Hacker News Reader

Comment by pwg | Hacker News Reader

> These emails are so bad and there is almost no chance of them finding their way through a spam filter, why are people still sending them?

Most likely because the cost to send per email is so low as to be essentially free.

As well, there may be some amount of 'Nigerian email' issue here. I've heard it said that the "Nigerian prince with 65M needs help moving money to your country" emails are so poorly worded on purpose to specifically filter out individuals who are not good marks for exploitation. I.e., if the recipient fails to notice the poor wording and grammar then they may also be easier to exploit. It may be the same with the spam. If an individual responds to the spam, then the spammer knows they have possibly found a very gullible individual that can be easily exploited.

View on HN

I'd pay for a service that responds to every spam email with a GPT-generated conversation, with the goal of wasting more of their resources than yours.

Maybe you could train it on a central database that keeps track of which conversations get the most consecutive replies from scammers.

You might not even need GPT. Maybe just the act of responding with a reasonable amount of text would create a large filtering burden for scammers.

I have to believe someone has tried this.

Sort of! I used to play around with Spamnesty [1] some 5 years ago. It looks like it's still around.

[1] https://spa.mnesty.com/

Does not even need to be that smart . "Lenny" and others are existing anti-telemarketer scripts and recordings. Reportedly 10-15 different answers from a "confused old man" are enough to waste their time.

So something similar with a bit of variation thrown in should turn that spammer advantage on it's head with the 65million spam emails getting 1million replies instead of 100.

I built it for a hackathon project.

https://devpost.com/software/test-7gzse1

It hooked into your email client, identified scams in a very mediocre way, and then sent back a autogenerated (not GPT-3 response).

The big problem is privacy. I am not sure how to effectively solve that one. Why I did not win, lol.

I see something similar as as a risk for the general internet. We are already seeing GPT-based content farms masquerading as product review sites. It's easy to monetize traffic via affiliate links.

It will get harder and harder to separate genuine (human-generated) content from AI-generated. Eventually I think we will have browser extensions designed to detect AI-generated content and it will result in another arms race.

I think this might be a good thing in the long run. It may force us back to smaller social communities, where we get recommendations by asking our friends and their friends, rather than trusting search engines to curate the internet for us.

There are risks. You don't want people to get into echo chambers. But that's happening anyway.

I don't think it'll waste much time, they'd just run their own conversation generating bot for the first few emails before passing to a human. It'll just be two NN models talking to each other.

> If an individual responds to the spam, then the spammer knows they have possibly found a very gullible individual that can be easily exploited.

This is absolutely key. The marginal cost of sending the spam is nil, but the marks self-select.

And you don’t need a lot if them to make a good income, especially if you’re in a developing country. In most of subsaharan africa, median income is under $1000. In some countries it’s under $500.

Due to conversion, minimum wage in Brazil is now $217,18

The cost for spammers can be actually free because they often use botnets and compromised accounts to send spam.

I reckon it’s just a numbers game.

Gmail will every now and then let absolutely ridiculous emails through the spam filter.

Badly spelled, bogus URLs, with $$$$ all over the place and All CAPS and just so clearly to any human a phishing attempt or just general scumminess and gmail takes a look at it and says “yup everything seems legit here. On to the inbox!”

My Gmail inbox from 2004 is 99% spam. I've tried everything I can imagine, nothing works. It is absolute chaos.

I've got the opposite problem - I get almost no spam to my inbox but regularly get legitimate emails routed to my spam folder, so I have to treat my spam folder like a second inbox anyway.

Google has to intentionally inbox a tiny fraction of spam, otherwise they'll never get corrective feedback on false positives, because basically nobody looks in their spam folder. Also, they have to permit a teensy fraction of stuff that would normally be temp-failed at SMTP time, for the same reason: otherwise it would be impossible for address and network reputation to improve. So sometimes if you see something completely ridiculous get inboxed, it might be that they are just trying to get data.

Ah, that explains some of what I see. I had noticed that my GMail spam folder went from thousands to dozens. I assumed they were zapping most of it but wasn't sure.

The overwhelming majority of suspected spam is rejected with temporary failure codes in SMTP and never makes it to the spam folder. The stuff in your spam folder is the cream of the crop!

There is a whole economy around spam at some scales: sometimes the spammer (the one actually sending the mail) has convinced a scammer (or slightly less illegitimate seller) to pay them to send mail because they claim to be good at getting through - in those cases the spammer actually gets paid a small amount (depending on their situation, a small amount to us could be a huge amount to them) no matter what.

botnets are not free. It takes time to develop and maintain. But it does make the cost of email using them very, very low.

sending a mailshot through a botnet has zero marginal cost. Maintaining the botnet has overhead, but that's not a cost of sending the email, it's an overhead.

I doubt this. Because even bot nodes will get blocked eventually so each email is degrading that value of that node. Again, very, very low, but not exactly zero.

I wished there would be some service like 'Lenny for phone spam' that will automatically keep on replying to spam emails and waste their time and resources.

This is an awesome idea. With something like GPT3 you could probably keep them going for a really long time. I have to imagine that if something like that became (somewhat) common, it would really cause some changes in how spammy sales works.

And then the spammers get AI to operate their side too...

Correct - because the cost of the spam itself is negligible, but the cost of developing a prospective mark into cash-in-pocket is quite high in terms of time & resources.

The last thing the scammer wants is for the mark to back out after the scammer has invested resources into them, and before they have realised any gains.

Effectively, insufficiently gullible marks reduce their profit margins.

I would like an email service where, when someone wants to send me an email, they have to first get some hash, e.g., "{hash}.myaccount@myprovider.com".

Then, in order to get the hash, they have to paypal me 20 cents.

If my attention is not worth 20 cents to them, then they should not be contacting me in the first place.

Further replies to the same thread should be free for some period of time (e.g. a month or so).

See discussion:

https://news.ycombinator.com/item?id=18100807

https://craphound.com/spamsolutions.txt

So according to that list, my idea would work.

According to the sibling comments, it does not only work, but some people are already using it in practice.

This was I think the basis for Dan Bernstein's "Internet Mail 2000"[0] idea - stop spam by making the sender responsible for mail storage.

[0]: https://cr.yp.to/im2000.html

Mail storage cost is negligible.

Per email?

Or per address/domain?

What's the best "altruistism wins" way?

Paid addresses, like ETH network?

Reply on news.ycombinator.com