Research on small-data poisoning effects

In episode 84 of The AI Fix, Graham and Mark stare straight into the digital abyss and ask the most important question of our age: "Is AI just a hungry ghost trapped in a jar?"

Also this week, we explore how a shadowy group of disgruntled insiders trying to destroy AI by poisoning its training data, how "vibe-coding" has stopped being a joke with even Linus Torvalds joining in, how Google’s AI health advice could have endangered lives, and why simply asking an AI the same question twice can turn it from clueless to near-omniscient.

Oh, and AI has managed to crack some famously unsolved maths problems in minutes, and Grok gains access to all of the Pentagon's networks? What could possibly go wrong?

Episode links:

• AI cracks fantastically hard maths problem.
• Google removes some AI Overviews after users’ health put at risk.
• Is AI just a hungry ghost trapped in a jar?
• Grok will be integrated into Pentagon networks later this month.
• Google reveals that you should always ask your AI twice.
• Employees waste 1.5 weeks a year ‘correcting’ AI output.
• Poison Fountain.
• AI industry insiders launch site to poison the data that feeds them.
• Small amount of poisoned data can influence AI models.
• Poison Fountain reddit thread.

The AI Fix

The AI Fix podcast is presented by Graham Cluley and 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!