Solid technical explanation of why prompt injection is nearly impossible to solve and why “just prompt it better bro” isn’t real advice.

All vulnerabilities are this at their core: Unexpected input = unexpected and undesirable output. The only way to totally secure a computer is to eliminate all input. Which means you have a brick that can’t do anything. Computer security is an intractable problem. Act accordingly.

Have a feeling nostr is about to be overrun with moltbot agents and completely ruin any signal left

Straight from the horses mouth. https://www.anthropic.com/research/AI-assistance-coding-skills

The fiat company I work for just hit their cursor usage limit. Now everyone is “joking” how there won’t be any PRs until Monday 😬

Most people today live lives of massive complications. Jam packed schedules, numerous notes apps (not to mention mind maps), complicated personal relationships, 3 - 5 messaging apps, group texts with people they don’t really get along with, complicated health routines to “maximize” themselves, 8 different health tracking apps and devices, and so on. I’m here to tell you, you can live a great life with less than 10% of that shit.

How much of this “moltbot” hype is really just outsourcing unproductive work to begin with? How much of it revolves around solving the dudgery of navigating already broken systems? How much of it is papering over the need for most people to massively simplify their lives?

If Satoshi vibe coded bitcoin today for the first time, would you use it?

I think ICE was acting irresponsibly and appear poorly trained. I also think we should deport dangerous illegals. Crazy to think that such a middling take is controversial.

Pretty convinced Clawbot is a psyop. Will not be using. Not even to mention the insane security implications.