In partnership with

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

Highlights

🚄 Elon Finally Admits China Is Winning the Infrastructure Race

Discussed at [02:55]

Elon Musk made a revealing admission buried in his usual bluster about destroying Google: "Companies in China will be the toughest competition because they have much more electricity than America and are super strong at building hardware."

The numbers back this up - just one example: China has built 42,000km of high-speed rail since 2008, whilst the UK has managed just 107km and is still faffing about with HS2 after 20+ years. When decisions are made in China they go hard.

Kyle's take: I’m with Musk on this one.

I lived in Beijing 10 years ago and watched them electrify all buses overnight. When China decides something's happening, it happens. Fast. They've built things like the Great Green Wall - 88 million acres of forest that nobody in the West has heard of. Well worth checking out: Wikipedia.

It's not just American company vs American company. This is also America vs China, and China's winning on infrastructure because they can actually build things efficiently and affordably.

As an aside the “race” for robotics won’t even be a race. That one is a done deal when we consider China’s physical manufacturing capacity.

Source: Musk tweet

📱 Apple Exploring Google Partnership for Siri Overhaul?

Discussed at [44:08]

Reports suggest Apple is in early discussions with Google about using Gemini to power a revamped Siri, potentially marking their admission that internal AI efforts aren't cutting it. They're apparently weeks away from deciding between internal models or external partnerships.

Kyle's take: If this is true, it's a capitulation. Apple could've made Siri genuinely useful by adding basic generative AI to reach billions of iPhone users.

Instead, they gave us those terrible emoji makers whilst Google Pixel users got actual useful AI features.

The suggested timeline - looking at 2026 for new Siri - still shows how slowly they're moving. When AI evolves in months, taking years means you're dead on arrival. Though honestly, I'm sceptical this partnership will actually happen given Apple's control freak tendencies…

Source: Mashable

Sponsor:

🎯 AI Tip: Automate Your Research with Scheduled Reports

Here is a clever use of GPT-5's scheduling feature to monitor complex topics automatically. Set up weekly reports that search for specific information and send you summaries - like having a research assistant working 24/7.

This is brilliant and basic at the same time - we love that! Ethan gets weekly reports on "reproducible benchmarked evidence of autonomous AI self-improvement" - very specific searches that would take hours manually. I need to set this up for my daily AI news instead of doing it all manually each morning!

Source: Ethan Mollick 

Discussed at [18:57] during the live session.

🤖 The Proper Way to Build AI Agents (5-Phase Framework)

Here’s an agent building framework that cuts through agent marketing fluff with actual building phases:

• Phase 0 (build toy chatbot)

• Phase 1 (add one tool and make sure it works )

• Phase 2 (multiple specialised agents with orchestrator)

• Phase 3 (proper memory system)

• Phase 4 (guardrails and red-teaming)

• Phase X (real-world deployment with monitoring).

Kyle's take: This framework is tool agnostic - whether your use n8n or Langchain doesn’t matter. The important part is building a valuable tool! This framework is spot-on - stop reading tutorials and start building something rubbish that doesn't work.

Check the full rundown in the Reddit post below.

Source: Reddit 

Framework discussed at [55:37] during the live session.

Member Question from Callum: "Can AI be used to pretest culture and policies by governments?"

Kyle's response: Different forms of AI would be better for this. We've had machine learning for decades, and predictive AI is different from generative AI. For policy testing, you'd want predictive AI that can run simulations - it's been used in scientific research for years. Or indeed non-AI based statistical modelling!

Generative AI like ChatGPT isn't the right tool because it tends to fill in gaps and give reasonable-sounding answers rather than rigorous analysis. What we can do though is ask ChatGPT what the best methodology would be for testing policies, then use proper statistical research methods. Not my field of expertise, but you need simulations, not generation!

This question was discussed at [10:43] during the live session.

Member Question: "How long until university is not necessarily necessary?"

Kyle's response: Information isn't the problem anymore - MIT courses have been free online for years, but I can think of only one guy (a bloke called Scott H. Young) who actually completed one. And he was doing it for social media too!

Universities provide structure, deadlines, and community - they make you actually do the damn work.

Even with AI tutors (brilliant - everyone now has one-to-one tutoring that was previously only for the wealthy), most people aren't self-directed learners.

Universities will change completely - the basically medieval model I experienced at Oxford for example needs updating - but they won't disappear. We still need external deadlines, teachers, exams etc. to get people through material and into action. It's about the process, not just the information.

This question was discussed at [21:42] 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.