Chat GPT and AI stuff

[camion]

Her LinkedIn hasn't been updated since spring 2022 when she was at Duke...She was a Rhodes Scholar and Young Trustees Finalist after interning at Credit Suisse as a Summer Analyst. Clearly, a very smart and motivated individual. No idea what she's done post graduation. She issued a half apology, blaming others for giving her "ideas" that were taken from the Harvard speaker but that she didn't know the source of those ideas (Even though it was the language that was used that was basically identical in many cases...).

Convergent Oration?

The oratorical equivalent of convergent evolution.

[PackMan97]

Her LinkedIn hasn't been updated since spring 2022 when she was at Duke...She was a Rhodes Scholar and Young Trustees Finalist after interning at Credit Suisse as a Summer Analyst. Clearly, a very smart and motivated individual. No idea what she's done post graduation. She issued a half apology, blaming others for giving her "ideas" that were taken from the Harvard speaker but that she didn't know the source of those ideas (Even though it was the language that was used that was basically identical in many cases...).

Take any significantly long sentence and google it. You'd be surprised how it won't turn up any results (unless it's been published on the interwebs).

For example "Clearly, a very smart and motivated individual". There isn't anything all that special about that sentence. Google it (with the quotes) and no results are returned. If someone gives a speech and duplicates sentences from someone else, it's pretty much a guarantee to be plagiarism.

[Bluedog]

Take any significantly long sentence and google it. You'd be surprised how it won't turn up any results (unless it's been published on the interwebs).

For example "Clearly, a very smart and motivated individual". There isn't anything all that special about that sentence. Google it (with the quotes) and no results are returned. If someone gives a speech and duplicates sentences from someone else, it's pretty much a guarantee to be plagiarism.

Yay, I didn't plagiarize!

[Skydog]

Take any significantly long sentence and google it. You'd be surprised how it won't turn up any results (unless it's been published on the interwebs).

For example "Clearly, a very smart and motivated individual". There isn't anything all that special about that sentence. Google it (with the quotes) and no results are returned. If someone gives a speech and duplicates sentences from someone else, it's pretty much a guarantee to be plagiarism.

Dam, you are rite! I just googgled the last centinse I wrot and their wir no matshes.

[crimsondevil]

I was playing with this a week ago. It’s scary that it responds with absolute confidence even if it’s wrong. It’s a matter of gold/garbage in gold/garbage out. If you can manage (manipulate?) it in the right way it’s frightening good. Like replace some human workers good. Did these people never watch Terminator?

I played with it a bit a few weeks ago - it was pretty decent at coherently explaining random biological processes, although there was often a small piece that was incorrect. With some back and forth, it would eventually realize its mistake.

I then had a hilarious discussion with it about how realistic the asteroid in the movie Armageddon was. It gave the (apparently) in-movie specs for the asteroid and then got very confused about the amount of damage it would have done - it didn't seem to understand the importance of mass as opposed to volume, and it was totally incapable of determining the density of the asteroid and comparing it to a real-life asteroid until I led it through it. (FYI the asteroid in Armageddon is way under-massed (500 billion tons or something) compared to its stated size ("Texas"), giving a ludicrously low density.) It eventually "got" it, but it took a long time.

I think the "getting ideas" is totally valid from one perspective (it would call out themes that could talk about), but also certainly is "unfair" in the traditional sense in that that is usually the most difficult part of writing a cohesive piece. In addition, most students are not stupid enough to copy/paste the result EXACTLY as there are now detectors to give a percentage likelihood that it was generated with AI, but one could very easily simply edit the language/word choice to get around this and basically keep all the sentences/themes/structure there and avoid detection I expect...I don't see how teachers are going to get ahead of it. Yes, schools/teachers are aware of it.

As a college prof, I think it will be an increasing problem, although mainly for papers and other such writing assignments. I fed it a couple of "short answer" questions from a recent exam of mine, and it was hit or miss. It got one question correct, fairly impressively, then totally screwed up another that was related to the first.

I also asked it to write me a paragraph explaining the significance of one of my lab's research topics, and it was... okay, but not anything great. To me, it sounded like someone trying but who didn't really have any in-depth knowledge about the subject. It is impressive how it spat it out in <30 seconds though.

[Bluedog]

I played with it a bit a few weeks ago - it was pretty decent at coherently explaining random biological processes, although there was often a small piece that was incorrect. With some back and forth, it would eventually realize its mistake.

I then had a hilarious discussion with it about how realistic the asteroid in the movie Armageddon was. It gave the (apparently) in-movie specs for the asteroid and then got very confused about the amount of damage it would have done - it didn't seem to understand the importance of mass as opposed to volume, and it was totally incapable of determining the density of the asteroid and comparing it to a real-life asteroid until I led it through it. (FYI the asteroid in Armageddon is way under-massed (500 billion tons or something) compared to its stated size ("Texas"), giving a ludicrously low density.) It eventually "got" it, but it took a long time.




As a college prof, I think it will be an increasing problem, although mainly for papers and other such writing assignments. I fed it a couple of "short answer" questions from a recent exam of mine, and it was hit or miss. It got one question correct, fairly impressively, then totally screwed up another that was related to the first.

I also asked it to write me a paragraph explaining the significance of one of my lab's research topics, and it was... okay, but not anything great. To me, it sounded like someone trying but who didn't really have any in-depth knowledge about the subject. It is impressive how it spat it out in <30 seconds though.

Yes it's definitely hit or miss, but students are also smart enough to know when it's a miss in all likelihood and then refine the questions or simply then research those out more. If it gives solid answers sometimes, that still is a worthwhile time saver (some say cheating).

[duke79]

The irony of it is that in the time the students spend agonizing over how to make it so that it can avoid detection, they could have just done it themselves.

This will only apply in a very limited number of situations, but one solution is to have students do their writing in class. This doesn't really work for research papers and the bulk of the other assignments for which this tool most applies, but it does help a little bit. Of course, it is also not the best use of class time.

Speaking of which, whatever happened to the woman who gave the Duke commencement speech last year that seemed to be plagiarized?

I believe she has now been elected a member of Congress. (See, e.g., George Santos)

[Bluedog]

I believe she has now been elected a member of Congress. (See, e.g., George Santos)

Hahahah, you gotta be 25 though to even be a representative and I don't think she forged her birth certificate.

She has kept a low profile on social media since graduated but may be in i-banking...although not a very good year for that.

[nmduke2001]

I just access ChatGPT last night. It is incredible and very scary. I had it write an essay that my daughter is working on for her 7th grade class. It did it in 10 seconds and, quite frankly, it was better than my daughters essay. I asked it to find some investments with a particular trait. 5 seconds later, I had a list of 10. I am hosting a dinner for CPAs to discuss business M&A. I asked ChatGPT to write the invite. It did it in 3 seconds and it's very good. I could see how this could be a huge time saver but also cause a lot of temptation to just have Ai do everything for you.

[Acymetric]

I just access ChatGPT last night. It is incredible and very scary. I had it write an essay that my daughter is working on for her 7th grade class. It did it in 10 seconds and, quite frankly, it was better than my daughters essay. I asked it to find some investments with a particular trait. 5 seconds later, I had a list of 10. I am hosting a dinner for CPAs to discuss business M&A. I asked ChatGPT to write the invite. It did it in 3 seconds and it's very good. I could see how this could be a huge time saver but also cause a lot of temptation to just have Ai do everything for you.

Here's the problem: if it wasn't worth someone's time to write it, why is it worth someone else's to read it? It's the birthday cardification of day to day discourse, I'm not here for it.

[Bluedog]

I just access ChatGPT last night. It is incredible and very scary. I had it write an essay that my daughter is working on for her 7th grade class. It did it in 10 seconds and, quite frankly, it was better than my daughters essay. I asked it to find some investments with a particular trait. 5 seconds later, I had a list of 10. I am hosting a dinner for CPAs to discuss business M&A. I asked ChatGPT to write the invite. It did it in 3 seconds and it's very good. I could see how this could be a huge time saver but also cause a lot of temptation to just have Ai do everything for you.

It's definitely going to change things and how we communicate here soon -- particularly, when it gets incorporated into apps we use everyday already (gmail, outlook). It has great potential for those with perhaps communication issues and thus is somewhat of an "equalizer" in that regard, but could also stunt the growth in those areas if people are overly reliant on it. So, defintiely pros and cons. It will be a key skill going forward as to knowing HOW to best use AI to do things more efficiently. I read an article about a call center using AI and they found that the weakest employees had the most benefits from using AI (it was a metric of answering caller questions and resolving them) whereas the already knowledgable/experienced resources had much less of a gain (but still had a small gain). The AI was used basically to help guide them in what to look for/next steps, so the experienced people already knew that and it was basically telling them something they already knew in most cases whereas the new people were able to use that guide to get to the point much more quickly. It's definitely going to be interesting...

I also just accessed ChatGPT for the first time yesterday and played around with some prompts. I did find it gave a contradictory response depending how I worded the prompt, but if you're smart enough, you certainly know not to take everything as gospel and can edit accordingly. It certainly can be a time saver and a starting point for idea generation.

[bundabergdevil]

I use it to write all my DBR pun posts.

[MChambers]

I use it to write all my DBR pun posts.

Sometimes they come off as a little robotic.

[OldPhiKap]

Sometimes they come off as a little robotic.

That explains the tin ear.

[bundabergdevil]

That explains the tin ear.

I’ve often been accused of artificial intelligence so checks out.

[rthomas]

Here's the problem: if it wasn't worth someone's time to write it, why is it worth someone else's to read it? It's the birthday cardification of day to day discourse, I'm not here for it.

I used to used a card catalog to look up references. Now I use google scholar.

[Acymetric]

I used to used a card catalog to look up references. Now I use google scholar.

Information lookup and interpersonal communication* are fairly different beasts, aren't they? I'm not anti-technology or anti-AI (that would be somewhat...self defeating?). It would be somewhat surprising if etiquette around AI generated content didn't develop naturally as it becomes more common, though, and this will surely not be the last time you hear about the idea that people aren't going to be bothered to read something you couldn't be bothered to write.

*I guess we are going to have to define that concept really loosely now.

[BLPOG]

You're all behind the times! You should be using Cat-GPT

[tommy]

100 million users:

Took mobile 16 years to get there.
Took internet 7 years.
Took ChatGPT 3 months.

AI is also going to supercharge hacking. Projection is $24 trillion (with a T) cost of cybercrime by 2027. If cybercrime cost were a country's GDP, it would rank 3rd in the world in GDP right now.

[Bluedog]

100 million users:

Took mobile 16 years to get there.
Took internet 7 years.
Took ChatGPT 3 months.

AI is also going to supercharge hacking. Projection is $24 trillion (with a T) cost of cybercrime by 2027. If cybercrime cost were a country's GDP, it would rank 3rd in the world in GDP right now.

Impersonating people with AI and phishing is definitely going to be out of control...It used to be that you could clearly tell fake messages by the horrible misspellings, punctuation, and general grammar (among other things). That won't be the case in the relative near future.

The growth is quite amazing, but ChatGPT also has a low barrier to entry as it's free and simply signup via internet. Mobile and internet had costs/calls/time commitment/infrastructure involved. Still, pretty amazing.