Can ChatGPT write an Eulerian Fluid Simulator in 200 lines of code?

In this video I experimented with ChatGPT to see whether my channel is now obsolete :-)

[patryk]

It doesn't get stuck, but it has a limit of tokens it can produce per user request. Sometimes I just ask it to "continue your code" to get rest of the answer.

[johnlin]

"Sure, I can do that!"
Gets stuck...

Sounds like OpenAI trained GPT to write code the same way I do.

[TenMinutePhysics]

Thanks for the hint about the possibility to generate more code. I might do a follow-up video and discuss some more findings... and if somebody gets the flipper or tutorial 3 working, let us know in the comments

[tekperson]

I found this channel recently, and it's become one of my favorites. I ran the first several problems from Project Euler through chatgpt3, and the results were mixed. More complicated requests give either mixed or poor results. That's very consistent with Ten Minute Physics results. I have used chatgpt3 to help prototype a technical presentation. The results are very wordy and often a little off base - but speed up initial development. Frankly, there needs to be more transparency about where the training material comes from and what restrictions the code it was trained on has. Normally licenses propagate with code, but somehow this short circuits that. This makes using chatgpt3 for business applications a bit iffy.

[MrButkam]

Keep up the good work ! Thank you for your valuable videos

[rodai]

to do more extensive code, just ask chatGPT "complete the code snipped where you left off in last answer", you can do it multiple times, and manually merge the code to run! it works!

[LordCogordo]

It has a limit of tokens, whenever he gets stuck just ask "keep writing" and he will

[bengineer_the]

If you ask it: to continue from the line: "contents of last line" then it will produce the rest of what it thought of. :)

[plasid]

Its possible make this using material nodes in UE5?

[compphys]

Something that's not trivial and cannot be found on the net (easily, bouncing balls are all over the place) to copy/paste cannot be generated (correctly, that is). I tried to get from it a function that computes the Clebsch-Gordan coefficients, got a function that spat out -nan. It's definitively far from a silver bullet.

[Ghostrider-ulxn]

ChatGPT absolutely can write complex simulations, I'm a physicist and I've done it. You just need to tell it to break the task into small steps and write fully completed blocks of codes for each steps. In other words, tell it to modularize the code and write blocks of codes for each steps. Not only will it comply, its accuracy will improve significantly.

[WizardOfAILand]

LOVE THE GREAT
Could you do: Moving Least Squares Material Point Method ????

[nielsdaemen]

You should try it again with the newest GPT4 version

[ksitau]

I think you need to write "continue" to get remaining code

[zebfross]

ChatGPT is way more useful to me than Dall-e 2. Such impressive output!

[bakmanthetitan]

Amazing video. Like the other commenter pointed out, there may be ways to get around the output limit. I would love to see a follow-up.

[samsaraAI]

I tried many easy things and it is not working for serious things.

[haluk]

You shouldn't expect semantics from a syntax machine

[takayamayoshikazu]

If chatgpt has such low processing capability, I can't use it for my engineering work...

[johnpelitidis]

This scares me... who will own and control this technology?