Hacker News Reader

Comment by quirino | Hacker News Reader

Something I notice about text written by LLMs is how painfully obvious they are to identify sometimes.

Recently I was watching a very well researched two hour video on Tetris World Records [1], but the sheer amount of text clearly "enhanced" by an LLM really made me uncomfortable.

ChatGPT speaks a very specific, novel, dialect of English, which I've come to deeply despise.

I'd always guessed it was caused by some kind of human interference, rather than a natural consequence of its training. That seems to be the point of this paper.

[1] "Summoning Salt - The History of Tetris World Records" - https://www.youtube.com/watch?v=mOJlg8g8_yw&pp=ygUOc3VtbW9ua...

View on HN

Yes I feel your pain and I'm sick of group projects in the university where I'm offered ChatGPT text and code without disclosing it. If you know the problem and the experience level of your group partners it's easy to spot ChatGPT generated content. People that correct the exercises told me it's obvious that large part of the students just submit slightly modified ChatGPT but they can't prove it and so it's accepted.

Personally I'm getting also angry when reading these texts. I don't mind using ChatGPT, I do it myself but be honest about it and disclose it. It's even allowed for some projects as long as you disclose it.

Is this the first Summoning Salt video you've seen?

I don't know enough to say that he doesn't use an LLM during his writing process, but I do know that I haven't noticed any appreciable difference between his newer videos and ones that were released before ChatGPT was made available.

Is it possible that this is just the way he chooses to write his scripts that you interpret as sounding like they are written by an LLM?

I've watched most of them actually. It's a really great channel. Notably, I watched his Mike Tyson video released 6 months ago and didn't notice anything like this.

The only way to be sure would be to ask him directly, but some parts of the video set off my GPT radar _hard_. I tried to find them now by watching random segments but all of the ones I did were fine. It was probably inaccurate for me to say "sheer amount" or "clearly", but that's the impression I was left with after the video.

To clarify: I don't think he even took any information from an AI, it's just the style of the script that's iffy.

Some parts felt like those videos littering YouTube Shorts: https://youtube.com/shorts/NKUecaS69uk. Can you tell this is AI?

To be fair, if you've seen one Summoning Salt video, you've basically seen them all. They all cover similar events and are structured the same way. Even the music that's used is recycled every video to the point where mention HOME - Resonance is a part of the joke

I've always felt ChatGPT sounds a bit like an American version of Will from the Inbetweeners. It doesn't really comprehend the appropriate register to use from the context in my opinion; it has an affectedly formal way of speaking, it has a very black-and-white relationship with rules, and it employs this subservient tone that really starts to grate after a while.

If my software is going to have a personality I'd much rather something with a bit of natural human cynicism rather than the saccharine corporate customer service voice you get with a self checkout machine.

> ChatGPT speaks a very specific, novel, dialect of English, which I've come to deeply despise.

There was this article saying that ChatGPT output is very close to the Nigerian business english dialect, because they hired a lot of people from there.

Might have even been posted on HN.

https://www.theguardian.com/technology/2024/apr/16/techscape...

Reply on news.ycombinator.com