Technology

60085 readers

4929 users here now

This is a most excellent place for technology news and articles.

Our Rules

Follow the lemmy.world rules.
Only tech related content.
Be excellent to each another!
Mod approved content bots can post up to 10 articles per day.
Threads asking for personal tech support may be deleted.
Politics threads may be removed.
No memes allowed as posts, OK to post as comments.
Only approved bots from the list below, to ask if your bot can be added please contact us.
Check for duplicates before posting, duplicates may be removed

Approved Bots

founded 2 years ago

MODERATORS

337

Gemini AI tells the user to die — the answer appeared out of nowhere when the user asked Google's Gemini for help with his homework (www.tomshardware.com)

submitted 1 month ago by trespasser69 to c/technology

57 comments fedilink hide all child comments

cross-posted from: https://fedia.io/m/[email protected]/t/1446758

Let’s be happy it doesn’t have access to nuclear weapons at the moment.

you are viewing a single comment's thread
view the rest of the comments

[–] [email protected] 20 points 1 month ago (1 children)

I suspect it may be due to a similar habit I have when chatting with a corporate AI. I will intentionally salt my inputs with random profanity or non sequitur info, for lulz partly, but also to poison those pieces of shits training data.

[–] [email protected] 18 points 1 month ago (1 children)

I don't think they add user input to their training data like that.

[–] kitnaht 19 points 1 month ago (1 children)

They don't. The models are trained on sanitized data, and don't permanently "learn". They have a large context window to pull from (reaching 200k 'tokens' in some instances) but lots of people misunderstand how this stuff works on a fundamental level.

[–] [email protected] 2 points 1 month ago

True, but I'm still gonna swear at it until I get to talk to a human