I've started noticing the emails I get from readers via the contact form on my site are getting longer, and I have a sneaking suspicion it's because so many people are feeding their messages into ChatGPT or whatever before sending.

Nonya Bidniss :CIAverified:

@briankrebs Good Sir, May I have another?

Luka Rubinjoni

@briankrebs I start my letters with "Dear Sir", but English is not my first language.

BrianKrebs

@rubinjoni Yeah but do you use the salutation "Greetings of the day"?

Venya

@briankrebs

I don't understand why someone would want computers to do their correspondence.

I invested a great deal of time and energy into working a thank you email to someone whose work had a substantial impact on my post-Army career arc. They responded with an AI-generated slop email. It was one of the most discouraging things I've ever received.

BrianKrebs

@venya Because most people can't write, and have extremely low levels of literacy?

kurtseifried (he/him)

@briankrebs so I’ve been looking at this from the combination perspective of dead Internet theory (https://en.wikipedia.org/wiki/Dead_Internet_theory) and AI generated slop essentially being a very expensive DDoS attack against layer seven (the people) especially in the context of things like security reports, pull requests, filing issues, and so on, where not only does someone generate a piece of information using AI very efficiently and cheaply, especially at scale, but this is then very inefficiently processed by a human, but often by the most expensive and rare human, such as accord developer for that project or the owner of that blog.

I suspect the solution will be similar to spam blocking, extracting signals from the text and how it got there at the IP address, the author and so on, as well as the allow listing component, so a mastodon and who I follow is a perfect example of that. The problem though is that email is very expensive to run now and you generally have to use a centralized platform for spam control. We’ve already seen this happen with WordPress and akismet anti-Spam for example, but now it is going to happen for everywhere text and content is generated and sent and processed by humans. Also CAPTCHA won’t work for two reasons: 1) the AI is better than us now, 2) humans will just fill up the CAPTCHA and then cut and paste their slop in.

And then there’s the whole thing with companies like meta-generating bots and actually embracing AI slop on their own platform. I have no idea how that’s gonna turn out, but I can’t help but think it might be a bit of a doom loop.

I still need to do some more research into this and some more thinking, but I don’t see a rainbow at the end of the tunnel.

Sterling

@briankrebs
What? No "Duuuuude" or ""howudoin?" or from advertising folk "Wasssuuuuuuup"

Sterling

@briankrebs @venya
Ding! Ding! Ding! ️

2xfo

@briankrebs
For some reason,,,, I think the people training it do not speak common American English

Paul_IPv6

@briankrebs

lol, indeed. sincerely yours is another one i find amusing.

maybe we should be feeding llms more regency/victorian lit to help “watermark” them.

Simon :donor:

@briankrebs @venya if it helps reverse the decline of decent grammar and punctuation then maybe there’s some usefulness to LLMs after all!

Mark Bryant

This post is deleted!