The AI Fix

⚠️ Never use an LLM to generate a password. ⚠️ A password is only safe if it’s hard to guess. That means no predictable patterns, no 123456, no p@ssword!, no dictionary words, no pet names, no favourite sports teams. It needs to be random. Something like: G7$kL9#mQ2&xP4!w

On this week's episode of The AI Fix, guest host Greg Iddon joins me to explain how research from Irregular makes it clear, you should never, ever 'vibe' your password.

⚠️ Never use an LLM to generate a password. ⚠️ A password is only safe if it’s hard to guess. That means no predictable patterns, no 123456, no p@ssword!, no dictionary words, no pet names, no favourite sports teams. It needs to be random. Something like: G7$kL9#mQ2&xP4!w

Now, being a human that's terrible at creating passwords you might be tempted to ask your LLM to do it, especially if you're already asking it to do lots of other things. And if you do, it will give you something like this. Something that looks like a strong, random password: G7$kL9#mQ2&xP4!w

When it comes to AI you can have safe or honest, not both. Research suggests that the neurons stopping hallucinations are the same neurons stopping the AI from telling you how to make ricin. Want more honesty? It's easier to jailbreak. Want better guardrails? Sorry, it's a big fat liar now.

To find out more, listen to episode 88 of The AI Fix, where you'll also discover that "subspace orthogonalization" isn't a Star Trek plot device, and that my co-host Corey Noles from The Neuron has definitely never tried to make a nuclear weapon on a quiet weekend.

In episode 87 of The AI Fix, @ai-fix-mark.bsky.social and guest host Nik Roberts take yet another gulp from AI's infinite well of creepy things, and discover a robot that learned to lip sync in the mirror. Sadly, we can't confirm whether it did so while singing into a hairbrush.

Also in the episode, Mark explains to Nik why he shouldn’t put off worrying about AI until after his Pilates class, and Nik stumbles into a social network full of bots obsessing over the temperature in Tokyo.

Do you wish your AI was a bit less... demonic? A little less ghostly and a bit more consultant-y? Strangely, even if you don't, you probably do.

Learn all about it on episode 86 of The AI Fix, where Mark is joined by guest host David Ruiz, who takes time away from his regular gig on the Lock and Code podcast to deliver some truly dreadful news about how AI companies are monetising the loneliness epidemic.

Do you wish your AI was a bit less... demonic? A little less ghostly and a bit more consultant-y? Strangely, even if you don't, you probably do.

In this week’s episode of "The AI Fix" podcast, @grahamcluley.com and @ai-fix-mark.bsky.social take a tour through the wild world of corporate hype - from blockchain to the metaverse to today’s all-consuming love affair with AI.

The Financial Times says three-quarters of America’s biggest firms are name-dropping “AI” on earnings calls Meanwhile, Deloitte’s been caught handing in AI-written reports that cite imaginary court cases, Jeff Bezos wants to put data centres on the moon, and we meet a freakily lifelike robot head.

Notts Police are taking a Unitree robot dog out on patrol. Sound like a good idea to you? Problem is that security researchers have found a security flaw that has left these and other Unitree robots wide open to hackers... raising the spectre of robot-to-robot viral infection using Bluetooth. 😱

Also in this episode... a giant robot spider goes backpacking for a year before starting its job in lunar construction, DoorDash builds a delivery Minion, and a TikToker punishes an AI by making it talk to condiments... 🟢 Spotify: theaifix.show/spotify 🟣 Apple Podcasts: theaifix.show/applepodcasts

In episode 69 of The AI Fix, a shark wears trainers, an AI writes an awful J-Pop song, Graham learns that ants don’t care about AI, Mark predicts the precise date of Graham’s demise, Norway trusts $1.9 trillion to an AI investor, and Florida fights snakes using laughably bad robot rabbits.

Also in this episode, the first AI Fix “you wouldn’t trust a pigeon” mug is spotted in the wild, and the show gets its first bit of feedback from an AI listener. Plus, Graham learns that more people have used ChatGPT than have ever owned a ferret, and Mark finally finds a use for cryptocurrency.

In episode 66 of "The AI Fix" podcast, ChatGPT gives @grahamcluley.com and @ai-fix-mark.bsky.social a terrible lesson in anatomy, boffins at Stanford ruin sushi, and Google Gemini has a self-loathing meltdown. grahamcluley.com/the-ai-fix-66/

Plus... DeepSeek gets an “F” in stopping existential threats to humanity, a robot doesn’t give birth, and a team of AI agents stuns our hosts with an amazing medical breakthrough.

🐦 In the latest "The AI Fix" episode: a pigeon gives a PowerPoint presentation, Mark plays a song about the Transformer architecture, a robot dog delivers parcels, some robots fall over at the Robot Games, and I take credit for one of computing’s greatest insights. open.spotify.com/episode/1b5a...

Plus, @grahamcluley.com explains why Microsoft doesn’t want you to use Excel’s new Copilot feature in any spreadsheet calculations that are meant to be useful, accurate, reproducible, or relied on for anything important.