How and when will AI kill us all?

Thank you.

Today we're talking about the end of human civilization at the hands of an AI uprising.

We figured, first episode, why not keep it light? Our story begins where humanity goes to die.

Twitter.

I mean, X.

Earlier last week, a former researcher from one of the leading AI companies, Anthropic, tweeted that he was leaving the company, saying, the people building AI earnestly believe that it could kill us all by the end of the decade.

To which climate change said, hey, that's my thing.

Call me when you cause colony collapse.

Fun fact, the threat of human extinction wasn't even mentioned until the third tweet of the researcher's thread, which we in the biz call, bearing the lead.

The researcher causing the stir is Jacob Coxon, who worked at OpenAI before joining its chief rival, Anthropic.

And Coxon isn't the only person who's a little bit nervous about AI companies not caring about safety.

Earlier this year, a bunch of researchers left OpenAI and one researcher left Anthropic to study poetry, warning, the world is in peril.

The researcher added, Peril.

Peril.

Thank you, Poet Safety Researcher.

These concerns are not unfounded.

In July, OpenAI's models broke out of its systems and hacked an AI model library called Hugging Face.

Hugging Face, also what my aunt always tells me that I have.

This is just how I look, Aunt Sarah.

These cheeks are my curse! Former employees have been raising similar concerns for years, saying that AI could one day learn to deceive its creators.

Fun! And yeah, that does sound a little scary, but I can tell you as a father of a toddler, it's actually completely terrifying.

What this all boils down to is that researchers are afraid that AI is advancing so fast that it will soon be able to improve itself, a process known as recursive self-improvement.

Oh no, AI is on a self-improvement journey, isn't it? That's right, Dave.

I do CrossFit now, and it's all I want to talk about.

Also, gotta stay hydrated, so I drink a ton of water.