Author | Lightnews

Reposted by Matt Berryman

Katie Martin @katie0martin.ft.com · 2d

We laugh about it now but that night took five years off my life. God love the medical staff of Cyprus

1 4

Matt Berryman @matt.berryman.social · Aug 22

Full analysis: transactionintelligence.net/uk-age-chec...

UK Age Checks: How a Win Became a VPN Marketing Campaign

Politicians love declaring victory. A new law passes, the metrics move in the right direction, and success is proclaimed. But in complex systems, the first metric to move is rarely the one that matters most.

transactionintelligence.net

1

Matt Berryman @matt.berryman.social · Aug 22

This matters for AI regulation. The EU's prescriptive approach (defining what AI must be) risks the same trap. The US prohibition model (defining what AI mustn't do) maintains adaptability.
In digital markets, second-order thinking isn't optional — it's survival.

1 1

Matt Berryman @matt.berryman.social · Aug 22

Now UK users route their data through opaque overseas VPN providers. Remember Facebook's Onavo? That "privacy" VPN that secretly monitored competitors for acquisition targets?
That's who people trust with their browsing now.

1 1

Matt Berryman @matt.berryman.social · Aug 22

When seat-belt laws came in (1983), compliance hit 90% within days. The London Congestion Charge (2003) cut traffic 15% in month one.
These worked because compliance became easier than resistance.
Age-verification did the opposite. It taught millions to vanish online.

1 1

Matt Berryman @matt.berryman.social · Aug 22

Politicians celebrated when UK traffic to adult sites halved after mandatory age-verification launched in July 2025. Plot twist: VPN sign-ups surged 2,000% in 72 hours.

The UK government had just run the country's most successful VPN marketing campaign. 🧵

1 1 1

Matt Berryman @matt.berryman.social · Aug 4

Anthropic have released a series of twelve lessons to improve your AI Fluency; www.anthropic.com/ai-fluency

Based upon The AI Fluency Framework by Dakan & Feller, it's designed to help improve how people work with AI to get the best quality of outcomes. Well worth 3-4h of you time.

AI Fluency: Frameworks and Foundations

Anthropic AI Fluency course and education

www.anthropic.com

1

Reposted by Matt Berryman

Janine Gibson @janinegibson.ft.com · Aug 1

What does my phone think happened to me four years ago?
Wait… you can hear me right?

7 2 51

Matt Berryman @matt.berryman.social · Jul 24

Sustainable AI can’t be an afterthought. #GreenAI #ResponsibleAI

Matt Berryman @matt.berryman.social · Jul 24

6/ 📊 As the race for ever-larger models continues, it’s time we ask:

Can we innovate efficiently; not just intelligently?

1

Matt Berryman @matt.berryman.social · Jul 24

5/ 🔍 Why it matters:

Most AI labs don’t disclose this info. Mistral’s transparency sets a precedent but also makes the case for environmental reporting standards in AI.

1

Matt Berryman @matt.berryman.social · Jul 24

4/ 🪨 Material Footprint

Training also used 660 kg antimony equivalent—a proxy for electronic waste and heavy metal impact.

That’s a weighty externality for digital progress.

1

Matt Berryman @matt.berryman.social · Jul 24

3/💧 Water Usage

It consumed 281,000 m³ of water—that’s enough to supply ~3,700 people for a full year.

Data centres and model training are thirsty.

1

Matt Berryman @matt.berryman.social · Jul 24

2/ 📦 CO₂ Emissions

Training Large 2 generated 20,400 tonnes of CO₂e—roughly equivalent to the annual emissions of 4,300+ petrol cars.

[Ref: mistral.ai/news/mistra...]

1

Matt Berryman @matt.berryman.social · Jul 24

1/ French AI startup Mistral just dropped the most detailed environmental report we’ve seen on a large language model to date.

Their “Large 2” model came with a serious carbon, water, and resource cost. 🧵 #AI #Sustainability #LLMs

1 1 2

Reposted by Matt Berryman

Katie Martin @katie0martin.ft.com · Jul 11

Amazing name of the day.