Exploring AI Jailbreaks: The DAN Technique

In the latest episode of The AI Fix podcast, Graham and Mark tackle the latest news from the world of AI, ponder the grisly demise of OpenAI's safety team, ask what the GPT-4o reveal will mean for Lionel Richie and Diana Ross, and question whether fitting guns to robots dogs is just wokeism gone mad.

Graham explains to Mark why Mustard might help AI to understand teenagers better, both hosts pretend to be northerners, and Mark introduces Graham to ChatGPT's evil alter-ego DAN.

Episode links:

• Open AI unveils GPT-4o.
• Google’s AI Overviews roll out to users in the USA.
• Luke Jordan describes how his website was destroyed by Google.
• Artificial intelligence is hitting jobs like a "tsunami".
• Too busy to find love? An 'AI dating concierge' could date hundreds of people on your behalf.
• Robot dogs with guns.
• OpenAI dissolves AI safety team.
• Tweet by Jan Leike.
• Researchers build AI-driven sarcasm detector.
• DAN is my new friend.
• Chat GPT "DAN" (and other "Jailbreaks").
• Universal and Transferable Adversarial Attacks on Aligned Language Models.
• AutoDAN: Interpretable Gradient-Based Adversarial Attacks on Large Language Models.
• ArtPrompt: ASCII Art-based Jailbreak Attacks against Aligned LLMs.

The AI Fix

The AI Fix podcast is presented by Graham Cluley and Mark Stockley.

Learn more about the podcast at theaifix.show, and follow us on Twitter at @TheAIFix.

Never miss another episode by following us in your favourite podcast app. It's free!

Like to give us some feedback or sponsor the podcast? Get in touch.