Pick Your Poison

So which is worse?

1. That LLMs are a complete security nightmare? This is from Bruce Schneier, a leading security expert:

   We Are Still Unable to Secure LLMs from Malicious Inputs

   Schneier quotes from this Wired article, which explains how easily a poisoned document can contain invisible-to-humans text that an LLM will read, & follow its instructions to send security keys to the bad guys.

   A Single Poisoned Document Could Leak ‘Secret’ Data Via ChatGPT

   Schneier concludes:

   It’s an existential problem that, near as I can tell, most people developing these technologies are just pretending isn’t there.

2. That LLMs will literally help troubled people pick the best poison to kill themselves with. From 404 Media:

   ChatGPT Answered 'High Risk' Questions About Suicide, New Study Finds

   The parents of a child who suicided with help from ChatGPT are suing OpenAI. Best of luck to them.

I'm going to go with #2. Killing children is IMO the worst crime there is.

---

Here's the home page/directory for my posts on Bullshit.