Hacker News Reader

Comment by nuker | Hacker News Reader

Soon email clients will be detecting GPT generated content and show warnings.

Will they? Gmail, which I am told is used by quite a lot of people, will already complete your sentences, encouraging you to copy the bad habits that it learned from reading the web. If they had a categorical objection to machine-composed messages they would not have implemented those features.

Gmail also offers to compose your entire response as long as you keep it brief (Smart Reply).

I guess I meant my client, Apple Mail app. Google won’t want to but will have to follow, to keep feature parity.

Detecting GPT generated content will take a fairly large language model to be reliable. Obviously this would need to be run in the cloud. Considering Apple's stance on privacy, sending private correspondence to the cloud is a huge no-no for them. Nobody is gonna implement GPT detection, atleast for emails.

Apple often boasts about on-device AI stuff. I bet they can make smaller, local model.

How will they do this? Is there any known way to detect AI writing (aside from human intuition), if you don't know the prompt?

https://huggingface.co/openai-detector

I typed a few paragraphs into this from OP's article and it judged them >60% real.

Edit: I typed a few sentences I made up myself and it judged them >90% fake. I don't think it works very well.

I tested three of my sent emails and all three were detected as 99% real. Searching for good examples of generated text now ...

Three full, generated, job related letters were detected as 99% fake. One was falsely detected as 99% real.

Sorry you had to find out this way...

Lol

"Describe in single words, only the good things that come in to your mind about your mother."

> I don't think it works very well.

That's the next level of hell: customer service rejects your request because it believes an AI wrote it (even though you wrote it yourself).

Coming soon to BigCorp.

Reply on news.ycombinator.com