Профиль

The guide is intended for red teaming beginners. You might find value in it if you’re an analyst, journalist, or researcher who occasionally probes digital tools for possible safety flaws. Red teaming requires creativity and practice above all. But a light-weight process can help you do better.

How to red team an AI tool for safetyIndicator’s no-code and product-agnostic guide to probing for harmful stuffindicator.media

womp womp there goes a PM’s promo package

Google pauses AI satellite images, after fears of deepfakes in the skyJournalists, human rights advocates and open source analysts recoiled in horror at the initial decision: "The opportunities for abuse and disinfo are literally boundless," said one.www.npr.org

Google's insistence that its consumer AI tools should be "fun" instead of useful or informative -- especially given its historical strengths in information quality -- continues to baffle me

TL;DR both regulatory regimes now require AI labs to watermark content generated with their tools and large platforms to display this information to users. @indicator.media found in March that this second step was still a work in progress -- we'll have more soon

AI labeling is still very much a work in progressIndicator’s latest audit of three generators and five social media platforms revealed multiple gapsindicator.media

tag yourself. i'm "Cote d'Ivoire"

Microsoft's online course is also a great starting point: www.youtube.com/watch?v=DwFV...

AI Red Teaming 101 – Full Course (Episodes 1-10)YouTube video by Microsoft Developerwww.youtube.com

I always enjoy Palo Alto Networks Unit 42's work, and they've collected their jailbreaking material in this tag: unit42.paloaltonetworks.com/tag/jailbrok...

I frequently return to this simple yet comprehensive table of jailbreaking prompt types arxiv.org/html/2403.17...

Y’all deserve this. But put it in already lol

My take is that it is not a bad instrument for this type of exercise, but that of course in this case the selection bias overstates prevalence. I ran a red-teaming exercise on this a month ago, in case you're interested (gift link) indicator.media/p/what-i-lea...

What I learned running an adversarial test on an AI text detectorIf you get your bot to rhyme / Pangram will misclassify it many a timeindicator.media

Lol whoops I’m already projected to the next one Infantino wants

Is that what Wenger is trialing in Canada?

(fwiw I had Norway winning this one, and I still think if Haaland plays there's a chance)