPA, it's going to hurt my chances/options for grad school. By the time I turn my homework in, I want to do so confident in knowing I fully understood everything, I was confident in my answers, and there wasn't anything I felt lost on that I just "hope" I get right and keep my fingers crossed.

I don't want to risk those things. I want to understand this stuff inside and out, and if I don't understand something, I just want a resource that will give me an explanation of what I'm lost on or not understanding, not ask me a ton of irrelevant questions or talk over my head with things I haven't seen yet.
 
Last edited:
I'm sorry you misunderstood what I was saying.

Think of it this way:

When taking a timed test, you pace yourself and skip problems that are too difficult, intending to return to them later. However, at times, you may get sucked into one problem, struggling with solving it correctly, only to realize you've spent more time on it than intended.

At that moment, you realize you have lost sight of the goal: To get the highest grade possible by maximizing your test points within the time limit.
 
Ascendant0 said:
I've had questions on here that had numerous posts over the course of literally days, with not a single shred of useful information that helped me figure out where I went wrong. And one time, after three days of the weekend with over a dozen replies and no actual help on here for a problem I was trying to figure out, I went to my professor's office, and he made perfect sense of it to me in less than 30 seconds. Literally, 30 seconds, 45 seconds at most. Meanwhile, I put at least 2-3hrs into that thread over that weekend directly (let alone hours outside of that working on the problem itself), all for absolutely nothing.
So did your professor know if you were in high school or a university graduate student? We don't because you have no information in your profile. Did he know what problem you were working on? We don't. Did he see how you tried to solve it? We haven't.
You expect people to guess at everything and spend more than 5 seconds trying to help. They won't.
And I don't think that you will find an AI program that will guess at everything and solve your problems either.
 
Last edited:
  • #10
jedishrfu said:
I'm sorry you misunderstood what I was saying.

Think of it this way:

When taking a timed test, you pace yourself and skip problems that are too difficult, intending to return to them later. However, at times, you may get sucked into one problem, struggling with solving it correctly, only to realize you've spent more time on it than intended.

At that moment, you realize you have lost sight of the goal: To get the highest grade possible by maximizing your test points within the time limit.
No need to apologize at all. Like I said, that was just a general blanket statement. Not specifically directed at you by any means.

And what you are saying there, that's pretty much exactly what I use the resources like AI for - if I realize I'm spending an unreasonable amount of time stuck on a problem. Don't get me wrong, I TRY before I look at those resources. I'll refer to the textbook, my notes, other chapters, sometimes just step away from it for a bit and come back to it later. But yea, just like you stated with doing a test as efficiently as possible in that regard, that's what I try to use the help for

To me, ChatGPT is like a personal tutor for when I'm stuck. I mean I can ask for clarification on something specific about an equation, or a topic, or a certain type of problem, and within seconds, full explanation. Usually, it's great with the explanations; it's the complicated problems it falls short on sometimes

But anyway, apologies if you took my message earlier as directed towards you. Not my intent at all, and you've been helpful and greatly appreciated. I appreciate everyone's time that tries to help, I just feel we could do it in a far more efficient way
 
  • #11
FactChecker said:
So did your professor know if you were in high school or a university graduate student? We don't because you have no information in your profile. Did he know what problem you were working on? We don't. Did he see how you tried to solve it? We haven't.
You expect people to guess at everything and spend more than 5 seconds trying to help. They won't.
And I don't think that you will find an AI program that will guess at everything and solve your problems either.
Something like that could be very easily addressed on these forums by having someone have some sort of tag attached to their user ID on here (to indicate their current education level). What you touched on though is a much smaller part of (what I feel is) the bigger issue

A solution to a problem is going to be the same ending solution (unless there's multiple solutions of course) regardless of where a person's at in their education. And a high school student might have been exposed to a particular unique equation or method that someone in upper-level undergrad may not have yet. So, I don't think that would necessarily help determine what they can and can't grasp. That's not the source of the biggest holdup though. The problem more so is before people even ask "have you tried this or that?", they immediately ask you to type out every single bit of work you've done, even the stuff you already know is incorrect, just to prove you tried to do it yourself. I mean I'm a very fast typer, but LaTeX, I don't feel it's so easy to do that fast and efficiently, especially when you have multiple pages of scribble trying to work on a problem

But anyway, I feel like we are getting way off track at looking at just a small aspect of it. I get what you are saying, but look through the sections where people are getting help for problems, and skim through those threads keeping in mind what I mentioned. It's glaringly obvious it's not as efficient as it could be imo. Again, if people are going to cheat themselves out of the learning process by just getting answers handed to them, that's their problem, and when test/quiz/exam times come, they'll pay for it. But instead, everyone else deals with things being done extremely inneficiently just in case you're one of those people that are cheating themselves out of learning. Hopefully that makes more sense of what I was trying to state earlier
 
  • #12
The problem with an education level tag is that it needs to be either updated by the member or automatically updated. However, in that case, we still wouldn't know whether you've progressed from high school to college, college to graduate school, or what level of degree you obtained.

Well, you get the idea. It's a trickier problem to automate because members won't be bothered to update the field.
 
  • #13
I've told this story many times.

In the early days of computers, I spent my hard earned money on a Radio Shack Model 1 computer. It was awesome in its day: black-and-white screen, 64-character line limit, and 20 lines per screen. The OS was BASIC, ie, you could type commands or BASIC line-numbered statements and then run to start your program. The audio tape recorder was your program storage media, and it was very finicky —i.e., the volume had to be just right —and it was slow.

My dad worked in the school district as a social worker. One day, he was talking with a school janitor who mentioned his son wanted to learn about computers. My dad shared that I was a programmer and that had a personal computer at home.

So a date was set, and the janitor's son came for a visit. As I started to show him the computer, he interrupted me with a question. “How can I enter my homework so the computer can do it?”

I was a bit surprised because I thought he had some idea of how they worked, but I had to patiently explain that no, the computer didn't work that way. It could do your homework if you taught it to do your homework by writing a BASIC program.

He was clearly disappointed, and after that, I heard he went back home to soup up his car. He was 16 at the time, and he was born too soon because now the computer could do his homework.

The End
 
  • #14
@ascendant

First: I suggest to use Googles Gemini too. For programming i heard Kimi is good, but did not tested it myself. With your question, that ChatGPT makes mistakes, well it is true. I don't know, if you are still reading this, but if so, GPT 5 is much much better at physics and programming. I can suggest some hints and tricks, you may know, but if you don't perhaps helpful.

Work with Gemini (go to aistudio.google, not the standard gemini) AND GPT to crosscheck their answers. Copy GPT into Gemini and say something like: "That seems wrong to me, what do you say?" and visa versa. Or, "I think, it is too difficult for it, can you offer me a step by step plan to solve this, what would you say to the AI-Assisstent here, so he can solve it?" Let them cross-check each other, if something seems important to you but you lack the ability to check it yourself.

Kontextmemory is like a mini-training. Copy the chat and make a pdf from it. Load it back into a new chat and ask; "Is there everything correct in it?" or do your followup work in putting keypoints into a pdf and upload it to go on working. With that you train the ai to follow the work and build up more and more, at the end it is like an evolutionary process like in biology.

Do not influence the outcome: This is a very difficult thing but a must! Do not ask: "Is this correct?" ask "Is this really correct, i think it could be wrong." This way the AI does not try to satisfy you. Be aware, that the AI tries to give you the preferred outcome. Try to formulate your questions and your talk so, that the AI does not really know, what you prefer, or better you convince him, that being correct is better than satisfying you.

Write a persona for the AI. Like: "You are an extremely precise physics teacher, who likes his students to understand everything step by step." or "you always prove your answers twice, because you are a very disciplined teacher." The more context the more specific intelligence emerges. If you work on Relativity or so, "you are a specialist worked decades on relativity. You love the theory and know every bits of it. You admire Einstein deeply and his way of thinking." Believe it or not, this makes a huge difference.

If you have to program, Make the AI to solve the problem, you trying to solve with your script, then look into its thought process, there you have the script which he wrote to solve the problem.

I hope this helps perhaps? Good luck.
 
  • #15
Ascendant0 said:
I use ChatGPT at this point as basically my personal tutor. It has been a huge help in learning Python code, fixing issues when I'd have them, verify it's working as intended, etc. Also helped me with some of my problems in other classes too (it's difficult for me to make it to office hours a lot of the time). Only problem is when it comes to more advanced concepts, sometimes it gets things wrong. I was wondering if there is something better out there that can take advanced physics and math problems?

I have tried MathGPT, and from what I've experienced, it's even worse than ChatGPT when it comes to difficult problems. The only thing I found that might have potential is Wolfram Alpha. But, not sure how decent that one is, and I don't have anything going on to test it out with currently. Anyone have any other suggestions, or experience with Wolfram vs ChatGPT to know how they compare on the more challenging stuff?
I've had a pretty good experience with Bing's Copilot, which I tested against material I knew pretty well.
 

Similar threads

  • · Replies 8 ·
Replies
8
Views
3K
Replies
10
Views
4K
  • · Replies 2 ·
Replies
2
Views
418
  • · Replies 3 ·
Replies
3
Views
3K
  • · Replies 36 ·
2
Replies
36
Views
2K
  • · Replies 7 ·
Replies
7
Views
2K
  • · Replies 5 ·
Replies
5
Views
2K
  • · Replies 13 ·
Replies
13
Views
571
Replies
14
Views
544
  • · Replies 21 ·
Replies
21
Views
3K