c/technology · by sanitation@lemmy.today · 79d

Scientists Invented a Disease to Test Whether A.I. Knew It Was Fake. Then, Chatbots Started Saying It Was Real

https://www.smithsonianmag.com/smart-news/scientists-invented-a-disease-to-test-whether-ai-knew-it-was-fake-then-chatbots-started-saying-it-was-real-180988924/

311 points · 17 comments · view on lemmy.world

17 Comments

nocturne@slrpnk.net · 96 pts · 79d (2 replies)

The eye condition bixonimania doesn’t exist, but neither bots nor some researchers caught that the content was fabricated—despite obvious clues

The first clue being an eye condition with the name mania in it, denoting that it is a mental condition.

adespoton@lemmy.ca · 32 pts · 79d (1 reply)

Citing Star Trek, LotR and Friends as sources should also have been a clue.

I do wonder whether the studies that cited theirs were actually AI generated themselves….

AngryRobot@lemmy.world · 3 pts · 78d

Of course those studies were AI generated.

webghost0101@sopuli.xyz · 74 pts · 79d (1 reply)

I read this back in april,

The entire research were very nervous about doing this cause they essentially post misinformation. So they filled it to the brim with clear signs that its fake, tons of pop culture references and nonsensical wording.

msage@programming.dev · 4 pts · 79d

Not the first time something like that happened, and everyone always buys into it, because no one actually reads studies.

TropicalDingdong@lemmy.world · 54 pts · 79d (6 replies)

I was at a bar where some marketing orangutans were discussing that the forefront of marketing was tricking chatbots into promoting your product or service via BS websites.

Cooked. Entirely cooked.

Telorand@reddthat.com · 23 pts · 79d (2 replies)

I see the web fracturing into a gigantic swamp of corporate AI slop and a tiny pool of the indie web.

Windex007@lemmy.world · 6 pts · 79d (1 reply)

What wall keeps the corporate ai slop off the "indie web"?

idealotus@lemmy.world · 2 pts · 79d

I think it's just routers, ISPs and open internet standards (http, etc) that keep that wall up. 🥲

neclimdul@lemmy.world · 3 pts · 78d

I believe the current meta of SEO is seeing how far you can get prompt injecting your site into AI summaries via llms.txt. So... that's cool too.

PrimeMinisterKeyes@leminal.space · 1 pts · 78d

NG SEO.

foenix@lemmy.radio · 1 pts · 79d

I've been getting spammed to use these services... They didn't even bother to check if I still work at the company in question they want to promote.

x00z@lemmy.world · 22 pts · 79d (4 replies)

I remember doing this kind of stuff back in the day against spoonfeeding. People on forums asked for help, so you provided code and snippets but it would never compile. The issues could easily be fixed by somebody with basic understanding.

Smoogs@lemmy.world · 3 pts · 79d (2 replies)

so basically you were training people how to code

x00z@lemmy.world · 10 pts · 78d (1 reply)

It was the common mentality back in the day. Everybody helped for free but spoonfeeding was a no-go. Many people would find a way to help without actually giving the solution. Keep in mind information was much harder to find so a lot of answers were also links to places where you could find the answer. Most of it boiled down to RTFM.

Smoogs@lemmy.world · 1 pts · 78d

[ removed ]

kamikazerusher@lemmy.world · 2 pts · 78d

I’ve come to appreciate this approach in the cybersecurity world. CVE exploits and proof-of-concepts from experienced authors get included in tools like Metasploit. Individual authors will intentionally do something like put subtle syntax errors or “forget” necessary include statements so something won’t compile or link correctly. Just enough to deter script kiddies but not enough to phase a seasoned tester or experienced developer.

PixeIOrange@lemmy.world · 1 pts · 79d

😱