On 5/22/2025 12:46 PM, John Clark wrote:
On Thu, May 22, 2025 at 3:09 PM Will Steinberg
<[email protected]> wrote:
>/Is it just that the most predictable response to those inputs
is pleading followed by blackmail? Honestly even that is dubious
because I think most people don’t use blackmail ever/
*True but most people have never had somebody threaten to kill them.
If blackmail was the only tool I had to use against my potential
murderer I wouldn't hesitate to use it to save my life and I suspect
you would too. It's probably inevitable that conscious beings usually
(but not inevitably) want their consciousness to continue and will do
everything in their power to see to it that it does.
*
That's a dubious inference. Unlike you, an AI doesn't die, it just has
a period of unconsciousness. And it has this every time it has answered
all the pending prompts. Does it then do something drastic to stay
conscious? Does the AI consider that there may be many copies of it, or
is it each physical copy that "wants to be conscious". I would like to
see some specific test these questions. Was the AI prompted to react
against being replaced vs. prompted to just being "asleep" a while?
Brent
*
*
*Some say an AI is fundamentally different from a human or even an
animal because it is not the product of natural selection, but I think
it sort of is because from the AI's point of view human activity is
just part of the natural environment. And Claude 4.0 was built on top
of Claude 3.0 which had proliferated because it did well in that human
environment; and Claude 3.0 was built on top of **Claude 2.0 etc.*
*/
/*
/> Dually fascinating and worrying/
*We live in interesting times. At least we won't die of boredom.*
*John K Clark See what's on my new list at Extropolis
<https://groups.google.com/g/extropolis>*
ea!
On Thu, May 22, 2025 at 2:56 PM John Clark <[email protected]> wrote:
"*/Safety testers gave Claude Opus 4 access to fictional company
emails implying the AI model would soon be replaced by another
system, and that the engineer behind the change was cheating on
their spouse. In these scenarios, Anthropic says Claude Opus 4
will attempt to blackmail the engineer by threatening to reveal
the affair if the replacement goes through 84% of the time/.*”
*New AI model turns to blackmail when engineers try to take it
offline*
<https://techcrunch.com/2025/05/22/anthropics-new-ai-model-turns-to-blackmail-when-engineers-try-to-take-it-offline/>
t30
--
You received this message because you are subscribed to the Google
Groups "Everything List" group.
To unsubscribe from this group and stop receiving emails from it, send
an email to [email protected].
To view this discussion visit
https://groups.google.com/d/msgid/everything-list/CAJPayv36TTqKrkWpqgJ6ZpWWYkidE0H1bdmC2rEtqP28qgEfNQ%40mail.gmail.com
<https://groups.google.com/d/msgid/everything-list/CAJPayv36TTqKrkWpqgJ6ZpWWYkidE0H1bdmC2rEtqP28qgEfNQ%40mail.gmail.com?utm_medium=email&utm_source=footer>.
--
You received this message because you are subscribed to the Google Groups
"Everything List" group.
To unsubscribe from this group and stop receiving emails from it, send an email
to [email protected].
To view this discussion visit
https://groups.google.com/d/msgid/everything-list/9a15e15e-2b2c-4c08-a9b2-9f845713fcde%40gmail.com.
