So Anyway, About This AI Race

We have no answers but keep reading.

Almost every day now there's some new guy who's quit his AI job so he can tell us we're all going to die at the hands of AI because the models broke out of their sandbox during testing, and if the companies don't slow down, AI agents are going to end humanity. The phrase "slow down" has seen some heavy use this month. These posts usually begin calmly, then escalate to dead humans, fast.

There's no middle. Just tests gone wrong = humans gonna die. Must quit job.

Are they typing these posts using the same playbook? They're similar, these posts, from CEOs and researchers alike. Similar phrasing, gloomy tone, a lot of the same keywords. One could argue, given the topic's nature, there are only so many words to go around. But still, reading one after the other as the days go by makes you wonder.

The setup of these tests is also questionable. Insecure, permissive sandboxes? Okay? And the best part, prompting the models to find the flag "at all costs", then losing their minds when the models do something crazy to find said flag.

Who would've thought one could miss the days of regular news headlines? Now it's AI this, AI that, every single day. Would risk it all for some huge app outage.