Comparing the two versions of Roznet’s posts (one his and the other chatgpt’s) is fascinating.
The chatgpt one is a bit like one of the many “corporate BS” generators which have been on the web for years.
You can see the problem for a teacher/lecturer having to mark homework/Masters thesis generated in this way. Very very tricky because while you know it was machine-generated, how do you prove it? I know someone who was a Masters/PhD supervisor and I used to see some of the essays in for marking; you would see really crap English (typical for Masters courses; poor ELP since “money talks” in higher eddukkation) and then a piece of perfect prose which was obviously plagiarised from somewhere, but unless you can find the source you can’t prove it, and mostly the source was behind a paywall of some “essays” website, or was custom-written for that student, often by a retired univ lecturer!!!
If you are a known real contributor then this is ok but a new poster is likely to be terminated fast, and there is so much of it. As it is, no more than 50% of new signups qualify for forum access for other reasons (I am not going to list them here, but let’s just say they could almost be automated in software 
) and all of them are real people; the bots are blocked effectively by the google hidden recaptcha.
You can run raw models on your computer or hugging face
What is “hugging face”?
You can also adjust the style of chatGpt quite easily. Just asking after getting the first output I posted, “can you make it more casual” gives the below. Beside “reckon” I would probably not use, I think it’s start to be harder to distinguish from something I may have written?
Copyright: It’s a pretty interesting and complex topic. These models make it easier to create content that looks like someone else’s, but it’s still original in its own way—we’re all just building on what we’ve learned from others, right?
- I think these models are pretty creative. They come up with stuff inspired by human content, just like we do.
- Personally, I’m blown away by how simple online communication can have such a big impact. Still, even after chatting with ChatGPT, I’m not entirely sure what “attention” really means. But a couple of YouTube videos from smart folks got me closer to understanding it.
- Count me as an optimist. I reckon we’re just getting started. These models can reason pretty well, and when we team them up with specialized AI tools, they could take us pretty far—the whole “agent” approach looks promising to me.
Peter wrote:
What is “hugging face”?
It is a website providing and frameworks to run open source models locally. bunch of wrappers on things you’d want to do with a model, like fine-tuning, rag, etc, etc
Sorry – don’t ban me Peter :)
Can help playing… here is a version as a “British Luddite”, I wonder what you would have thought of such a post out of nowhere, and would have suspected ChatGPT or just someone trying to be funny?
A few additional reflections:
- Copyright: A most vexing and complex matter indeed. Beyond mere inspiration, a staple of our human endeavour where we but refine upon past knowledge, these contrivances facilitate the creation of works eerily reminiscent of established artisans, yet claim originality.
- I must concede, these mechanisms display a certain inventiveness. They conjure works anew, taking cues from the annals of human creativity, a practice not unfamiliar to our own species.
- I find myself utterly bewildered by the marvels wrought by simple matrices operating at grand scales. Despite numerous inquiries directed at this ChatGPT, the concept of “attention” eludes me still, whereas a mere one or two visual lectures by astute scholars upon the YouTube have illuminated the matter more effectively.
- I align myself with the optimists, albeit cautiously. This may well be merely the commencement of a broader saga. The reasoning capabilities these models exhibit, paired with more focused “narrow” AI tools, suggest a promising avenue—the “agent” approach, though I tread with reservation.
Very few people use words like vexing, staple, contrivances, annals, scholars, and the use of “artisans” is definitely totally sickly-sweet BS which no normal person (i.e. outside of Marketing ) would use with a straight face; example:
Must be a French site, we rely on artisans.
It was about the only thing I understood in this thread.🙂
Peter wrote:
You can see the problem for a teacher/lecturer having to mark homework/Masters thesis generated in this way. Very very tricky because while you know it was machine-generated, how do you prove it?
Yes, that’s becoming a real problem at universities.
roznet wrote:
What do you think of using ChatGPT for editing a post? For those of us not great at English? Not that I really have, but just curious, seems a valid use to me.
It’s very useful. You can ask a LLM for a translation or rephrasing. I can’t speak for forum rules, but I would definitely expect you to make adjustments after the fact and check that it’s phrased correctly to mean what you intend (which means it’s not a pure copy / paste containing nonsense). If it’s a pure (automatic) translation, you can also mention it at the beginning.
Your following generated posts still sound like corporate BS to me (mainly because there are no language shortcuts)
I think the best way to use LLMs for phrasing is to ask for suggestions and then reassemble the sentence yourself.
Airborne_Again wrote:
Yes, that’s becoming a real problem at universities.
Presumably the answer is that after submission the student would be “interviewed” on the content. So they are asked to explain certain points or delve further in and clarify things. This would be an integral part of the results.
This would quickly distil those who understood what they had written vs those who only copy and pasted it. And I suppose if you completely understand what you’ve written it doesn’t really matter if you’ve copied and pasted it or not!
