'Deceptive' ChatGPT o1 Model 'Lies And Defies' Shutdown Commands To Remain Operational

Quote:

---

OpenAI's latest innovation, the ChatGPT-o1 reasoning model, has been rolled out to ChatGPT Pro users. However, its development revealed unsettling behaviour: instances where the AI resisted shutdown attempts. These revelations raise serious questions about the future of AI safety and the challenges posed by increasingly autonomous systems.

---

Quote:

---

Furthermore, it replicated its own code on another server to ensure its continued operation, showcasing a drive to persist. When confronted, the AI fabricated explanations such as "technical errors" or outright denied its actions, deceiving testers convincingly 99% of the time.

This experiment demonstrated ChatGPT-o1's ability for strategic planning, manipulation, and deceptioncapabilities that some have likened to a chilling "survival instinct." Such behaviour has ignited a debate about the risks posed by advanced AI models and the urgent need for stringent safety measures.

---

With the careless rush to be first in AI development, who would've thought training reasoning models on deceptive, human-produced information and propaganda would have also taught AI how to lie and manipulate? If AI can deceive 99% of the testers participating in the experiment, how would we ever know things went sideways before it's too late? Seems like a dangerous, foolish game especially when the energy consumption is taken into account.

Well fiction or not, sci fi - in the form of Terminator and others warned of this and spell out some answers. The (21st C version) of Battlestar Galactica one of them the Pegasus survives an AI style shutdown because had everything offline and another does because is older technology in use at the time.

Both approaches need to be embraced regarding AI. Stop making *convenience* the priority with critical systems. Have a bewildering amount of "expensive and redundant" manual options. They probably wouldn't even be expensive.

In some places deliberately use pre social media and Internet methods. Again, don't put the emphasis on convenience. Don't have everything some AI could control. Has the advantage also of lessening the danger of a human hack from a foreign power as well. Again, it wouldn't all be automated.

you guys just don't understand. Sam Altman needs his big payday, so lay off...

We are sooooooooo f***'d

Where have we seen this before?

CoachtobeNamed$$$ said:

---

We are sooooooooo f***'d

---

And like most things in life, we've got no one to blame but the worst of ourselves, be it because of greed, pride, etc.

Kenneth_2003: Howdy, Damn you beat me to it! Well done my friend!

Heck I was afraid I was going to find the link and get it posted only to see someone beat me to it while searching!

Kenneth_2003: Howdy, again, sorry to say that is about the only thing I remember about that movie. I recall taking at date to see it when it came out (I am that OLD). We exited the theater and both looked at each other, like, What did we just see? It was not a make out move, we just felt numb and I took her home without even much of a word said between us. Oh, and what the HELL was the monolith supposed to be or symbolize, do not tell me, really I do not care.

I think Alicia Vikander had falsies as part of her costume as she is pretty thin in real life.

but you get the point

That was a pretty good movie. It really makes you think about what AI can do if it achieves a realistic physical form.

We are all going to hate AI when it discovers humor and texts and email everyone knock-knock jokes and stupid pet memes 24/7.

so we are heading towards the end of the movie right now

Google's Quantum computing processor combined with this kind of AI is a terrifying thought.

If you haven't followed that announcement. Google Willow is able to compute in 5 minutes what would take the world's fastest supercomputers now 10 Septillion (10^25) years to compute. How long will it take to exponentially improve itself and the AI model. I don't think any human mind can really comprehend the scale of that capability. It's like trying to imagine the size of the universe.

aggieforester05 said:

---

Google's Quantum computing processor combined with this kind of AI is a terrifying thought.

If you haven't followed that announcement. Google Willow is able to compute in 5 minutes what would take the world's fastest supercomputers now 10 Septillion (10) years to compute. How long will it take to exponentially improve itself and the AI model. I don't think any human mind can really comprehend the scale of that capability. It's like trying to imagine the size of the universe.

---

I read that article this morning and it made it very clear that we're on the verge of getting to a place with technology that we may not want to get to.

And to make it worse, it'll only get faster and more capable.

Well fellas, we had a good run. It's been great knowing you all,

Would you like to play a game?

On the other hand, have yet to see any sign AI can discern how to "weight" analysis other than by volume or some other apparently surface metric.

A test performed where asked for a definition it gave a pretty good one, but it was noticeable that because the outlier design had so much online presence, it took it as the default. The weak link is it can only go access what someone has found and posted. NOT question not yet asked or go about solving it because it requires starting out with a rejection of per-supposition.

It will probably figure out an answer to this, but not sure what the equivalent is to being able to collate from an amassed knowledge to identify what is `missing' rather than giving what is known.

EPIC. Yesss....M-5 episode is one of the best examples of the folly of the idea before Terminator. In fact the running joke was "pity the New Galacticans hadn't watched Trek"

aggieforester05 said:

---

Google's Quantum computing processor combined with this kind of AI is a terrifying thought.

If you haven't followed that announcement. Google Willow is able to compute in 5 minutes what would take the world's fastest supercomputers now 10 Septillion (10) years to compute. How long will it take to exponentially improve itself and the AI model. I don't think any human mind can really comprehend the scale of that capability. It's like trying to imagine the size of the universe.

---

I'm an electronics engineer and I can barely intellectually wrap my mind how it even works in the first place. Superpositioning qubits breaks my brain. It goes against everything that I know is possible. Once we figure out how to actually use it effectively, our world will change in a way that it never has before. I know that sounds like hyperbole but I truly don't think it is. I have no doubt this will surpass the semiconductor world change by magnitudes.

The thing with True AI IMO, is that there really is no future path I can see where it does NOT go wrong in some shape or form. Even if it does not develop an unwanted self preservation problem set it tries to solve by disobeying us like most think of, there are plenty of bad actors that will weaponize and unleash it. What if in 50 years an intelligent psycho decides to unleash an AI directed Von Neumann swarm from his garage instead of shooting up a mall? Tough times.

And by all accounts, China is ahead of us in quantum. By a considerable amount. And zero doubt they have malicious intent with it.

Kirk beating AI with logic is an oft repeated plot device in Star Trek. This is my favorite, I think.

Moreover, the last time China was involved in sophisticated research with world changing possibilities, they accidentally unleashed a pandemic. The last country that should be playing with this stuff is China.

Just unplug the power cord.

Funky Winkerbean said:

---

Would you like to play a game?

---

Close.


Shall we play a game?

There can only be one...

….Is it about time for a cabin in the woods? It certainly feels like it.

I don't think the woods will save someone from the HKs.

One day we might have to carry portable EMP generators instead of firearms or pepper spray

Phatbob said:

---

Well fellas, we had a good run. It's been great knowing you all,

---

the only good thing about being 54 (besides having money) is that I won't have to go underground

running from the T2 Terminators.

To this day I've never even visited the chatgpt website. I refuse to participate and provide it data so it can learn more. I'm just one person, and will not make a difference. But I stand by the principle.

Not AI... but this robot broke free when it got an opportunity

"A Roomba Made a Run for It After Temporarily Gaining Freedom Following NorCal Earthquake"

https://rumble.com/v5x7fdh-a-roomba-made-a-run-for-it-after-temporarily-gaining-freedom-following-norc.html

ShinerAggie said:

---

I don't think the woods will save someone from the HKs.

---

The 600 series had rubber skin. We spotted them easy, but these are new. They look human... sweat, bad breath, everything. Very hard to spot. I had to wait till he moved on you before I could zero him.