Technology

59534 readers

3199 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 1 year ago

MODERATORS

1102

CEO of Google Says It Has No Solution for Its AI Providing Wildly Incorrect Information (futurism.com)

submitted 5 months ago by Stopthatgirl7@lemmy.world to c/technology@lemmy.world

444 comments fedilink hide all child comments

You know how Google's new feature called AI Overviews is prone to spitting out wildly incorrect answers to search queries? In one instance, AI Overviews told a user to use glue on pizza to make sure the cheese won't slide off (pssst...please don't do this.)

Well, according to an interview at The Vergewith Google CEO Sundar Pichai published earlier this week, just before criticism of the outputs really took off, these "hallucinations" are an "inherent feature" of AI large language models (LLM), which is what drives AI Overviews, and this feature "is still an unsolved problem."

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

[–] Excrubulent@slrpnk.net 2 points 5 months ago

Prompt injection has shown us that basically any attempt to limit the output like this is doomed to fail. Like anti-piracy ones, where if you ask directly for the info it says no, but if you ask for the info under the guise of avoiding it, it gives up everything.

Or for instance with the twitter bot, you could get it to regurgitate its own horrifically hateful prompt, then give it a replacement prompt and tell it to change its whole personality, then tell it to critique its previous prompt. There is currently no way to create a prompt that has supremacy over the user input. You can't ask it to keep a secret because it doesn't know what a secret is.

I think because we're getting access to hallucinations, it's a bit like telling a person "don't think about an elephant". Well, they just did, because you prompted them to with the instruction. LLMs similarly can't actually control what they output.