🤔 A good question. We might have to talk about that in episode 90...

⚠️ 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

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.

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.

😂😂😂

Anthropic's Assistant Axis

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.

IF ONLY HE'D LISTENED, RIGHT???

We need a word for "good advice in the wrong place" in the way that "weed" means "perfectly good plant growing where I don't want it".

Mwaahahahahahahahahaha!

Cover art for The AI Fix with Mark's head at the correct size, finally.

Finally, I get the cover art all to myself.

🤖 Apple Podcasts podcasts.apple.com/us/podcast/t...
🤖 Spotify open.spotify.com/show/2Ozk8uN...
🤖 Pocket Cast pca.st/kg11t9po

#AI #ArtificialIntelligence

Plus, we discover brain rot, an AI writes a terrible J-Pop song, Norway trusts $1.9 trillion to an AI investor, @grahamcluley.com learns that more people have used ChatGPT than have ever owned a ferret, and I look at whether AI will destroy the economy.

In episode 69 of @theaifix.show, Florida decides to solve its python problem with some laughably bad robot rabbits.

What's most unsettling about this is—I can't get it out of my head. I don't want it in there, but it's always there, singing away, taking up a little corner of my brain.

Is this the future? AI goes out to work and we just replay endless brain worms?

Also in episode 69, Graham learns that more people have used ChatGPT than have ever owned a ferret, and I finally find a use for cryptocurrency.

🤖 Apple Podcasts podcasts.apple.com/us/podcast/t...
🤖 Spotify open.spotify.com/show/2Ozk8uN...
🤖 Pocket Cast pca.st/kg11t9po

It's utter drivel. Catchy nonsense. Glorious slop.

In other words, it's what every generation of kids inflicts on its parents. And once you hear it, you'll have to smooth your brain folds out with a belt sander before it stops playing in your head.

Has AI created the most annoying thing in the world?

In episode 69 of The AI Fix, @grahamcluley.com and I discover Brain Rot, the Gen Alpha AI craze that's even more annoying than Crazy Frog ringtones.

AI can be vaccinated against evil, and the "Rumble in the Silicon Jungle"

Plus, @grahamcluley.com digs into the spat between Elon Musk and Sam Altman, and I learn that humanity could be saved by vaccinating AI against evil.

🤖 Apple Podcasts podcasts.apple.com/us/podcast/t...
🤖 Spotify open.spotify.com/show/2Ozk8uN...
🤖 Pocket Cast pca.st/kg11t9po

In episode 64 of @theaifix.show, a robot folds some laundry, a robot crab looks for love on the beaches of Portugal, the “Godfather of AI” thinks our only hope is to build motherly AI, the UK government has a terrible idea, and our hosts discover a long-lost sixties soul hit, “Dusty Plasma”.

Is this AGI?

"Image created"

A row of computers, one of which is under protected.

I'm asking GPT-5 to create some illustrations for me, and I'm starting to wonder if its much heralded "tool use" includes MS Paint.

52.8 is greater than 69.1

Apparently, GPT-5's facility with "sustaining unrhymed iambic pentameter" is going to make it better at helping you with "everyday tasks like drafting and editing reports."

As long as those reports don't contain charts, I guess.

Find out more in episode 45 of @theaifix.show, with me and @grahamcluley.com—get it in all your favourite podcast apps.

🤖 Apple Podcasts podcasts.apple.com/us/podcast/t...
🤖 Spotify open.spotify.com/show/2Ozk8uN...
🤖 Pocket Cast pca.st/kg11t9po

Yes, they are predicting the next word in a sentence, but they aren't improvising at each word, they're thinking about the whole sentence and planning ahead.

As Geoffrey Hinton said, "you have to be really intelligent to predict the next word accurately".

Large Language Models plan ahead

Large Language Models (LLMs) don't work the way we thought.

LLMs are sometimes described as next-word predictors or likened to a fancy autocomplete.

Recent research from Anthropic shows that they are far more than that.

AI robot throwing up nuts and bolts.

In our next episode, we find out how AIs often vomit up unpleasant stuff.

Follow "The AI Fix" in all podcast apps.

🟢 Spotify: theaifix.show/spotify

🟣 Apple Podcasts: podcasts.apple.com/gb/podcast/t...

🛑 Pocket Casts: pca.st/itunes/17533...

Everywhere else: theaifix.show/follow/

#podcast #ai

The AI Fix Podcast · Graham Cluley and Mark Stockley · Join tech veterans Graham Cluley and Mark Stockley as they dive headfirst into the hilarious, bizarre, and downright mind-boggling world of artificial intel...

open.spotify.com/show/2Ozk8uN...

The AI Fix Technology Podcast · 42 Episodes · Updated Weekly

Get the latest episode of The AI Fix in your favourite podcast app today!

podcasts.apple.com/us/podcast/t...

All is revealed in episode 42 of @theaifix.show, when I introduce @grahamcluley.com to the "reversal curse".

Graham also has a brainwave about the creepy Neo Gamma robot ads, a crow plays peek-a-boo, humans give up writing, and an AI designs a drug.