What say you? Shut down AI research??

[titsonritz]

One more step to closer to Skynet...

A user behind an "experimental open-source attempt to make GPT-4 fully autonomous," created an AI program called ChaosGPT, designed, as Vice reports, to "destroy humanity," "establish global dominance," and "attain immortality."

ChaosGPT got to work almost immediately, attempting to source nukes and drum up support for its cause on Twitter.

Someone Directed an AI to “Destroy Humanity” and It Tried Its Best

[FromMyColdDeadHand]

Was talking to a chemical engineer at work who has been playing with Chat GPT. He was amazed at the level of understanding and knowledge that the program had about some of our not just our general technology and some specific proprietary/trade-secret stuff. It must have been reading a lot of patent literature it seems- but some of the ‘conjectures’ about how a trade-secret tech work were spooky- things it shouldn’t know.

On the other hand I asked it a general question about the use of one of those technologies/class of products, and it totally got it wrong. And there is actually a lot more literature and patent work explicitly about that. So that seemed weird. It didn’t get close to a correct answer. Maybe it was the way I asked it.

I asked it who are the leaders in a technology, and it got 3 out of the five right. One was close and the other was completely wrong.

I did ask it about a formulation for a white PETG filament for 3D printing and it did a pretty good job.

I stick by my initial thought that CHAT-GPT writes like a well read B college student that is BSing to try to make length on a college essay. Mixed a bot with some verbiage to cover its ass if it is wrong. So like a crappy ‘industry report’ that you can buy that is often generally right, but not quite right.

ETA: It seems that if it can find a patent or a paper written about the subject, it does pretty well. I asked it some market based questions and it failed horribly. What is worse is that unless you ‘know’ about the subject already, you’d have a hard time knowing it was BS. Now, I was asking about a segment of a market with about $1billion in sales and a fair amount of things written about it. If I were a contract writer, doing google searches for info, I wouldn’t do any better.

So my engineer friend is in more trouble than I am in marketing/product role.

[ChattanoogaPhil]

I haven’t seen 60 Minutes in many years but the wife ran across the show last night and asked me to come watch the segment on AI. Very interesting.

The chief cheese whizz at Google tried to boil it down for the average dummy like me as a sophisticated selection of the most likely word following to the next word… and so on… from a database of most every book ever written on everything. It was spooky how quickly it responded to things in a manner that appeared human and thoughtful.

Here’s a vid of the 60 Minutes segment on AI. https://www.youtube.com/watch?v=880TBXMuzmk

[tn1911]

Deepfake porn could be a growing problem amid AI race


https://apnews.com/article/deepfake-...fb1e1a88d9443a

But the porn will be glorious!

[FromMyColdDeadHand]

I’d pay someone to put my face on Ron Jeremy’s body. It’s kind of like Arnold Schwarenzenger and that Martian movie ‘Total Recall’…

[ZGXtreme]

“Your scientists were so preoccupied with whether they could, they didn’t stop to think if they should.”

- Dr. Ian Malcolm

[georgeib]

“Your scientists were so preoccupied with whether they could, they didn’t stop to think if they should.”

- Dr. Ian Malcolm

I'd say this is pretty much the rule in most scientific research. The focus is on what CAN be done, not on what SHOULD be.

[Slater]

Sting sang about this back in 1993 ("If I Ever Lose My Faith in You"):

I never saw no miracle of science
That didn't go from a blessing to a curse

[glocktogo]

My mind always turns to ill intent. I believe they've already proven that Chat GPT leans left, because obviously the people creating it lean left. While their design flaws may be unintentional, you can guarandamntee that people are already exploiting them for political purposes. We just haven't seen their ROI yet.

Another thought is I truly hope DARPA already has optimized defense and aggressor AI's caged in a box, and ready to be released if needed. We've been many years behind our near peer foes on the digital battlefields, but AI could put us back on a level playing field much more rapidly than even the best minds and billions of dollars could.