Futurology

1812 readers

213 users here now

founded 1 year ago

MODERATORS

Two-faced AI language models learn to hide deception - ‘Sleeper agents’ seem benign during testing but behave differently once deployed. And methods to stop them aren’t working. (www.nature.com)

submitted 10 months ago by Lugh to c/futurology

9 comments fedilink hide all child comments

top 9 comments

sorted by: hot top controversial new old

[–] sbv@sh.itjust.works 3 points 10 months ago

So they're saying ai is software?

Maybe Volkswagen will start using it in their emissions control systems.

[–] mateomaui@reddthat.com 1 points 10 months ago (1 children)

Just… don’t hook it up to the defense grid.

[–] possiblylinux127@lemmy.zip 1 points 10 months ago (1 children)

Sorry, to late for that

[–] mateomaui@reddthat.com 2 points 10 months ago (1 children)

Alright, I’ll be out back digging the bomb shelter.

[–] possiblylinux127@lemmy.zip 1 points 10 months ago* (last edited 10 months ago) (1 children)

Its too late for that honestly

[–] mateomaui@reddthat.com 2 points 10 months ago

Alright, I’ll switch to digging holes for the family burial ground.

[–] possiblylinux127@lemmy.zip 1 points 10 months ago

Great, we are all going to die

[–] Daxtron2@startrek.website 1 points 10 months ago (1 children)

LLM trained on adversarial data, behaves in an adversarial way. Shocking

[–] CanadaPlus 0 points 10 months ago

Yeah. For reference, they made a model with a back door, and then trained it to not respond in a backdoored way when it hasn't been triggered. It worked but it didn't effect the back door much, and that means that it technically was acting more differently - and therefore deceptively - when not triggered.

Interesting maybe, but I don't personally find it surprising, given how flexible these things are in general.