Hacker News Reader

Comment by walthamstow | Hacker News Reader

I've even grown to enjoy spelling and grammar mistakes - at least I know a human wrote it.

You can prompt/train the AI to add a couple of random minor errors. They're trained from human text after all, they can pretend to be as human as you like.

Barring simple typos, human mistakes are erroneous intention from a single source. You can't simply write human vagaries off as 'error' because they're glimpses into a picture of intention that is perhaps misguided.

I'm listening to a slightly wonky early James Brown instrumental right now, and there's certainly a lot more error than you'd get in sequenced computer music (or indeed generated music) but the force with which humans wrest the wonkiness toward an idea of groove is palpable. Same with Zeppelin's 'Communication Breakdown' (I'm doing a groove analysis project, ok?).

I can't program the AI to have intention, nor can you. If you do, hello Skynet, and it's time you started thinking about how to be nice to it, or else :)

Making it feel like there's no reliable way to discern what's truly human

There is. Be vehemently against AI, put 100% AI free in your work. The more consistent you are against AI, the more likely people will believe you. Write articles slamming AI. Personally, I am 100% against AI and I state that loud and clear on my blogs and YouTube channel. I HATE AI.

Hate to tell you but there is nothing stopping people using AI from doing the same thing.

AI cannot build up a sufficient level of trust, especially if you are known in person by others who will vouch for you. That web of trust is hard to break with AI. And I am one of those.

Are you including transformer based translation models like Google Translate or Deepl in your categorical AI rejectio?

Yeah.

The funny thing is that the things it refuses to say are "wrong-speech" type stuff, so the only things you can be more sure of nowadays are conspiracy theories and other nasty stuff. The nastier the more likely it's human written, which is a bit ironic.

No, you can finetune locally hosted LLMs to be nasty.

Maybe the future of creative writing is fine tuning your own unique form of nastiness

> The nastier the more likely it's human written, which is a bit ironic.

This is as everything else, machine produced has a flawlessness along some dimension that humans tend to lack.

There was a meme along the lines of people will start including slurs in their messages to prove it wasn’t AI generated.

A few months ago, I tried to get Gemini to help me write some criticism of something. I can't even remember what it was, but I wanted to clearly say something was wrong and bad.

Gemini just could not do it. It kept trying to avoid being explicitly negative. It wanted me to instead focus on the positive. I think it evidently just told me no, and that it would not do it.

Yeah all the current tools have this particular brand of corporate speech that’s pretty easy to pick up on. Overly verbose, overly polite, very vague, non assertive, and non opinionated.

Next big thing: AI that writes as British football hooligans talk about the referee after a match where their team lost?

I mean, it's not a meme..

I included a few more "private" words than I should and I even tried to narrate things to prove I wasn't an AI.

https://blog.dijit.sh/gcp-the-only-good-cloud/

Not sure what else I should do, but it's pretty clear that it's not AI written (mostly because it's incoherent) even without grammar mistakes.

I liked the "New to AWS / Experienced at AWS" cartoon.

People could prompt for authenticity, adding subtle mistakes, etc. I hope that AI as a whole will help people writing better, if reading back the text. It is a bit like "The Substance" movie: a "better" version of ourselves.

Whenever somebody misspells my name, I know it's legit haha

Way back when we had a landline and would get telemarketers, it was always a sign when the caller couldn’t pronounce our last name. It’s not even that uncommon a name, either

Guess what? Now the computers will learn to do that so they can more convincingly pass a turing test.

yesss. my thought too. All the variations of English should not lost.

I enjoyed all the belter dialogue in the expanse

> Write a response to this comment, make spelling and grammar mistakes.

yeah well sumtimes spellling and grammer erors just make thing hard two read. like i no wat u mean bout wanting two kno its a reel person, but i think cleear communication is still importint! ;)

Reply on news.ycombinator.com