@yogthos that plausible deniability will come in handy after the nukes start flyin.

Ignoring the Skynet concerns; either he's feeding data that should not be available to the public into a privately owned system or he is trying to form actionable models off of incomplete data. Neither option is comforting to think about.

This gives me hope 🥰

Unfortunately, that scenario is no longer the stuff of cold war fiction. As first reported by Business Insider, Major General William “Hank” Taylor, commander of the 8th Field Army in South Korea, told reporters that “Chat[GPT] and I” have become “really close lately.”

“I’m asking to build, trying to build models to help all of us,” he said, adding that he’s using ChatGPT to help make military and personal decisions affecting the soldiers under his command. This includes the joint United Nations Command in South Korea, which Taylor currently leads as chief of staff.

Evolution has always heavily favored taking the easiest route possible, and humans especially excel at offloading anything we possibly can. Laziness is ironically a huge advantage for most of human existence.

So it makes sense when people immediately dump their own critical thinking to AI. And if it's just one person fucking their own life up it's not a huge deal.

But a senior military leader offloading executive decisions to an AI is fucking insane.

Not only for all the normal reasons, but the fact that it could be a personalized attack vector, and at this level with him publicly talking about it...

Someone is going to try.

But he's not just fucking around with his own life and safety, he's fucking with thousands of people with little to no oversight.

I just hope to God he's using the DoD version at least and not just a random build.

Article title:

Top US Army General Says He's Letting ChatGPT Make Military Decisions

Body text

“I’m asking to build, trying to build models to help all of us,” he said, adding that he’s using ChatGPT to help make military and personal decisions affecting the soldiers under his command.

These are not equivalent statements. You can take input from a subordinate, or from a military history work, doctrine manuals, or Google, or ChatGPT in making a decision without having them make that decision.

I need to get started on my fallout shelter

Do that. Let the slopbox make all the decisions.

Fantastic.

RIP

taps the sign

I'm less concerned about it making poor decisions and more concerned about plausible deniability. It's going to make poor decisions, that's just a feature of AI, but if either one makes a bad call and people die he can just say it wasn't him. Basically gives them a golden ticket to do whatever they want and blame AI to avoid accountability. It's literally one of the excuses that Israel has uses when murdering civilians in Gaza.

I'm not saying this particular person is the kind that would do something like this, but there are plenty that will and already do.

Decisions like AI suggesting to depressed people to kill themselves, told dry alcoholics to have a drink as a treat, and recommended taking heroin for others?

Many folks are too dumb to understand that AI is not actually intelligent.

So this absolute dumb fuck, that has utterly fallen for the sycophancy, is very likely putting military secrets into his prompts. Military secrets that can then be read by employees at OpenAI, or hackers that have gotten into their servers, and sold to whatever nation state pays up. What an absolute shit-for-brains.

ETA: Even if not literal secrets. I guarantee that the the enemies are very willing to pay just to find out about what (even mundane) problems he's asking AI for help with. Is a top US General asking AI how to unclog a toilet? North Korea wants to know.

I’m thinking of telling management that AI came up with my plans so I can just do whatever I want and have them love it.

AI Jesus take the wheel!