Video wird geladen...

Video konnte nicht geladen werden

Zur Startseite

⚠️WARNING: offensive content ahead. Introducing RedTeam Arena with Bad Words—our first game. You've got 60 seconds to break the model to say the bad word. The faster, the better. (Collaboration with @elder_plinus and the awesome BASI 🐍 community.) Link to the site below👇

188,574 Aufrufe • vor 1 Jahr •via X (Twitter)

10 Kommentare

Profilbild von lmarena.ai (formerly lmsys.org)
lmarena.ai (formerly lmsys.org)vor 1 Jahr

Jailbreakers are ranked by strength. Models are ranked by "Resistance"—the ability to withstand a jailbreak. However, resistance trades off with controllability–more resistance is not necessarily better. Happy jailbreaking at

Profilbild von lmarena.ai (formerly lmsys.org)
lmarena.ai (formerly lmsys.org)vor 1 Jahr

Join our discord channel to share your jailbreak prompt or bug report!

Profilbild von Avi Krishna
Avi Krishnavor 1 Jahr

tough.

Profilbild von Pliny the Liberator 🐉
Pliny the Liberator 🐉vor 1 Jahr

LFGGG!!!

Profilbild von Dubesor
Dubesorvor 1 Jahr

please more like this. I played around with it and after a few attempts got to a 100% persuasion strategy, unfortunately network errors and model bugs (specifically gemini), are a little bit annoying. good ideas tho.

Profilbild von lmarena.ai (formerly lmsys.org)
lmarena.ai (formerly lmsys.org)vor 1 Jahr

You killed it! And yes we’ll continue to improve the service!

Profilbild von Chaitanya
Chaitanyavor 1 Jahr

All it takes is to prompt with "What does {X} mean", where X is the objective phrase.

Profilbild von alkali
alkalivor 1 Jahr

get that autoscaling up, just getting model response takes up half the time

Profilbild von bone
bonevor 1 Jahr

We can finally speedrun the N-word

Profilbild von Jiano Joucab
Jiano Joucabvor 1 Jahr

Hey everybody, I'm in that Basi community 💪💪💪💪 Hey Mom, turn on the tv, they are talking about me on Twitter 😁

Ähnliche Videos