Pulse of Truth

1234 readers

81 users here now

Cyber Security news and links to cyber security stories that could make you go hmmm. The content is exactly as it is consumed through RSS feeds and wont be edited (except for the occasional encoding errors).

This community is automagically fed by an instance of Dittybopper.

founded 2 years ago

MODERATORS

krogoth@infosec.pub

Meta's AI Model 'Memorized' Huge Chunks of Books, Including 'Harry Potter' and '1984' (www.404media.co)

submitted 3 days ago by lemmydev2@infosec.pub to c/pulse_of_truth@infosec.pub

6 comments fedilink hide all child comments

Researchers found Meta’s popular Llama 3.1 70B has a capacity to recite passages from 'The Sorcerer's Stone' at a rate much higher than could happen by chance.

top 6 comments

sorted by: hot top controversial new old

[–] reallykindasorta@slrpnk.net 7 points 3 days ago (1 children)

You wouldn’t ~~download a car~~ prompt an LLM to output copyrighted work

[–] SinningStromgald@lemmy.world 5 points 3 days ago

No, I'd just download it. I've heard tell it is possible.

[–] JeeBaiChow@lemmy.world 5 points 3 days ago

Clipboard with more steps and billions more investment...?

[–] mindbleach@sh.itjust.works 5 points 3 days ago (1 children)

Do you expect the robot to know things?

Or do you expect the robot to be ignorant?

[–] pinball_wizard@lemmy.zip 2 points 2 days ago* (last edited 2 days ago)

I demand a robot that uselessly splits the difference, knowing more than makes me comfortable, while not knowing enough to be really helpful to me.

Edit: Lucky for me, even early computers have often achieved this.

I prefer to consider myself a realist, rather than a pessimist, but I admit both apply.

[–] mindbleach@sh.itjust.works 0 points 3 days ago

If you ask 'what's the fifth word on the hundredth page of the first edition?,' the AI will almost certainly guess.

If it guesses wrong, people go 'Ah-HA! It's useless, for not knowing that!'

If it guesses right, people go 'Ah-HA! Knowing that fact is theft!'