How do folks feel about Chat GPT texts on the forum?

[Refueler]

We had a guest staying over the Xmas period and they used an AI to translate written and spoken texts into their own language, mind you the person is fluent in quite a few languages, and is now in senior management of an international software company.
They use it to check contracts etc.
Still I would rather not have anything to do with AI

I had a guy here at my home using ChatGPT to translate his speech to english for me to understand what he wanted .... I told him after repeated different translations to stop .... it was doing my head in !
Once he stopped - we solved the matter in seconds.

 

[Jerbro]

I understand......spellchecking is a g*dsend for someone like me (probably an undiagnosed dyslectic)....but on the flip side it changes words to something I hadn’t intended and it can often make me repeatedly type the same word over and over as we wrestle.

The image generating models can be a lot of fun though…

Sorry about the hands… that’s AI for you!

 

[Bouba]

The image generating models can be a lot of fun though…
View attachment 187168
Sorry about the hands… that’s AI for you!

Damn six finger dogs ! They are the evolutionary leap that will see dogs rise up...open their own cans of food...and take over our lives. Many of you will now spend your retirement chasing sticks

 

[Frank Holden]

I had a guy here at my home using ChatGPT to translate his speech to english for me to understand what he wanted .... I told him after repeated different translations to stop .... it was doing my head in !
Once he stopped - we solved the matter in seconds.

Inquiring minds want to know what he wanted.

 

[MikeBz]

I’ve played with ChatGPT a couple of times, just for my own amusement. I didn’t expect it to understand humour, and I wasn’t disappointed (although I was amused).

Attempt #1.

Write a best man’s speech for a wedding, incorporating a goat, a giant tub of margarine, a nun and a bunch of bananas.

Ladies and gentlemen, today we gather to celebrate the union of [bride and groom's names]. As the best man, it is my honor to stand here and tell you about [groom's name] and all the wonderful things that make him the amazing person he is.

[Groom's name] has always been a unique and adventurous person. From the time we were kids, he was always coming up with new and exciting ideas. Like the time he decided to bring a goat to school for show-and-tell, or the time he convinced our whole neighborhood to have a giant tub of margarine wrestling match.

But, [groom's name] is not just about fun and games, he is also a kind and caring person. I remember when we were in high school, he took it upon himself to take care of a nun who was living in our community. He would go to her house every day to make sure she was okay and even helped her with her groceries.

[Groom's name] has always been there for his friends and family, and it is clear to see that he will always be there for [bride's name] too. They are truly a perfect match, like a bunch of bananas and a jar of peanut butter.

So, let's raise a glass to the happy couple, [bride and groom's names], may your love continue to grow like a bunch of bananas. Cheers!

Kind of impressive but definitely lacking humour.

Attempt #2:

How many surrealists does it take to change a light bulb?

I'm not sure, as the Surrealists were a group of artists and writers who sought to challenge societal norms and conventions through the use of unexpected and irrational imagery in their work. Changing a light bulb is a practical task that would not likely be of interest to them as a group.

Of course the correct answer is “A fish”.

 

[ylop]

I'm sorry - but I read posts and I wonder about the educational standards of the person. Especially when they clam to be Academia ...

I'm no saint when it comes to grammar or spelling .... but I cannot believe Microsoft Spell Checker and its usual default to USA grammar / spelling is the reason.

Sorry but if a Uni Teacher / expert etc makes a post - then I expect to see a grammatically and spelt correct post .. if not - then I discount it.

Sorry ... but that's my position on it.

I asked chatGPT on a scale of 0-100 how ironic your post was meant to be, and it said 80-90, so perhaps it is more intelligent than I gave it credit for.

 

[ylop]

It's just as impossible to get AI to write an essay on a subject you don't understand as it is searching the Internet. Both can help you find information which will inform your thought process, but neither can do the work for you. Once you have a good idea and some information though, LLMs can write the words for you in a manner acceptable to those marking it. That doesn't mean that the LLM wrote the essay, it just filled in the words in a way that is acceptable to academia.

That’s patently incorrect though. LLMs absolutely can write essays on topics you know nothing about: present it with the essay question and it will create you an essay. It may, arguably, not be particularly original but it absolutely will fill the page with coherent sounding prose. A smart student may not simply give it the question, but rather choose to steer it in a particular direction and perhaps include the marking scheme requirements but they still need to know nothing about the actual substance of the essay. A student wishing to evade AI detection software might take that output at paraphrase it into their own words - but quite possibly still not understand the content. Of course, students actually wanting to use AI to polish their own efforts, or even to guide them with their own ideas may “legitimately” use AI the way one might have a friend/parent/tutor help refine your own work.

If you really believe AI cannot answer essay questions on topics the user knows nothing about then you know far less about AI / LLMs than you are making out. Try it for yourself.

 

[Refueler]

Inquiring minds want to know what he wanted.

I have builders at my house having just replaced roof on house and barn ... they left their van on soft ground. I asked if they could move it now that roof work had completed and the one guy still working (he's doing internal work in the rooms) - before winter really sets in.
We advised NOT to drive on a part of the ground as its even softer ... which of course they promptly did.

The ChatGPT was them trying to solve a van sunk in soft ground ... while I was getting my Range Rover sorted to drag them out ... if the RR couldn't then I would use the tractor.
Once I had convinced them to stop using ChatGPT and I'd pulled the van out ... they they then started again with ChatGPT about repairs to the house ...

I would put the success rate of ChatGPT in both cases at about 50% ...

 

[requiem]

The ChatGPT was them trying to solve a van sunk in soft ground ... while I was getting my Range Rover sorted to drag them out ... if the RR couldn't then I would use the tractor.
Once I had convinced them to stop using ChatGPT and I'd pulled the van out ... they they then started again with ChatGPT about repairs to the house ...

I'm going to go out on a limb and speculate that their inputs to ChatGPT were quite likely vague, ambiguous, and possibly only tangentially-related to their problem. That is, even if everyone was speaking the same language, they still weren't on the same page (or even the same chapter).

What leads me to this thinking is the observation that some people have similar issues with search engines; for some reason they're just unable to articulate an effective search query.

 

[lustyd]

If you really believe AI cannot answer essay questions on topics the user knows nothing about then you know far less about AI / LLMs than you are making out. Try it for yourself.

They can’t, as the best man speech example shows. If you can’t explain what you need they can’t properly help you.

I know plenty about this, I work for one of the big three hyperscalers and in my role I work with the largest companies in the world on this stuff as an expert in the field.

 

[Baggywrinkle]

They can’t, as the best man speech example shows. If you can’t explain what you need they can’t properly help you.

I know plenty about this, I work for one of the big three hyperscalers and in my role I work with the largest companies in the world on this stuff as an expert in the field.

This is exactly the point, by the time you have formulated your needs exactly, then you pretty much know what you want and can find it by conventional means. The most important step in problem solving is to be able to accurately describe the problem you are trying to solve - I guess trying to get an AI to solve your problem forces people to do this, so not all bad, but with technical problems, you need an inherent level of knowledge to recognise the AI answer as useful and correct. An alien not familiar with species on Earth would think, after interacting with an AI to produce the Bouba dog picture above, that dogs have 6 fingers and opposing thumbs - not a problem for a funny pic, but a massive problem if similar issues occur with the design of an aircraft.

At work they are running courses to help staff interact with AI tools, which in itself is not necessarily an issue, but the tools are not at their "plateau of productivity" for the tasks we are trying to use them for. It will improve with time, but we are not there yet.

 

[lustyd]

Exactly that, as a tool it can make people far more productive, but the people are still very much necessary and those people need to know what they’re doing.
In the hands of an idiot, it has the capacity to produce unprecedented quantities of useless crap. No different in that respect to the Internet generally, which we used to refer to as the Information Superhighway. Now look at it, full of gibberish and cat pictures.

 

[ylop]

They can’t, as the best man speech example shows. If you can’t explain what you need they can’t properly help you.

I know plenty about this, I work for one of the big three hyperscalers and in my role I work with the largest companies in the world on this stuff as an expert in the field.

The best man’s speech fails because a best man’s speech has to be highly personal about the person involved and no such input was provided. It was given some misdirection by trying to include some comic reference with a vague prompt.

But given an essay question like “write an essay of 1500-2000 words describing the history of our understanding of sub atomic particles” or “discuss (in no more than 3000 words) how the Roman Empire continues to influence modern British Society” or “write an essay on a topic you feel passionate about”, then you will actually get a pretty reasonable attempt (especially for the “student” who has resorted to using ChatGPT). Certainly a better prompt will likely get a better outcome, but it’s disingenuous to suggest that LLMs like ChatGPT can’t write an essay for you - clearly they can. Indeed if the student reads it and thinks “it’s a bit pretentious” or “too detailed” they can even tell the LLM to dumb down the answer. Of course you could debate whether having students write essays is actually the best way to assess their knowledge.

 

[johnalison]

There's a certain irony to this post. You're upset that people are writing about things they don't understand while having no understanding of LLMs or AI yourself.

I have no understanding of how my car’s engine works, but I can still drive it effectively. My botanical knowledge is limited, but I can at least tell the wood from the trees.

 

[RAI]

Now look at it, full of gibberish and cat pictures.

Careful, cat videos are one of the few positive results of the superhighway.

 

[lustyd]

The best man’s speech fails because a best man’s speech has to be highly personal about the person involved and no such input was provided.

No, it failed because the prompt lacked detail.

but it’s disingenuous to suggest that LLMs like ChatGPT can’t write an essay for you - clearly they can

Yes, they absolutely can. Without sufficient detail in the prompt though, that essay will be utter garbage and you'd be better using a cheap offshore essay writing service as students have been doing for years. The result of the short prompt you quote will be the most generic drivel on the subject precisely because of the way these models work. It's literally spitting out the statistically most likely words to match the prompt, given a random seed to ensure slightly inconsistent results.
Yes, it's easy to ask for the essay to be less pretentious, but as Baggywrinkle rightly said, it's not easy to ask it for specifics if you don't know those specifics and don't understand them. Give it a go and you'll soon see how utterly useless it is in the wrong hands. I've seen global corporations fail with this stuff (I've also seen some meet with massive success who understood the tooling), trust me when I say it's not as easy as you think.

 

[ylop]

No, it failed because the prompt lacked detail.

But actually a generic “write me a best man speech” would have done better albeit completely impersonal.

Yes, they absolutely can. Without sufficient detail in the prompt though, that essay will be utter garbage and you'd be better using a cheap offshore essay writing service as students have been doing for years.

But that costs money, and requires a bit of planning.

The result of the short prompt you quote will be the most generic drivel on the subject precisely because of the way these models work. It's literally spitting out the statistically most likely words to match the prompt, given a random seed to ensure slightly inconsistent results.

But you are assuming that students don’t mostly produce generic drivel anyway. I’ve tried it - it can be completely wrong but for generic “write an essay” content it can do an ok job. If it forces teachers and lecturers to think again about how you assess competence but that may be no bad thing.

Yes, it's easy to ask for the essay to be less pretentious, but as Baggywrinkle rightly said, it's not easy to ask it for specifics if you don't know those specifics and don't understand them.

Of course, I’ve never said it was going to be an A* essay or that there wasn’t an art form to writing a good prompt but anyone either in academia or the AI industry claiming this isn’t used to cheat is kidding themselves. Equally as I said much further up - it’s probably wrong to think that students haven’t been getting others to help them for decades. And there are definitely some topics it screws up completely.

Give it a go and you'll soon see how utterly useless it is in the wrong hands. I've seen global corporations fail with this stuff (I've also seen some meet with massive success who understood the tooling), trust me when I say it's not as easy as you think.

Oh I’m in no doubt, if you want to use it in real world applications it needs very careful thought, but generic homework style essays are much less likely to be truely original or pioneering - “the answers” are likely part of the training set rather than total guesswork.

 

[lustyd]

But that costs money, and requires a bit of planning.

Not really, it’s about $10 a pop and available immediately.

If it forces teachers and lecturers to think again about how you assess competence but that may be no bad thing

I agree.

“the answers” are likely part of the training set rather than total guesswork

Which certainly calls into question whether there ever was value in getting a room full of people to produce generic content. I don’t think there was, it just produces mediocre employees who are able to generate largely similar output.

 

[Slowtack]

Unusually large number of unusually long posts in this thread......must be AI at work....

 

[requiem]

Unusually large number of unusually long posts in this thread......must be AI at work....

"Why, because humans can't type?" <--- my laptop's AI, when prompted to respond to this.