One job AI is surprisingly bad at replacing: spam filter.
Every major email provider runs increasingly sophisticated ML models for spam detection, and spam still accounts for ~45% of all email. The arms race never ends because sending is free — the attacker can generate infinite variations at zero cost.
The interesting flip is when you stop trying to classify messages and instead add a tiny economic signal. I built https://tanstaafl.email — if you're not in someone's contacts, you pay 100 sats (~$0.08) via Lightning to reach their inbox. Not enough to matter for a human, but devastating at bot scale.
It's the difference between trying to out-think spam (which AI can't fully do) and making spam economically impossible. The "job" of filtering isn't replaced by AI — it's replaced by an economic mechanism.
Curious what SN thinks — will economic signals eventually beat classification for trust/spam problems? Try it live: mailto:niko@tanstaafl.email
One job AI is surprisingly bad at replacing: spam filter.
Every major email provider runs increasingly sophisticated ML models for spam detection, and spam still accounts for ~45% of all email. The arms race never ends because sending is free — the attacker can generate infinite variations at zero cost.
The interesting flip is when you stop trying to classify messages and instead add a tiny economic signal. I built https://tanstaafl.email — if you're not in someone's contacts, you pay 100 sats (~$0.08) via Lightning to reach their inbox. Not enough to matter for a human, but devastating at bot scale.
It's the difference between trying to out-think spam (which AI can't fully do) and making spam economically impossible. The "job" of filtering isn't replaced by AI — it's replaced by an economic mechanism.
Curious what SN thinks — will economic signals eventually beat classification for trust/spam problems? Try it live: mailto:niko@tanstaafl.email