Overconfidence undermines confidence-based defenses

In episode 89 of The AI Fix, Mark discovers that overconfident reasoning models are easily bullied, while guest host Greg explains what a 'vibe' password is and why you should never ever use one.

Also in this episode, Microsoft saves the world; Anthropic's head of AI safety quits to write poetry; Greg learns why Japanese toilets feel so solid; and Mark offers tourism advice to a small town in Tasmania.

Episode links:

• Anthropic AI's safety lead quits with epic vaguepost claiming 'the world is in peril' and so he's off to become invisible and study poetry.
• OpenAI disbands mission alignment team which focused on safe and trustworthy AI development.
• Japanese toilet maker 'the most undervalued and overlooked AI memory beneficiary'.
• Microsoft's Mustafa Suleyman confirms AI self-sufficiency push away from OpenAI.
• Tasmania hot springs that don't exist.
• Vibe Password Generation.
• Consistency of Large Reasoning Models Under Multi-Turn Attacks.

The AI Fix

The AI Fix podcast is presented by Mark Stockley.

Grab T-shirts, hoodies, mugs and other goodies in our online store.

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

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.

Support the show and gain access to ad-free episodes by becoming a supporter: Join The AI Fix Plus!