What do you think are the most recognisable signs of AI? How can you identify it + scams?
Over the entire time of its existence, AI has improved a lot, which is problematic.
The amount of dishonest creation and scams is on the rise.
How do you protect yourself and distinguish AI from reality?
Here are a few examples I noticed:
In texts, the words "although, while, however, genuinely, honestly" really bother me - they smell like AI.
In videos, it's the imperfect letters. A family member of mine sent me this video thinking it was real. But when I looked at the font on the price tags, I knew it was AI.)
There was a case on Twitter where a guy was at a job interview with a recruiter, and the guy asked the recruiter to hold his fingers in front of his face – that job interview was an AI scam... it also affects this sphere; you need to ask for nonsensical moves that can expose AI.
Sometimes, I use Pangram to identify AI-generated texts.
Replies
I would like to share details about two recent incidents involving AI in Japan. I hope this proves useful as a reference.
A company caught an AI in a job interview. Three things gave it away: it mispronounced the company name in a way that looked like a text conversion error, there was a lag before every answer, and when challenged it said "ご自身がお話ししています" (go-jishin ga ohanashi shite imasu) — using an honorific that only applies to someone else to refer to itself. A Japanese speaker would never make that mistake.
The lag is the one I’d actually use. In a text chat you can force it: fire several questions in a row and watch the response times. A person speeds up on easy ones and stalls on hard ones. Generation time doesn’t vary that way.
The honorific mistake doesn’t exist in English. Your list — although, however, genuinely — is an English list.
And a security researcher said the biggest thing AI changed for attackers is that it crosses the language barrier. We’ve had over 105 million records leaked since September, and he said Japan is the only country spiking like this. I’m one of those records, which is how I ended up reading about this in the first place. Japanese was quietly acting as a filter, and that filter is gone.
So the language-specific tells are the ones disappearing first. They work because the model is bad at that language, not because it’s AI.
@ark_y_k maybe not trained on japanese data enough? :)
Hi, I'm the person who's used to use "although, while, however, genuinely, honestly" :D How do I suppose to survive in anti-AI checks now?
@fedulovelena try to find alternative words. E.g. I use instead "however" – "But!" :D
For me, three biggest giveaways of AI-generated texts are, 1. unnecessary defense. Basically overuse of "it's NOT because A, it's because B". But nobody thinks it's A in the first place. 2. Overuse of hyphens. 3. Using abstract language when it needs to be specific. For example, AI would say something like "Concerns were raised regarding the timeliness of the process" while human beings would just say: "They said they’d reply by July 13 but didn’t.”
@looqreadit_26 and don't forget em dashes.