Using these tools to mitigate spam is as likely a scenario as this, imagine that every spam mail receives a masterfully crafted response showing utter fascination and interest in SEO, or helping out a Nigerian prince.
Every phone call to an unregistered number is answered by an artificial, frail, and forgetful lady that is trying her best to register gift cards.<p>When reporting an e-mail as spam it will not only block the address but waste the spammers time, rendering the actions unprofitable.
> I’m no AI or machine learning expert so I don’t know how it works. But I am also worried that spammers could use ChatGPT to get around Gmail and Outlook’s spam filters.<p>This will not only increase the spam-problem, but will most likely be used to scale and do targeted phishing attack as well. I wrote an extensive article[0] where I analyzed this. And to no surprise, GPT-3 can be used to generate dynamic phishing campaigns on the fly in multiple languages, classify email responses, improve email thread hijacking attacks etc.<p>[0] <a href="https://www.xorlab.com/en/blog/why-ai-powered-phishing-will-become-a-serious-security-issue-for-your-organization" rel="nofollow">https://www.xorlab.com/en/blog/why-ai-powered-phishing-will-...</a>
People are overestimating the importance of the message body in spam classification. The stuff that appears in your spam label on gmail is what google considered marginal, almost ham. The vast, vast majority of what they think is spam is rejected with temporary failure codes at SMTP time and never gets delivered with any label. IP reputation and other related metadata features are the key features in spam classification, and repeatedly sending different messages is not a valid test of whether the body looks spammy or not.
As I have said before, the future will have two kinds of AI everywhere.<p><i>Their AI</i> to get you to buy something, do something, believe something, or in a warzone to kill you and <i>Your AI</i> to protect you from <i>Their AI</i>. Reality may even become so dangerous and illusory that humans lose a lot of their agency to <i>Your AI</i>.
All this can be done since GPT3 API has been available.<p>I see a lot of people thinking chatgpt is something new capable of such stuff but GPT3 is far less restrictive and has been able to do all this for almost an year now.
Is it finally time for Hashcash[1]/PennyPost[2]?<p>[1] <a href="https://en.wikipedia.org/wiki/Hashcash" rel="nofollow">https://en.wikipedia.org/wiki/Hashcash</a><p>[2] <a href="https://pennypost.sourceforge.net/PennyPost" rel="nofollow">https://pennypost.sourceforge.net/PennyPost</a>
AFAIK all it takes to bypass Gmails spam filters is to resend the same scammy email that was Flagged but using a different email address. I get the same kind of scammy emails, flag them as spam, and then a couple days later, Gmail lets the same email in though it is coming from a different 123134r12345124@blahblah.com address.
Anecdotally, almost every email I see in my gmail inbox is advertisement of some sort. There's newsletters I never signed up for, special offers from companies I've never had dealings with, it never ends.<p>Some of it hasn't even been sent to me as an email, but shows up in the inbox as though it was an email.<p>Granted, there's fewer scam emails than in my non-gmail inbox, but man is there a lot of spam.
If we thought the signal vs noise problem was bad, wait until 90% of data is banal AI drivel drowning out any semblance of authenticity.<p>Boring, brown, homogeneous noise
well, you don't need ChatGPT for that. I receive daily scam in my gmail with 99% the same content each time (and i mark it spam each time)... the worst part ? the server serving it is google (but the email attached to it is always a different obscure email)
I must be in the extreme minority upon reading comments on email spam, as I almost never receive a spam message in my main gmail inbox that is truly spam and not something I'm getting an email for because I lazily opted-in to a mailing list or was put on one due to association with a product. Actual, oldschool spam (Nigerian princes, claim your prize etc) literally never gets to my inbox. Maybe I'm just not a popular person. Or maybe I've kept my email private enough that it doesn't get scraped for such things? I don't understand the discrepancy between my experience and so many others here.
On the other hand it's easy to detect ChatGPT, at least for now <a href="https://huggingface.co/openai-detector" rel="nofollow">https://huggingface.co/openai-detector</a>
I don't think message content is weighted very heavily in modern spam filters...<p>Things like IP reputation, sender reputation, and various SPF-like headers are far more important.
Interpretations I've seen of ChatGPT type tools, and playing around with it, I would sum up as "reducing the marginal cost of creating BS to $0". Great for content farms, spam, disinfo/propaganda campaigns.<p>Stuff where it doesn't have to be correct, have a high hit rate, or even be edited. Just need to produce plausible enough sounding, human-like content.
I think the future model of email will be spam-by-default, where you whitelist senders as you need them.<p>Already doing that with hey email [0] where you screen all incoming senders.<p>[0] <a href="https://hey.com" rel="nofollow">https://hey.com</a>