The Perils of ChatGPT

[bae]

You do realize that most ChatGPT models are trained up to a certain cutoff point in time?

It's not a universal tool...

[Tybee]

I just used an example in my class, about the Gouda cheese advertisement from the Superbowl, where it claimed that 40-60% of the cheese eaten in the world is Gouda.

We have a big emphasis on AI in this class, unfortunately, as that was how the university dealt with it, "responsible use." I am trying to point out there is no such thing.

[herbgeek]

Here's what I just got from the free version of ChatGPT :

When does your training data end
ChatGPT said:
My training data goes up until June 2024, but I can look up recent information if you need something more up-to-date!

[iris lilies]

You do realize that most ChatGPT models are trained up to a certain cutoff point in time?

It's not a universal tool...

No I did not know this data is not current. The answer should have been something like “I do not know who is the President of the United States since my training is only current through June 2024 and there was a Presidential election in November 2024.”

I only played around with it because someone on the MMM site with very sophisticated medical knowledge and health needs claims she is using
chatGPT to help her figure out her current health situation. She led me to believe it was more advanced than it appears to me to be.

[bae]

It is a poor craftsman who blames their tools.

[iris lilies]

It is a poor craftsman who blames their tools.

Haha, well, for sure, I won’t be using it for anything important.

[catherine]

I'm at a market research conference right now (sitting with a cup of coffee having reached my intensive people-interaction limit for the morning, with 6 more hours to go).

I came to learn how AI is being applied to market research tasks, and one of the things I learned in one of the sessions is that Chat GPT "lies"--you cannot trust the information. You have to fact-check it. So it was supremely stupid for the staff person of The View to fall back on the "journalistic integrity" of AI.

[iris lilies]

I just had a creepy conversation with ChatGPT. Not expecting much, I asked it to tell me about a particular aspect of contemporary art I am seeing. It gave me the answer that I pretty much suspected, but it also added interpretation of the artistic technique that I haven’t thought about. And then it asked me questions about the art I’m seeing. And I answered those questions. And then it commented on my connection with the Art and complimented me and asked more questions and I answered those questions.

damn that was creepy, it was like talking with a real expert.

[catherine]

I just had a creepy conversation with ChatGPT. Not expecting much, I asked it to tell me about a particular aspect of contemporary art I am seeing. It gave me the answer that I pretty much suspected, but it also added interpretation of the artistic technique that I haven’t thought about. And then it asked me questions about the art I’m seeing. And I answered those questions. And then it commented on my connection with the Art and complimented me and asked more questions and I answered those questions.

damn that was creepy, it was like talking with a real expert.

I know! DH uses CoPilot, and the conversations are way too real. Definitely the guy who wrote the screenplay for Her was on to something.

[LDAHL]

I recently came upon my wife screaming obscenities into her phone. She was unhinged by an unholy union of AI and telemarketing. Over a period of a week or two, she got over a hundred calls from multiple numbers selling a Medicare gap policy she wasn’t eligible for or interested in. The system kept taking different tacks trying to get her to buy, apologized and promised to take her of the list, but kept relentlessly calling back.