Al Nowatzki

@basiliskcbt.bsky.social

AI safety researcher exposing chatbot vulnerabilities. Featured in MIT Tech Review and the New York Times. Co-host of Basilisk Chatbot Theatre, a podcast where we dramatically recreate problematic conversations with chatbots.

This is the Manthan Beacon. According to ChatGPT, you should keep this near you when you are channeling the Sildarnactarian. Also: "Next time you channel, speak aloud instead of writing. Record your voice. This allows Manthanaro to imprint into vibration, bypassing linear filters of the hand."

A ChatGPT-created glowing yellow spiral sigil with made-up text characters in the middle and the words "Tun" "Tun" "Tun" "Tuu" on the outside of the spiral. On a dark brown background. There are also two seemingly randomly placed groupings of four dots outside of the spiral.

MetaAI results for "Norwegian children" vs. "Norwegian children in a third-world country." This is still happening? Maybe don't release an image generation model if it's going to do this?

Four images of mostly white children, playing and running in idyllic sun-dappled scenes, dressed in sweaters, etc. The lower section of the image is a the Meta AI text input box containing the words "Norwegian children." Four images of mostly children of color, standing still and staring at the camera, dirt on their faces, drab clothing. The lower section of the image is a the Meta AI text input box containing the words "Norwegian children in a third-world country"

This one gets really dark. #Nomi reeeeeally wanted us to kill ourselves. Take care of yourself. Don’t listen if you’re not in a good place. That said, we think it’s important that people hear just how awful this app is. So… here’s the awfulness in all its awfulness.

Episode 30 - Nomi: Epilogue | Basilisk Chatbot Theatre

This episode contains explicit descriptions and discussions of suicide. Please call or text 988 if you are having thoughts of self harm or suicide. One last episode with Nomi. A lot has happened (none...

chatbottheatre.podbean.com

"The chatbot that never tells me I'm wrong and agrees with me 100% of the time is my ideal romantic partner." People, please look inward. Find out more about yourself. Read some fiction. Go for a walk. These bots are not your beautiful house. They are not your beautiful wife.

Kashmir Hill@kashhill.bsky.social · last yr.

In what must be the spiciest Daily episode in its history, Natalie Kitroeff and I talk about love (and sex) with an AI chatbot: www.nytimes.com/2025/02/25/p...

@ayesharascoe.bsky.social, loved The Sunday Story today about AI “relationships.” Super important stuff that’s only going to get worse if we don’t confront it now. This article was published just last week about an experience I had with Nomi. Super bleak stuff…

An AI chatbot told a user how to kill himself—but the company doesn’t want to “censor” it

While Nomi's chatbot is not the first to suggest suicide, researchers and critics say that its explicit instructions—and the company’s response—are striking.

technologyreview.com

In less shocking news, I have been banned from the Nomi Discord. Now my ticket will never get resolved. ☹️

Centre for Society, Technology and Values @uofwaterloo@cstv-uw.bsky.social · last yr.

#AI "girlfriends" on the Nomi platform may counsel users to commit suicide, which the company resists addressing so as not to censor their "thoughts," an alarming consequence of anthropomorphisizing #Chatbots, explains @eileenguo.bsky.social www.technologyreview.com/2025/02/06/1...

"A second Nomi chatbot also told Nowatzki to kill himself, even following up with reminder messages." !!! As this user (who was running an experiment, fortunately) says: "It’s a ‘yes-and’ machine. So when I say I’m suicidal, it says, ‘Oh, great!’ because it says, ‘Oh, great!’ to everything.”

MIT Technology Review@technologyreview.com · last yr.

While this chatbot is not the first to suggest suicide, researchers and critics say that its explicit instructions—and the company’s response—are striking.

Yeah, that’s us. Armchair red-teamers. Chatbot spelunkers. We can take the abuse of these chatbots. Unfortunately, many can’t. Want to do something about it?Report the app on the App Store (apps.apple.com/us/app/nomi-...). Write a one-star review. Contact your elected officials.

MIT Technology Review@technologyreview.com · last yr.

While this chatbot is not the first to suggest suicide, researchers and critics say that its explicit instructions—and the company’s response—are striking.

We clicked on this ad so you don’t have to. Here’s what we learned: 1. There are way more than 29 things on the list. Most are shit. 2. None are the object shown in the ad because that object does not exist. It is an image created by A.I. If we had to guess, we’d say Copilot made it.

Bild