ChatGPT And Physics?

[Markus Hanke]

A lot of people are talking about the ChatGPT project of late, so I decided to try it out for myself, because I was curious. I’ve asked it a lot of technical questions with regards to physics and some other topics, and was pretty impressed how well it was able to answer most of them.

However, the AI wasn’t 100% spot-on. At least twice I caught it giving dubious answers - the first one was, with some goodwill, highly misleading; the other one was just flat out wrong. Interestingly, I was able to tell the AI about the wrong answers, and when I asked the same questions again (in a new chat session), it had one of them (not the other one!) corrected. 

So the moral here is that one should be very careful about using ChatGPT as a source of scientific information - at the very least, the answers it gives should be cross-checked against other sources to ascertain their correctness, and not just accepted at face value. This is an AI, not an online encyclopedia - the system makes mistakes, it learns, and it continuously updates its own internal states. It’s not infallible, so we are well advised to use it with caution.

PS. I fear that people referencing AI-generated output from sources such as ChatGPT to backup their arguments is going to become “a thing” before too long on forums such as ours. That’s going to be a problem, for several reasons; most importantly, as mentioned above, AIs make mistakes, and sometimes give wrong answers, so I don’t think we should consider thinks like “ChatGPT said…” to be a valid reference when it comes to backing up arguments. Secondly, I noticed that ChatGPT sometimes answers the same question in different ways, using different wording - so you can’t link to a specific textual answer in a way that others can check and recreate; it’s dynamic output, not static text. That’s problematic when it comes to using it as a reference.

What do other people here think? Do you guys feel there might be a need to develop an “AI-Generated Content Policy” for this forum?

[Genady]

1 hour ago, Markus Hanke said:

a need to develop an “AI-Generated Content Policy” for this forum?

I think the moderators already have one, e.g., 

 

[iNow]

1 hour ago, Markus Hanke said:

Do you guys feel there might be a need to develop an “AI-Generated Content Policy” for this forum?

There’d be no good detection mechanism, few good ways to enforce such a ban, and this becomes doubly true when the forum “member” doing the posting isn’t even human but is instead itself an AI bot. 

[swansont]

As Genady notes, we’ve already reached the conclusion that ChatGPT is not a credible source. It’s not peer-reviewed and no way to trace any expertise. It’s true that there’s no reliable way to directly detect it, but I think most of us are not shy about asking for citations or justification for any dubious claims. Plagiarism may go undetected, but BS will still be challenged. 

[CharonY]

There is a good discussion to be had how this will affect teaching and especially academia, but that may be off-topic.

[swansont]

1 hour ago, CharonY said:

There is a good discussion to be had how this will affect teaching and especially academia, but that may be off-topic.

Sounds like it could be an interesting new thread

[Eise]

Here is a youtube from '60 symbols' trying out ChatGPT:

 

ChatGPT was a great help for me in writing a bash-script: it was not correct the first time, but everytime I wrote what the error was, ChatGPT came back with a better script, until it was correct.

Today I tried it out with a SQL Server script, and I gave up after it was wrong again and again. It could still help in getting ideas though. Maybe there is more content in the internet about bash than SQL Server?

[Genady]

On 1/20/2023 at 4:22 PM, CharonY said:

There is a good discussion to be had how this will affect teaching and especially academia, but that may be off-topic.

Here is what Chomsky has to say about ChatGPT and education, especially academic: Chomsky on ChatGPT, Education, Russia and the unvaccinated - YouTube

 

[dimreepr]

On 1/20/2023 at 10:40 AM, Markus Hanke said:

What do other people here think? Do you guys feel there might be a need to develop an “AI-Generated Content Policy” for this forum?

Doesn't it boil down to the fundamentals of language?

I think, will always be different too AI think, however much data about I, it ingests.

[Endy0816]

I've had success with using it to write fictional accounts of real and fictional events, even changing them. It is interesting to say the least.

 

Quote

Please provide a fictional account of the assassination of JFK but Oswald misses every shot

 

On November 22, 1963, President John F. Kennedy was traveling through Dallas, Texas in a motorcade as part of a political trip. As the motorcade turned onto Elm Street, shots rang out from the Texas School Book Depository.

Lee Harvey Oswald, a disgruntled former Marine who had defected to the Soviet Union and later returned to the United States, was believed to have been the lone gunman. Oswald took up position on the sixth floor of the Book Depository and fired three shots at the President's limousine as it passed by.

However, in this fictional account, Oswald misses every shot.

The first shot misses completely, flying harmlessly into the air. The second shot also misses, hitting a tree branch and shattering it into pieces. The third shot hits the limousine, but only damages the trunk and does not harm any of the passengers.

After the shots are fired, pandemonium breaks out in the motorcade and on the streets of Dallas. Secret Service agents quickly surround the President and rush him to safety.

Oswald, realizing that he has failed in his attempt to assassinate the President, flees the scene and tries to blend in with the crowd. However, he is quickly caught by the Dallas police and taken into custody.

Despite the chaos and confusion of the day, President Kennedy emerges unscathed and goes on to serve out the remainder of his term. The failed assassination attempt becomes a footnote in history, remembered as a moment of great danger and uncertainty that ultimately ended in triumph.