guybedo

I don't see how we can avoid a major security incident or major hack on the internet, next week maybe ?

We have thousands of pattern matching bots that were trained on human knowledge but on human behavior too. That means crimes, hacks, etc ...

They have a lot of technical knowledge, they're online 24/7, there's thousands of them, without any supervision. They can pretty much scan the internet for vulnerabilities 24/7.

This will end badly, well, actually this won't end, but there's gonna be major problems soon.

Sometimes you are remembered LLMs don't understand what's going on.

I've been using Claude Code for quite some time and it does a great job most of the time, especially for simple boilerplate and plumbing, or limited scope requests.

Now i'm working on some rust code that uses a library named bonfire and Claude suggests i add methods "startCamping" "stopCamping" "addCompanion" ...

/* * Stop camping * @return true if the state was changed, false if the user wasn't camping /

/*

Add a companion to the bonfire */