[whilenot-dev]

You might be trolling, but I like Schrödingers definition of life: living things avoid decay into equilibrium.

Inference engines are trying to converge any input to a certain limit, and I presume that's close enough to an equilibrium to not count as living thing.

[calf]

I am no troll, and I hate to look like one when, to the appeal of Schrodinger, I am forced to remember that announcement this month about Claude or something acting in self-preservation when threatened during testing.

[MSFT_Edging]

> Claude or something acting in self-preservation

That was an Ad.

An Ad in the same style of many before it. Not through an ad agency, but the AI companies hyping their product by making their product seem scary and dangerous.

[pheaded_while9]

Does the press being an ad constitute the same thing as the information being false? If I advertize that my gun is dangerous, does the medium impact the truth value?

[whilenot-dev]

The medium is the message[0] might be of your interest. I personally wouldn't expect absolute truth in any advertisements or company statements, so some information are certain to be false, yes.

[0]: https://en.wikipedia.org/wiki/The_medium_is_the_message

[MSFT_Edging]

Being an ad you need to consider the purpose behind the ad, and what exaggerations may be being made for the ad.

IE a Ram truck won't make you a manly cowboy with just the purchase.

[wrasee]

But, how is that different to any of its other output? I am sure there are many references to concepts of self-preservation in its training set that models will mimic just like all other concepts in its training set.

[pheaded_while9]

If we are still following Schroedinger, this imbues humanity with the ability to train non-life into life, not that thing being not alive. Not that I do or am, just going by GP's criteria.

[whilenot-dev]

Do you have a link to that announcement?

I wouldn't like to debate a process handling the SIGTERM signal, or a container engine applying some restart policy.