The fundamental problem is that there is no difference between a spammer and a new user. LLMs will make them indistinguishable eventually.
What other solutions are there except to impose a significant onboarding or ongoing cost to new users? There could be many ways to impose this cost but it seems we have to do it.
Spam filters, pattern matching, manual blocking is all a waste of dev resources because attackers will just escalate over time.
quotingI suspect it wont be long until there is a fresh npub with each post using an LLM so that there is no text pattern. It will just look like a new user.
note1vgv…ggjv
Eventually each npub will have a full LLM created profile too, complete with an image icon. We won’t be able to mute or block, except to block new users.
The attacker might also fill up people’s DMs with useless messages to make DMs unusable too.
It does cost something for the attacker to maintain the attack, but they can reduce their costs by focusing it on large accounts and new users. If the attacker just focuses on new users, and makes the apps unusable for them, we might not even realize the severity of what’s going on.
Anyway, this is what I could do, and I’m assuming the attacker is more skilled than me.
We need to start putting in these defenses ahead of time, thinking like the attackers, so that we aren’t surprised by the ongoing escalations.