your take on ChatGPT: thoughts?

[manoeuver]

I'm still sorting through mine.

Who else here has played with it so far? Any interesting results?

my first take is that the previous generation text generators were way more fun.

more later but I want to hear your takes.

[Chip]

Screenshot:

[Jon Szanto]

more later but I want to hear your takes.

Honestly, I was interested in taking it for a spin, but the sign-up wanted a lot of personal info and while it may be out there, I didn't want to feed an AI device with more of me. I'll come around to it at some point, if it doesn't get to me first.

Addendum: just finished this article with some unsettling thoughts on the AI revolution.

[dneal]

What Jon said, and that Vox article reminded me of two AI machines developing their own language - to the point that researchers pulled the plug because they didn't know what they were saying.

https://www.ign.com/articles/2017/07...s-own-language

[catbert]

[bunnspecial]

I've played some with it, and in fact was feeding it exam questions a few weeks ago and was surprised to see that it was able to get around 85% of the completely correct(even ones where I threw in a twist that threw off a lot of students).

When it failed, though, it failed big and for reasons not immediately obvious to someone who doesn't know the subject well.

As an example, I fed it a fairly standard question for me-asking it to calculate the pH of a buffer system and then explain based on the provided conditions whether or not this was the expected answer. This is a moderately difficult question for the level of classes I teach, but the the students who actually pay attention in class when we're talking about buffers conceptually and when we work through examples typically have no problem answering it. Calculating the pH is quite simple, as there's a well known equation that you plug everything into(and the only potential trip-up is not correctly identifying what is what). ChatGPT did that fine. Where it fell apart was the "why", where it gave me a wordy explanation of how buffers work but failed to actually step back, look at the the information provided, and say "There is more acid than conjugate base, so the pH should be lower than the pKa, which it is".

I've talked to other professors who are worried about it, but for a lot of purposes it has reached the level of "C level student work" with ease, but with better grammar and more informed arguments than typically seen from those students.

[bunnspecial]

Tried it again out of curiosity. This time the explanation was exactly correct, but the math was incorrect. It made a classic mistake I see students make, which is not converting the Ka to pKa...Screen Shot 2022-12-17 at 7.11.25 AM.png

But tried it again and it did it correctly

Screen Shot 2022-12-17 at 7.14.35 AM.png

[manoeuver]

I've played some with it, and in fact was feeding it exam questions a few weeks ago and was surprised to see that it was able to get around 85% of the completely correct(even ones where I threw in a twist that threw off a lot of students).

When it failed, though, it failed big and for reasons not immediately obvious to someone who doesn't know the subject well.

As an example, I fed it a fairly standard question for me-asking it to calculate the pH of a buffer system and then explain based on the provided conditions whether or not this was the expected answer. This is a moderately difficult question for the level of classes I teach, but the the students who actually pay attention in class when we're talking about buffers conceptually and when we work through examples typically have no problem answering it. Calculating the pH is quite simple, as there's a well known equation that you plug everything into(and the only potential trip-up is not correctly identifying what is what). ChatGPT did that fine. Where it fell apart was the "why", where it gave me a wordy explanation of how buffers work but failed to actually step back, look at the the information provided, and say "There is more acid than conjugate base, so the pH should be lower than the pKa, which it is".

I've talked to other professors who are worried about it, but for a lot of purposes it has reached the level of "C level student work" with ease, but with better grammar and more informed arguments than typically seen from those students.

Thanks for posting this. I'm of a similar mind.

No science fiction author (to my knowledge) has come up with a villain as diabolical as a system that can (and does) create an infinite stream of ~85% reliable information.

Now is the time to get to work improving your BS detectors, folks.

[Chip]

It could certainly replace most political speech writers. Also ad copywriters.

[MarciBriggs]

It could certainly replace most political speech writers. Also ad copywriters.

Ja, jag höll med dig. Förresten, vill du spela onlinespel med riktiga pengar? Om ja, då kan du besöka https://roligacasinon.se/casinos/casino-types/roulette här där du hittar de sajter genom vilka du kan tjäna mycket pengar genom att bara stanna hemma. Jag använder också den webbplatsen för att hitta spelsajter för riktiga pengar online.