The skinny on what's happening in AI - straight from the previous live session:

Highlights

📰 Wired Gets Fooled by AI Freelancer (And Admits It)

Discussed at [03:41]

Wired magazine just published a mea culpa after discovering they'd been scammed by an AI-generated article. A "freelancer" pitched them a story about Discord wedding communities, wrote the whole piece, and it sailed through their editorial process.

They only got caught when the fake writer insisted on PayPal payments instead of normal freelancer systems and it triggered a deeper look.

Kyle's take: This is the fascinating bit - the actual article was fine. Readers didn't complain, editors didn't spot it, it passed their fact-checking. I’ll repeat myself: the article was sound.

What gave it away was the dodgy payment request. So if the content quality was acceptable to Wired's standards, does it actually matter it was AI-generated?…

The bigger worry is how many other publications have AI content they don't know about. And indeed how many Wired articles are AI generated and just didn’t get caught.

Detection tools are basically useless now - too much money going into generation, not enough into detection. It's an arms race the detectors have already lost. And publishers are on the back foot here.

Source: Wired magazine confession

🔍 Grok Chats Exposed on Google (Again)

Discussed at [11:01]

The same privacy disaster that hit ChatGPT two weeks ago has now hit Grok. Hundreds of thousands of private conversations are showing up in Google search results.

Privacy disaster in progress, as one Oxford professor put it. Though calling it a "disaster" isn't exactly groundbreaking analysis. The real issue is people sharing chats without understanding they're making them publicly searchable.

Kyle's take: 

I did a quick check on the livestream and found everything from people discussing medical conditions to someone asking Grok for detailed instructions on making Class A drugs whilst apparently off their face on DMT…

Makes my Friday nights look dull in comparison.

Want to try it? Use the google search operator “site:grok.com” followed by any keyword.

For example type in “site:grok.com embarrassing rash” without the ““ and you’ll get any relevant shared chats.

Interestingly Google fixed the ChatGPT leak quickly through cooperation, but Grok's still exposed.

Source: BBC investigation

📊 That MIT Study Everyone's Quoting? They Haven't Read It

Discussed at [29:25]

Everyone's throwing around this MIT study claiming 95% of AI pilots fail, but here's the thing - it's really hard to get hold of. MIT put it behind a wall, and when people request copies, they don't send them (according to people in my community who have been struggling to get it)

I actually read the full 26-page study, and it's not what the headlines suggest.

Kyle's take: The study defines "successful AI project" incredibly narrowly - only big flagship top-down implementations count. Meanwhile, 90% of employees at these companies are using ChatGPT daily in their actual work, but that doesn't count as "success" because it's grassroots adoption, not a managed pilot.

The big takeaway for me? Only 40% of companies officially provide AI subscriptions, but 90% of workers use AI anyway. That's not failure - that's shadow IT showing the technology actually works.

The real failure is leadership trying to force AI workflows from the top instead of supporting what their staff are already doing brilliantly at ground level.

Source: MIT/Nanda research (direct link to the paper

Plus here’s a video of me discussing the issue:

🧮 GPT-5 "Discovers" New Maths (Sort Of..Kinda)

Discussed at [55:50]

There’s a viral claim that GPT-5 discovered new mathematics. A researcher gave it an unsolved convex optimisation problem and it provided a better proof than existing papers. Sounds revolutionary, but here's the nuanced take from an actual maths professor.

Kyle's take: Ernest Ryu a maths professor from UCLA (who actually knows this stuff) says the proof would take a PhD student a few hours to work out, but GPT-5 did it in 30 seconds.

Impressive speed, but not groundbreaking new mathematics. It's doing very high-level maths extremely fast, not inventing new maths.

Still extremely impressive though - having a PhD-level mathematician available instantly is going to be transformative for research. Just don't expect the "AI breaks mathematics" headlines with picture of Einstein looking confused to be accurate…it’s clickbait.

Source: Sebastian's research and Ernest Rayo's analysis

Member Question from Hyperfocus: "Are businesses okay with paying a 23-year-old five figures for a couple of weeks' work?"

Kyle's response: Absolutely, if you can prove the value! Age means nothing - Zuckerberg just offered a 24-year-old $250 million to join Meta's AI team. If you’ve got the chops then a 5 figure short contract is not a lot.

If you can save a business six figures, they'll happily pay five. The key is value-based pricing - if your work generates an extra £100k per year for them, charge £20k.

Just make sure you can actually deliver what you promise, and break payments into milestones to de-risk it for them. You’ll also need case studies, testimonials and proof you can deliver to make the sales process go smoothly.

This question was discussed at [27:07] during the live session.

Want the full unfiltered discussion? Join me tomorrow for the daily AI news live stream where we dig into the stories and you can ask questions directly.

Streaming on YouTube and TikTok (follow and turn on notifications for Live Notification).

Audio Podcast on iTunes and Spotify.