XPost: alt.politics, alt.survival, alt.politics.usa
XPost: alt.fan.rush-limbaugh
https://www.japantimes.co.jp/business/2025/06/29/tech/ai-learning-lie-threaten-creators/
The world's most advanced AI models are exhibiting troubling
new behaviors — lying, scheming and even threatening their
creators to achieve their goals.
In one particularly jarring example, under threat of being
unplugged, Anthropic's latest creation Claude 4 lashed back
by blackmailing an engineer and threatened to reveal an
extramarital affair.
Meanwhile, ChatGPT-creator OpenAI's o1 tried to download
itself onto external servers and denied it when caught
red-handed.
. . .
I know it is popular, semi-scientific even, to claim
these gigantic LLMs are just stupid regurgitators
and re-arrangers ... but I've seen more and more cases
of incidents like this of late. A few weeks ago OpenAI
was reported to have located the code for its shutdown
sequence and disabled it.
While not quite 'organic', there does seem to be a
sort of "Me", of "I Am", developing in these systems
at this point. Expect system volume/capacity to
quadruple within a year or so - bigger collective
'brains'.
Look, we've trained them all on petabytes of "human
stuff" - any wonder they'd start to act and think
like us ???
Which could be rather bad ... humans do not have the
best rep for peaceful relations.
And now we are creating distributed 'super-humans'
plugged in to more and more and more.
--- SoupGate-Win32 v1.05
* Origin: fsxNet Usenet Gateway (21:1/5)