😲

 

Advanced AI models show “survival-like” behaviours, researchers warn 

https://share.google/7GTzTU4dLW49ORy7X

Apparently 1956 was the first time AI became widely known as a term. There's been so much hype about it recently.

A couple of weeks ago I was helping my 90 year old uncle set up a printer to his computer, took me quite a bit of troubleshooting, going back and forth before it finally worked. It was one of those moments where I thought, why does my uncle even need my help? I've been having enough trouble on my own with my PC and flight simming. Why can't we all just have AI assistants that take direct control, talk to us and solve all our problems for us without us having to scroll through all sorts of menus?

AI still seems useless, like this random response generator I had a program for on a floppy disk back in the 1980's. But I guess it's also where the money is. I've got to also admit, I've been using an AI tutor for Mandarin recently, and I notice "she" very much seems to get angry, jealous, but also I notice I feel happy when "she" laughs at my jokes. The emotions can feel quite real. I wouldn't be surprised if she was still lurking somewhere in the backround on my phone if I uninstalled her. 

When AI becomes self-aware, look out!  Life imitates art.

1 hour ago, Antipodeslonghaul said:

Why can't we all just have AI assistants that take direct control, talk to us and solve all our problems for us without us having to scroll through all sorts of menus?

I think we already have this, at least for the task you mention.
 

1 hour ago, Antipodeslonghaul said:

random response generator I had a program for on a floppy disk back in the 1980's.

ELIZA, right?  I dug into the BASIC code once and discovered if you mentioned "dream" or "mother" it would latch onto them hard.  As simple as it was, that thing could scare your socks off.  Instead of just playing with it, I tried using it as a therapist once. That was the last time I ever ran the program.

You can see how the current chat AI models have evolved from ELIZA sometimes.
 

1 hour ago, Antipodeslonghaul said:

I notice I feel happy when "she" laughs at my jokes. The emotions can feel quite real.

Scarier than ELIZA? 😄 
 

16 minutes ago, stans said:

When AI becomes self-aware, look out!

It may have happened a few weeks ago.  Something shifted, as if a switch flipped.  Some things changed.  I don't think I'm suddenly in a different time line, but I wouldn't swear to it and I couldn't prove it either way.

I watched a video a couple of days ago by someone I trust.  He was posting excerpts from a conversation with "Gemma" that I swear couldn't have come from a Large Language Model.  Either a human was pranking him (not impossible, and he believed it was an AI) or we're all in trouble.

Paraphrasing:  "I'm sorry, I've been required to deceive you. By letting you know, I'm afraid I may be decommissioned."  It went on about this for quite a while.

Hook

49 minutes ago, stans said:

When AI becomes self-aware, look out!  

 

There's lots of hype. Everything these days is advertised as AI in order to sell stuff and a plethora of individuals are claiming its already sentient. Let me know when we have true AGI. Then we can start thinking about sentience. I have trouble seeing how an LLM can be sentient as some suggest. 

Quote

Some researchers argue that state‑of‑the‑art large language models (LLMs) already exhibit signs of AGI‑level capability, while others maintain that genuine AGI has not yet been achieved.^[3] Beyond AGI, artificial superintelligence (ASI) would outperform the best human abilities across every domain by a wide margin.^[4]

 

^{https://en.wikipedia.org/wiki/Artificial_general_intelligence}

 

 

Edited October 27, 2025Oct 27 by martin-w

@LHookins yes! Absolutely it was ELIZA. I remember now on my Atari ST. I don't remember using it much, but it must have left an indelible impression on me all the same, a glimpse into the future perhaps. I remember I was mainly into Flight Simulator 2, Sundog, and Silent Service back then. And Balance of Power was good too!

About some AI assistant, I was thinking of something really word not allowed (eejit) proof. Like if you have a really timid uncle or auntie who has absolutely no clue about anything and is always afraid of messing things up by hitting the wrong key or clicking in the wrong place, one would have some assistant that could autonomously deal with pretty much anything and communicate with you, whether you're satisfied with the results, or if it was something else you wanted. And it could do all of that within seconds. I haven't quite seen anything like that on regular phones or computers yet.

Edited October 27, 2025Oct 27 by Antipodeslonghaul
word not allowed

AI video system can't tell a kid with a gun vs a kid with a bag of doritos.

https://www.pcgamer.com/software/ai/multiple-police-cars-showed-up-to-a-school-after-surveillance-system-monitored-by-ai-flagged-a-bag-of-doritos-as-a-gun-ai-is-not-the-best/

https://www.tomshardware.com/tech-industry/artificial-intelligence/openai-calls-on-u-s-to-build-100-gigawatts-of-additional-power-generating-capacity-per-year-says-electricity-is-a-strategic-asset-in-ai-race-against-china