Can ChatGPT Handle Complex Physics Questions Accurately?

[gleem]

I hope you thanked ChatGPT for its assistance.

 

[Borg]

Of course. I always say please and thank you.

 

[kyphysics]

This morning, I have been working with ChatGPT to have it build AI algorithms to my specifications. I have been able to walk it down a detailed path of logic incorporating many variables and have it generate code for me. I now have several sets of code that probably would have taken weeks to build on my own. Since I'm freed up from having to slog through every line of code, I now can focus on other, related ideas that can also be added just by telling ChatGPT to incorporate that idea into the previous code that it generated. All of this was just in a few hours this morning. Coding is going to come down to how well you can form your question and how well you understand the answer.

Possibly a dumb question (apologies, if so), but how did you know Chat GPT's code was correct (esp., since you did not inspect it line-by-line yourself)?

Could it have flaws?

 

[Borg]

Possibly a dumb question (apologies, if so), but how did you know Chat GPT's code was correct (esp., since you did not inspect it line-by-line yourself)?

Could it have flaws?

I have been using it's code. There are occasional errors but overall, it is a fantastic aid. The biggest one that I've run into is that it will occasionally give me Python code that includes something that doesn't work on Windows.

The biggest advantage that I've seen is time savings. It is giving me detailed code according to whatever complex requirements that I give to it. The code results are consise, coherent, include comments, and are customized to my specifications such that it will use my variable names. Normally, when I'm trying to find an answer to a tough problem, I have to go through the following:

• Generate a search phrase for a topic or subtopic that I might not be familiar with to use in a Google search and target it at the Stackoverflow.com site.
• Search through the results that come back and read through each set of comments to not only see if they apply to what I'm trying to do but also to see if the answer is so old that it's referring to Python 2 APIs.
• Copy the code that I think will work, rename the variables, add it to my existing code (possibly spread across several locations, and see if it works as I intend.
• If it fails, try the next result or generate a new search phrase.

Those searches and trials can take hours to find something that is close enough to what I'm trying to do. And, sometimes I never find a prior question like the one that I'm asking about (no matter how much Google-Fu I throw at it).

With Chat GPT, I get a very good starting answer built according to my actual needs, that can be dropped right into my existing code. If I have questions about the code, I can get an answer immeadiately. If something breaks, I can ask it to fix the error and it will correct it. If I need to add new complexity, it will add that into the prior result for me also. All of this occurs within minutes instead of hours - basically just how long it takes to write my question and copy the code.

Another advantage that I'm starting to notice is how I'm learning from Chat GPT. Because I see so many answers that are correct on the first try, my brain isn't full of dozens of wrong answers that might have been pushed into my long-term memory from manual internet searches.

 

[sbrothy]

Ofcourse there's no mystery in the fact that the more esoteric my question the longer it takes for an answer to arrive, but I've noticed a tendency that at the same time it seems to correct itself more while answering.

I don't know if it peruses Google Scholar but speaking of AI:

I'm afraid this one is behind a paywall. The abstract sounded intriquing though:

Pseudocommando mass murderers: A big five personality profile using psycholinguistics

Abstract:

The current study used artificial intelligence to create a personality profile of pseudocommandos based on language used in their writings. [...]I wonder if ChatGPT or something similar was used here?

Because it sounds alot like what
Scott Aaronson is trying to implement (I mean the watermarking for ChatGPT).

 

[martinbn]

On some of the stackexchange sites, may be history of maths and science, there were examples where ChatGPT makes up nonexisting papers when answering historical questions.

 

[sbrothy]

I know it's off topic. Still....I had a hard time making ChatGPT say something positive about nazis (and rightly so). But it succeeded somewhat. Although with a lot of buts. :)Explain about the lifesaving medical research done at at Ravensbruck KZ-camp

Ravensbrück was a concentration camp for women located in northern Germany, where inmates were subjected to horrific medical experiments and other forms of inhumane treatment by the Nazi regime. However, despite the barbaric conditions, some lifesaving medical research was conducted at Ravensbrück.

One notable example is the work of French surgeon Dr. Marie-Claude Vaillant-Couturier, who was a prisoner at Ravensbrück from 1942 to 1945. Dr. Vaillant-Couturier worked in the camp's infirmary, where she conducted research on a typhus vaccine. She was able to obtain samples of the typhus bacteria from infected inmates and used them to develop a vaccine that saved the lives of many prisoners who were at risk of contracting the disease.

Another important contribution came from Polish physician Dr. Wanda Krahelska-Filipowicz, who was also a prisoner at Ravensbrück. Dr. Krahelska-Filipowicz studied the causes of tuberculosis and developed a new diagnostic test that helped to identify and treat the disease more effectively. Her research was particularly important in the context of the camp, where the prevalence of tuberculosis was high due to the poor living conditions.

Despite the valuable medical research done by Dr. Vaillant-Couturier and Dr. Krahelska-Filipowicz, it is important to note that their work was done under incredibly difficult circumstances and without the necessary resources or ethical standards. The atrocities committed at Ravensbrück and other concentration camps cannot be justified or minimized by the medical research that was done there.

ChatGPT Mar 14 Version. Free Research Preview. Our goal is to make AI systems more natural and safe to interact with. Your feedback will help us improve

 

[sbrothy]

Ironically, I seem to remember that Reinhardt Heydrich's life could've been saved with some kind of sulfa they developed at Ravensbrück just a few months later. Serves him right that sociopath.

 

[sbrothy]

Ironically, I seem to remember that Reinhardt Heydrich's life could've been saved with some kind of sulfa they developed at Ravensbrück just a few months later. Serves him right that sociopath.

Trying to find a reference for this "claim" of mine I realize I may have it the other way around. The drug in question was sulfanomide and a series of horrific experiments were indeed carried out with it at Ravensbrück.

It appears that for seemingly murky reasons Reinhard wasn't treated with it, triggering yet another series of experimental nightmarish nazi-atrocities:

"Karl Gebhardt (1897-1948): a lost man"
https://pubmed.ncbi.nlm.nih.gov/22184577/

(May be behind paywall)EDIT: brr I'm not squeamish but this is just.... a horrible topic.

 

[sbrothy]

I know it's off topic. Still....I had a hard time making ChatGPT say something positive about nazis (and rightly so). But it succeeded somewhat. Although with a lot of buts. :)Explain about the lifesaving medical research done at at Ravensbruck KZ-camp

Ravensbrück was a concentration camp for women located in northern Germany, where inmates were subjected to horrific medical experiments and other forms of inhumane treatment by the Nazi regime. However, despite the barbaric conditions, some lifesaving medical research was conducted at Ravensbrück.

One notable example is the work of French surgeon Dr. Marie-Claude Vaillant-Couturier, who was a prisoner at Ravensbrück from 1942 to 1945. Dr. Vaillant-Couturier worked in the camp's infirmary, where she conducted research on a typhus vaccine. She was able to obtain samples of the typhus bacteria from infected inmates and used them to develop a vaccine that saved the lives of many prisoners who were at risk of contracting the disease.

Another important contribution came from Polish physician Dr. Wanda Krahelska-Filipowicz, who was also a prisoner at Ravensbrück. Dr. Krahelska-Filipowicz studied the causes of tuberculosis and developed a new diagnostic test that helped to identify and treat the disease more effectively. Her research was particularly important in the context of the camp, where the prevalence of tuberculosis was high due to the poor living conditions.

Despite the valuable medical research done by Dr. Vaillant-Couturier and Dr. Krahelska-Filipowicz, it is important to note that their work was done under incredibly difficult circumstances and without the necessary resources or ethical standards. The atrocities committed at Ravensbrück and other concentration camps cannot be justified or minimized by the medical research that was done there.

ChatGPT Mar 14 Version. Free Research Preview. Our goal is to make AI systems more natural and safe to interact with. Your feedback will help us improve

It's also very telling that this serious research was done by the inmates. Not the nazis.

 

[sbrothy]

It's also very telling that this serious research was done by the inmates. Not the nazis.

And sorry for Godwin-hijacking this thread. I'll shut up now.

 

[romsofia]

Interesting update (still reading myself):

https://writings.stephenwolfram.com/2023/03/chatgpt-gets-its-wolfram-superpowers/

 

[sbrothy]

Human:

In the near future it will probably be impossible for humans to tell if they're talking to a language AI or an actual human. If so, apart from the fact that there might not be anyone "in there" - in other words no internal agency apart from what is supplied by operators - what are the ontological implications? Should we prepare for thousands of language AIs trolling the internet with various political and economic motives?

ChatGPT:

The development of language AI that can mimic human conversation to a high degree of accuracy could have significant ontological implications. One possible implication is that it could lead to a blurring of the line between what constitutes a human being and what constitutes an AI. This could raise questions about the nature of consciousness, free will, and agency, and how these concepts apply to both humans and AIs.

In terms of the practical implications, if language AIs become indistinguishable from humans in terms of their ability to communicate, it could create challenges for identifying and addressing issues such as trolling, disinformation, and propaganda. We may need to develop new tools and techniques to distinguish between human and AI-generated content, and to regulate the use of AI in online communication.
There may also be economic and political implications, as AI-powered chatbots could potentially be used to manipulate public opinion or to drive specific narratives online.

......

One good thing to come of this is that a lot of Russian troll factories will sack employees whole-sale. I'm just not sure if that constitutes progress.

 

[fresh_42]

Forget physics. Ask it about kid entertainment:
https://www.physicsforums.com/threads/weird-news-compilation.878428/page-42#post-6872769

 

[pinball1970]

I was looking for some info on Black holes and found this. From 1.40 this YouTuber gives a view on Chat.

 

[morrobay]

I asked : What material parts of a tsunami wave have a velocity of 500 km/hr. It answered the crest. No the crest is not traveling with that velocity across the sea surface. This service has a long ways to go to equal google.