You are viewing a single comment's thread from:

RE: Chasing shadows: Is AI text detection a critical need or a fool's errand?

in Popular STEMlast month

I like the topic and I agreed with you that we need a different approach to dealing with "AI".

The comparison with the pocket calculator is very apt. In the meantime, our sons' grammar school uses a computer algebra system in lessons. Why should pupils have to (repeatedly) draw graphs or repeatedly calculate derivatives or zeros when an app can do it much faster and more efficiently? Of course the methods need to be known, but do you have to repeat them x times? That bored me even as a pupil. In the time saved by using the app, you can perform completely different mental feats...

I see it similarly with the language models: The human provides the concepts and the "AI" formulates this, helps with structuring, etc.
That would of course be the ideal situation.
A kind of watermark would be very interesting. However, gorilla's argument that this can be circumvented with private language models cannot be dismissed out of hand.
The fact that this is possible in principle is shown by research into the protection of copyrighted material through invisible modification, so that "AI" cannot use this as training data (I can't find the link for this at the moment).

Have you entered your observation into the Globe At Night application or web site?

Unfortunately not yet. I haven't had any motivation in the last few days (I even switched on my PC today for the first time in several days) and now the sky is full of clouds. But as I have seen, there are still some campaigns to come :-)

Sort:  
 last month 

I see it similarly with the language models: The human provides the concepts and the "AI" formulates this, helps with structuring, etc.

That's how I am starting to see it, too. Another thing that occurred to me after posting is that AI writing styles are nearly certain to influence human writing styles. It's not a 1-way street. Even things that are written by humans may start to resemble AI texts.

Watermarking (i.e. white box techniques) is great for AI that wants to be recognized, but yeah with Open Source and private models, I can't imagine a way to make it universal. At some point, I think we'll just have to assume that everything we're reading was written or assisted by an LLM.

I haven't had any motivation in the last few days (I even switched on my PC today for the first time in several days) and now the sky is full of clouds.

I was counting on clear skies last night, but I didn't get them. Maybe tonight.

Coin Marketplace

STEEM 0.29
TRX 0.12
JST 0.032
BTC 63571.79
ETH 3068.45
USDT 1.00
SBD 3.97