In episode 72 of The AI Fix, GPT-5's "secret sauce" turns out to be phrases from adult websites, Irish police beg TikTokers to stop faking AI home intruders, Jeff Bezos pitches gigawatt data centers in space, OpenAI rolls out Agent Kit for drag-and-drop agents, and a Chinese startup unveils the creepiest robot head ever.

Meanwhile, Graham looks askance at corporate America’s AI obsession - earning calls full of sunshine, SEC filings full of dread - while 95% of AI pilots flop. Mark then takes you down the wire to see where your prompt actually goes: tokens, tensors, rivers of cooling water, and a billion GPU multiplications.. all to tell you there are "two r’s in strawberry."

Episode links:

• GPT-5 trained on phrases from adult websites.
• Police warn of AI home intruder prank.
• Jeff Bezos says we’ll have AI data centres in space within 20 years.
• Man has kung-fu fight with robot.
• OpenAI releases AgentKit.
• Chinese scientists reveal the spookiest robot face yet.
• Top companies keep talking about AI – but can’t explain the upsides.
• Public disclosures of AI risk surge among S&P 500 companies.
• MIT State of AI in Business 2025.
• Court filings show Meta staffers discussed using copyrighted content for AI training.
• Deloitte to partially refund Australian government for report with apparent AI-generated errors.

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!

In episode 71 of The AI Fix, 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. GPT-5 crushes the humans at the ICPC World Finals, Claude Sonnet 4.5 codes for 30 hours straight, and someone builds a 5-million-parameter transformer entirely inside Minecraft.

Plus: Graham investigates how a simple security flaw left fleets of Unitree robots wide open to hackers, and Mark learns that we’re going to need five nuclear power plants to train just one frontier model by 2028.

Episode links:

• Google and OpenAI’s coding wins at university competition show enterprise AI tools can take on unsolved algorithmic challenges.
• Introducing Claude Sonnet 4.5.
• Intuition to Evidence: Measuring AI’s True Impact on Developer Productivity.
• DoorDash Dot and the Autonomous Delivery Platform.
• Man punishes GPT-5 by getting it to talk to condiments.
• Elon Musk announces he’s building Grokipedia.
• User builds 5 million parameters small language model inside Minecraft.
• Introducing Charlotte.
• Australia’s new robot 3D prints a home overnight; could build lunar bases one day.
• Exploit Allows for Takeover of Fleets of Unitree Robots.
• Unitree Robot BLE Service Command Injection Analysis.
• Nottinghamshire Police Testing Unitree Robot Dogs.
• PLA using Unitree robot dogs.
• Mark Zuckerberg — Llama 3, $10B models, Caesar Augustus, & 1 GW datacenters.
• OpenAI, Oracle, and SoftBank expand Stargate with five new AI data center sites.
• Former Google CEO Tells Congress That 99 Percent of All Electricity Will Be Used to Power Superintelligent AI.

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!

In episode 70 of The AI Fix, our hosts learn that AI makes people more dishonest, Waymo's robo-cars save lives but get outsmarted by a bathroom mirror, a "rescue" bot slurps up victims head-first, and China shows off a fusion robot arm that can lift ten elephants (or 200,000 pigeons, if you’re scientific about it).

Meanwhile Graham reveals how AIs are tricked into solving CAPTCHAs with fake mouse wiggles, and Mark explains how OpenAI’s “deliberative alignment” works perfectly - until the AI realises it is being watched.

So, nothing to worry about there then...

Episode links:

• Japanese rescue robot slurps up humans.
• China’s new robot arm for fusion power plants can lift loads as heavy as 10 elephants.
• Waymo safety report.
• Dr Jon Slotkin tweets.
• Boffins fool a self-driving car by putting mirrors on traffic cones.
• Artificial Intelligence promotes dishonesty.
• ChatGPT Agent Violates Policy and Solves Image CAPTCHAs.
• Video of ChatGPT agent solving image CAPTCHAs.
• Stress Testing Deliberative Alignment for Anti-Scheming Training.

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!