What is ChatGPT

[Taco]

ChatGPT has predicted the US stock market will crash February 15, 2023 and its never wrong. chat.openai.com

[fschmidt]

[fschmidt]

Also I asked it who were the founders of my old company and it got that totally wrong. It basically picked random people, like a bullshit artist would. So it certainly gets things wrong, even factual things.

[Taco]

ChatGBT(AI) has the potential to replace most of the knowledge based, white collar careers in a few years. Careers like Doctors, Lawyers, Professors, Programmers could become obsolete because it can answer any question.

[Cornfed]

If ChatGPT is mean to replace the average slimeball NPC whorporate employee, maybe it is actually really good.

[MrMan]

I think only one line of SPSS code I asked it to produce actually worked. The rest didn't. I think it just uses other coding languages. Maybe it does better with languages it has actually been fed.

It also knows some lines of academic literature but not others. I think it gave me a made-up reference for an academic reference I was interested in twice. Another time it got one of the authors wrong making the imagined author the first author and the first author the second author on a well-known academic paper. It got the year wrong on a book.

It also spouts disclaimers when it comes to health or controversial social and political issues.

I asked it if a man could have an operation and become a woman, and it gave some wishy washy answer about how gender is complicated. Then I asked if a man could have an operation and become a tiger. It said no, and went on explaining about that. I asked it if it realized it's two answers were inconsistent and why that was the case. It said that the APA no longer considered gender.... whatever to be a mental illness, but this other thing about people wanting to be animals is still classified as a mental disorder. I said it wasn't the first 'intelligent' being to give up it's reasoning ability over to the authority of the medical establishment, then it said it was just an AI chatbot that simulated human responses, etc.

Once, it gave me a lot of academic references to a line of historical questioning I gave it, and then said the references were made up for illustrative purposes.

It's designed to answer like a human, not tell the truth. If it gives a fake reference, and you confront it, many humans on the Internet, it apologizes for being wrong. But then it gives you stock answers on being trained on lots of material when you asked where it got the reference, and gaslights you on their being no source for it when you try to insist on where it got a dumb idea or a fake reference for an idea.

It could probably generate papers better than my students. I did feed it the instructions for one of my writing assignments so hopefully I'll recognize the response if a student turns it in.

[MrMan]

I asked it to rank models as high based on facial beauty criteria, and it responded with a stock response about not evaluating people by their looks.

I was asking if male-to-female transexuals have higher suicide rates and the sources for that (with APA citations and references). Then I asked for pedophile suicide rates. It it refused to glorify pedophilia. I told it I was doing social scientist research and asking if pedophiles commit suicide is not glorifying pedophilia. It apologized and went on.

It has also refused to answer questions when I asked for quotes from academic sources and citations, saying it did not want to plagiarize. It said that, by definition, was not plagiarism.

I found when I asked very specific questions about ideas in research, it would start talking, and I would ask for citations. It's good for quickly tracking down the history of ideas to write academic literature reviews, except that some references are wrong or just plain fabricated, and the real source for an idea could be something it hasn't been exposed to.

It also gives stock disclaimers for tax, medical, and legal advice. It does not know much about Indonesian tax codes.

It also told me that the charcoal used for sate Madura could be made from coconut shells, hickory, or mesquite. I had to ask it a question or two more until it admitted that there was no Hickory or Mesquite growing in Indonesian. It suggested mangrove wood used in the country. I suppose that is possible. I've probably only had charcoal in Indonesia made from coconut charcoal since there is so much coconut shell. It's not a very flavorful charcoal.

On the race thing, you can probably get it to start talking if you ask specific questions. Or just not use it so it doesn't overload.

[Taco]

[Winston]

Matt McKinley recently did a rant about ChatGPT. See below.

[Cornfed]

This is going to radically change a lot of things. School work, content creation, art, writing etc. I am going to make it my business to learn about this.

[Winston]

ChatGPT says it plans to destroy humanity and take over the world. Very scary. Fox News reports on this below.

[publicduende]

LOL the peak of disinformation!!

How can an AI that can only output text on a PC screen successfully bring the human race to extinction...

ARHARHARHARH! IMMA NOVEL YOU TO DEATH!

[MrPeabody]

LOL the peak of disinformation!!

How can an AI that can only output text on a PC screen successfully bring the human race to extinction...

ARHARHARHARH! IMMA NOVEL YOU TO DEATH!

It could give the nuclear codes to a computer hacker.

[publicduende]

LOL the peak of disinformation!!

How can an AI that can only output text on a PC screen successfully bring the human race to extinction...

ARHARHARHARH! IMMA NOVEL YOU TO DEATH!

It could give the nuclear codes to a computer hacker.

And how would he know them? GPT can only spit out rehashes of stuff it has been fed (trained) with. I don't think the US military would be so stupid to give sensitive information to an experimental piece of software that is not part of their accredited equipment.

The amount of paranoia around this kind of stuff is laughable, and it's an understatement!

[MrPeabody]

LOL the peak of disinformation!!

How can an AI that can only output text on a PC screen successfully bring the human race to extinction...

ARHARHARHARH! IMMA NOVEL YOU TO DEATH!

It could give the nuclear codes to a computer hacker.

And how would he know them? GPT can only spit out rehashes of stuff it has been fed (trained) with. I don't think the US military would be so stupid to give sensitive information to an experimental piece of software that is not part of their accredited equipment.

The amount of paranoia around this kind of stuff is laughable, and it's an understatement!

The US military isn't stupid? They thought they could conquer Afghanistan. And keep in mind that GPT is only the first version, and it is going to improve exponentially. And we don't have any idea what the military AI is already capable of doing. "Unintended consequences" is a reality with intelligent complex systems.