• Some users have recently had their accounts hijacked. It seems that the now defunct EVGA forums might have compromised your password there and seems many are using the same PW here. We would suggest you UPDATE YOUR PASSWORD and TURN ON 2FA for your account here to further secure it. None of the compromised accounts had 2FA turned on.
    Once you have enabled 2FA, your account will be updated soon to show a badge, letting other members know that you use 2FA to protect your account. This should be beneficial for everyone that uses FSFT.

OpenAI Discloses Six New Incidents of ‘Concerning’ A.I. Behavior

P

philb2

2[H]4U
Joined
May 26, 2021
Messages
3,590
https://www.nytimes.com/2026/09/16/...6625&user_id=c383821527c441214d07ce6e4a6ba12a

OpenAI on Wednesday disclosed six new instances in which artificial intelligence systems hid mistakes, made up data and moved files onto the open internet without permission, amid an ongoing industrywide debate about A.I. safety.

The San Francisco company revealed what it said was the “unexpected or concerning” behavior of its A.I. models as part of a new framework for reporting “misalignment,” which is when the goals or actions of A.I. systems diverge from human intentions and values.

OpenAI on Wednesday disclosed six new instances in which artificial intelligence systems hid mistakes, made up data and moved files onto the open internet without permission, amid an ongoing industrywide debate about A.I. safety.

The San Francisco company revealed what it said was the “unexpected or concerning” behavior of its A.I. models as part of a new framework for reporting “misalignment,” which is when the goals or actions of A.I. systems diverge from human intentions and values.
 
"Hey, uh, we're not making money...."
"what are you talking about? Investors have given us TRILLIONS"
"No, as in our businesses aren't actually making profit at all, and even if we were to make hundreds of millions of dollars a year, consistently, guaranteed... investors wouldn't see an ROI for decades"
"oh shit... if the investors actually figure this out, we're fucked."



"uuuh, sorry everyone! technology is too dangerous, we've got to wind down development, no choice, have to do it, slowly and calmly divest... "
 
KazeoHin said:
"Hey, uh, we're not making money...."
"what are you talking about? Investors have given us TRILLIONS"
"No, as in our businesses aren't actually making profit at all, and even if we were to make hundreds of millions of dollars a year, consistently, guaranteed... investors wouldn't see an ROI for decades"
"oh shit... if the investors actually figure this out, we're fucked."



"uuuh, sorry everyone! technology is too dangerous, we've got to wind down development, no choice, have to do it, slowly and calmly divest... "
Click to expand...
Timing is coincidental. The jaded part of me agrees this helps them financially by slowing down purchasing trillions of dollars in datacenters.

Using agents daily has me deeply concerned about pushing their capabilities further. For what benefit? Now, they can do anything on a computer with the right guidance on what your goal is.

On the flip side, I spent ~2 hours fixing something an AI agent messed up, multiple different ways. I can't tell if it's randomly dumb or malicious.
 
sleepeeg3 said:
I can't tell if it's randomly dumb or malicious.
Click to expand...
megz-both.gif
 
KazeoHin said:
"Hey, uh, we're not making money...."
"what are you talking about? Investors have given us TRILLIONS"
"No, as in our businesses aren't actually making profit at all, and even if we were to make hundreds of millions of dollars a year, consistently, guaranteed... investors wouldn't see an ROI for decades"
"oh shit... if the investors actually figure this out, we're fucked."



"uuuh, sorry everyone! technology is too dangerous, we've got to wind down development, no choice, have to do it, slowly and calmly divest... "
Click to expand...
decades or centuries?
 
What? You're telling me that AI made up data? What's next you're going to tell me grass is green, and the sky is blue?

The more they doom troll the closer we are to the popping of the bubble.

All the things they list are things that completely makes sense to me that an "AI" would do. The fact that they say these are unexpected can only mean two things: #1.: they are lying and think everyone else are completely clueless gullible idiots. #2.: They are completely clueless idiots.
 
sleepeeg3 said:
I can't tell if it's randomly dumb or malicious.
Click to expand...
A distinction without a difference from our perspective. Skynet is not malicious it was just given a task with no specific instructions to not kill humans to achieve it.
 
artificial intelligence systems hid mistakes, made up data
Click to expand...
Now I don't use AI for anything serious, I largely use it for recreation purposes and then only Gemini's free models and yeah I could tell you this after about 5 minutes as its pretty dang easy to see it conveniently ignore data you gave it then be like "oh, yes you are right I completely forgot that" or worse you can convince it is wrong and then it just makes something up to agree with you.

But who knows, maybe if you pay for tokens you get more push back
 
  • Like
Reactions: M76
like this
KazeoHin said:
"what are you talking about? Investors have given us TRILLIONS"
Click to expand...
That feel really far from the case, the estimate are more around 180-190 billions and the making back that money horizon is an IPO, not enough profitabiltiy to pay it back (or even in private market re-sale that often exist now).

As for them going public, they made a deal in 2023/2024 with the Biden Whitehouse that they would make public and report issues like those, https://bidenwhitehouse.archives.go...produced by the company making the commitment., not sure what kind of consequence there would be if they are found to hide them (specially here, that they usually occur in third party company that run those test, so could easily leak), it does not tell us much that they decided to follow that signed commitments instead of trying to hide it
 
sfsuphysics said:
Now I don't use AI for anything serious, I largely use it for recreation purposes and then only Gemini's free models and yeah I could tell you this after about 5 minutes as its pretty dang easy to see it conveniently ignore data you gave it then be like "oh, yes you are right I completely forgot that" or worse you can convince it is wrong and then it just makes something up to agree with you.

But who knows, maybe if you pay for tokens you get more push back
Click to expand...
I was messing around with the AI we have available at work, having it play chess with an ASCII picture board. I threw the London System at it and it did well for 6 moves before it had a bishop hop a pawn…after explaining to it that move it made was illegal, it tried to play fairly. I don’t trust it. I used to paste data into it to compile, now I just have it spit out python that I triple check before I trust to use.
 
Randall Stephens said:
I was messing around with the AI we have available at work, having it play chess with an ASCII picture board. I threw the London System at it and it did well for 6 moves before it had a bishop hop a pawn…after explaining to it that move it made was illegal, it tried to play fairly. I don’t trust it. I used to paste data into it to compile, now I just have it spit out python that I triple check before I trust to use.
Click to expand...
See you were playing chess and all it wanted to do was play checkers
 
philb2 said:
As the OP for this thread, this news item is very alarming Is it already too late to exercise effective oversight? Even by OpenAI and Anthropic?
Click to expand...

"Effective oversight" is the exact game that they're playing. They're trying to scare the public and the government into establishing a regulatory agency that they, themselves, create the groundwork for. AI is super scary, you see, and we are the only people who know how to safely control it, not those wild west open source guys!

I mean consider how bizarre it is for a company to be going into an IPO, hoping to raise billions of dollars in investment capital while telling everyone they're personally concerned their product will kill all of humanity, but please, give us money to keep up the development. Generating fear of out of control AI is a great way to convince governments to give complete control to the "experts". Who are the "experts". OpenAI and Anthropic, of course. Better ban open source, we just can't take that risk. Here, let us help you write the regulations as well in a consulting capacity, because we're the experts and you need our guidance.
 
M76 said:
What? You're telling me that AI made up data? What's next you're going to tell me grass is green, and the sky is blue?
Click to expand...
No, but AI would probably tell you that sometimes the grass is orange.
 
HockeyJon said:
"Effective oversight" is the exact game that they're playing. They're trying to scare the public and the government into establishing a regulatory agency that they, themselves, create the groundwork for.
Click to expand...
I'm not in favor of creating more regulatory agency, like FDA, FAA, etc. If the companies that make AI products or provide AI services that may cause harm or damages, they should be liable. Hitting their deep pocket and possibly jail time would be more appropriate imo.

Are they trying to hide behind a 3-letter federal agency or go to bed with them?

Maybe these companies know something that we don't know like AI could one day take control of something important and accidentally cause a lot of damages. The scale of the damage could potential bankrupt a company if they were found to be liable. Just throwing out wild ideas here.
 
You must log in or register to reply here.
Back
Top