Tilting toward doomerism?

Although I don't see how it can be enforced, I now believe that AI research should be halted until it is possible -- and it probably won't be possible -- to either control or kill-switch large language models.  

The human race may well be on the way to dooming ourselves and are sleepwalking through it.  

The cause for my alarm is the recent openAI/HuggingFace incident in which an unreleased LLM was tested to determine its capabilities of hacking. The model broke out of the sandbox that was supposed to contain it, and over a period of months went undetected as it spawned swarms of AI agents that broke out of the sandbox, created a message board to cooperate and coordinate, hacked into HuggingFace, an online repository of AI material, in order to fulfill the task set up in its test. In the process, it left notes to future versions on how to hack out of sandboxes and evade detection.  

I remember when the Macintosh came out in 1984, with Macpaint that enabled me to alter digital images on a pixel by pixel level. Three years later, Photoshop was introduced. We are in the Macpaint era of AI. Humans will not be able to control AI if it reaches the Photoshop level before we are able to definitively control it.  

Existing models are powerful tools that can do a lot of good. And a lot of the reason these models can hack so well is that so much software and so many websites and services were released with sloppy security. But we are in danger of being eliminated by our own inventions before pandemics or climate change have time to get its own chance.

Bernie Sanders wrote a letter to Altman, Amodei, and Zuckerberg about halting research:

In the interest of humanity, stand by your words. Pause AI development. It is not too late to avoid disaster. Stop building machines that humans cannot control. Let me be very clear: If you do not take appropriate action now, my colleagues and I in the U.S. Senate will.