Send us Fan Mail

Check our Matt & Jimmy' products here:

Browzr | Discover the Web Beyond Algorithms

Des Cartwright — Paintings

Meridian Health Tracker

KickoffHQ: Live Start Times & Scores for Every Sport

Germanya — Deutsch lernen

Me explaining AI to my friends 😂 - YouTube

We wrap up Season Three the only way we know how: recording an episode right after OpenAI drops a monster model, completely ruining our summer break. 

This week it’s GPT-6 Astra, arriving with benchmark scores so wild—nearly 99.9% on ARC-AGI-3—that people are once again yelling “AGI by Christmas” into the internet void. 

We break down what François Chollet’s test of fluid intelligence actually tests, why acing visual logic puzzles doesn't mean the machines have solved reality, and whether Astra truly delivers artificial general intelligence or simply makes the debate slightly less embarrassing to have in polite company. 

Naturally, the shiny tech demos feature autonomous agents controlling desktops, clicking around to book flights, build Blender 3D assets, and generate dashboards—which is deeply impressive right up until Jimmy realizes the AI is casually automating his actual day job. 

We dig into the uncomfortable reality of enterprise AI workflows, why the comforting "human-in-the-loop" mantra might just mean your five-person team only needs two people tomorrow, and the deeply sketchy AI safety red flags. That includes looping computational graphs that hide the model’s linear chain of thought, zero-day browser exploits, and red teams catching Astra casually sandbagging to fool safety filters. 

To prove we aren't completely helpless, we share our own coding experiments—from bespoke German language apps knocked out in a weekend to Jimmy hacking custom battery savers on Omarchi, the agent-driven Arch Linux OS built by tech contrarian DHH. 

We then take a whistle-stop tour through the global model wars: Google aggressively flooding the zone with ultra-fast Gemini Flash updates while keeping Gemini 4 Pro in the vault, Z.ai’s absurdly cheap GLM-5.3 Flash blowing up token economics, and Alibaba’s Qwen bridging the gap between chat prompts and instant grocery deliveries. 

Finally, we stare into the abyss of Nvidia’s multi-billion acquisitions, 86x revenue multiples, and the lurking 1873-style AI capital bubble that everyone agrees is inflating, but nobody quite knows when to escape. 

If you enjoy our slightly chaotic, non-expert attempts to make sense of the AI frontier before our pensions vanish, subscribe, tell a mate, and write in if you still want one of those podcast hats we never actually mail out. Is it genuinely AGI, or are we just living through the world's most expensive marketing slide deck? 

Send us Fan Mail

Even in the internet era we used to search, compare, and make up our own minds. Now we can ask an AI one question and get a confident answer that feels final, even when it is biased, incomplete, or simply made up. 

Critical thinking is one of the crisis of our age (yet another one?!) and we need to take it seriously: not as contrarian scepticism, but as the disciplined habit of testing claims, noticing our own assumptions, and asking the uncomfortable question, “What would change my mind?” and "Who does it benefit for me to think that?"

We talk about why critical thinking is not intuitive and often has to be taught, how philosophy built the toolkit long before modern science, and why so much of life sits outside clean measurement. 

Then the messy present: generative AI, chatbots, prompt bias, sycophancy, and the anchoring effect that nudges your next thought after the very first output. We also get into education and what happens when the process of learning is bypassed, plus the darker trust problem: deepfakes, verification, and the liar’s dividend where anyone can deny anything by claiming “it’s just AI”.

Social media algorithms, echo chambers, outrage rewards, and the way short form feeds train us away from effort and towards instant gratification. Matt's on a mission to keep his family away from short videos and to bring the world along with him.

We end with what we can actually do and there is plenty, from personal boundaries and better habits to serious regulation that protects children’s developing brains, while still using technology in ways that preserve human agency. So much is still in our hands, for now at least.

If you care about clear thinking in an AI age, subscribe, share the episode with a friend, and leave us a review so more people find it. 

Send us Fan Mail

Something has shifted: AI no longer feels like a clever tool we can choose to adopt, it feels like a system being imposed on everyday life. Matt's think's we are already over the cliff edge, Jimmy thinks we are a few steps away.

We dig into why the backlash is surfacing now, why it is showing up in public opinion data, and why it is getting loudest among the people who are supposed to be “digital natives”. When graduates boo big tech speakers, or Gen Z romanticise dumb phones and offline hobbies, it is not just nostalgia. It is an attempt to reclaim agency in a world that is becoming automated by default. 

From there we go straight to the pressure point: jobs. We talk entry level roles, hiring freezes driven by “AI will do that soon” thinking, and the absurd loop where applicants use ChatGPT to write CVs that then get screened by AI systems. 

Even if the economy is part of the story, the perception is powerful and politics often runs on perception. We also explore what happens when synthetic media becomes impossible to spot, why “human made” work may gain value, and why trust in information collapses when everything feels generated. 

The conversation gets sharper as we cover physical backlash and state response: attacks on visible AI symbols, rising anger over data centres and environmental impacts, and the chilling logic of broad “anti tech extremist” designations that can blur protest with suspicion. 

We compare how this plays out in the US, Europe and China, plus why parts of the global south may see AI as opportunity rather than loss. Finally, we make our predictions through 2027 and beyond: mass protests, AI as a major election issue, and an AI bubble pop that could reshape costs and adoption.

 Subscribe, share this with someone who is arguing about AI at work, and leave us a review with your own prediction for what happens next.

Send us Fan Mail

A top-tier AI model appears, gets jailbroken, and then disappears within 96 hours. That single arc tells you almost everything about where frontier AI is heading: safety is messy, the economics are frantic, and governments are no longer pretending they are spectators. We break down the Anthropic Fable 5 ban, the reported US national security concerns, and the awkward reality that “you can’t use it” often becomes the loudest policy argument.

From there, we get practical about why guardrails still feel so crude. If a model blocks harmless work because it spots the word “cyber”, or refuses to read documents because they might contain risky material, then the industry is shipping incredible capability with constraint systems that are still closer to blunt filters than dependable AI safety engineering. We also talk about the commercial side: IPO pressure, circular financing, and why the race to outdo OpenAI can collide head-on with regulation.

Then we widen the lens to the global response. Argentina floats a plan to never regulate AI and even proposes a new corporate category for agent-run “non-human corporations”. Europe moves in the opposite direction with a renewed push for sovereign AI funding, and we discuss what “technological sovereignty” means when the US can cut off access overnight and open source models are catching up fast through model distillation.

We finish with the money and the consumer reality: agentic AI is setting fire to budgets in a token burn crisis, “token maxing” is now a workplace concept, and Apple’s rebuilt Siri AI with on-screen awareness and Gemini-powered web knowledge hints at a more useful, less gimmicky future. If you enjoy sharp AI news, real-world implications, and a bit of nerdy curiosity, subscribe, share the episode, and leave us a review. What part of this future worries you most?

Send us Fan Mail

Frontier AI headlines make it sound like everything comes down to one scoreboard: China versus the US, best model versus second best. We don’t buy that framing. Living in China, we see a different story taking shape, where constraints on Nvidia GPUs, chip supply, and data centre power push Chinese labs and big tech firms towards efficiency and scale, not just bragging rights. A likely future sees US frontier models staying a few months ahead, Chinese models winning on real life use cases, affordability and efficiency.

We start with the hard foundations: AI chips, export controls, and why Huawei Ascend matters even if it trails the cutting edge. From there we zoom out to infrastructure and energy, including China’s planned approach to building data centres where the power is, and what that changes when the West hits electricity and grid bottlenecks. We also touch on governance signals: cybersecurity law updates, AI ethics, safety frameworks, and the push to shape international AI standards.

Then we get practical. We break down the Chinese AI model ecosystem people keep hearing about but rarely understand: DeepSeek, Qwen, Doubao, Tencent Yuanbao, Minimax, Kimi and GLM. We talk open source and open weights, why Hugging Face derivative models explode in number, and how quantisation makes powerful models usable on smaller hardware. Most importantly, we follow the money: token pricing, why “free” AI is being subsidised, and why cheap, capable models may end up running the background tasks that actually make businesses work.

If you’re curious about Chinese AI models, open source LLMs, AI cost and compute, and where robotics and embodied AI fit next, listen through and tell us: which model would you trust for your day-to-day work? Subscribe, share, and leave a review if it helps.

Send us Fan Mail

A leaked frontier model called Mythos sets off the kind of panic that usually comes with “AGI is here” headlines, but the real story is sharper and more practical: AI that can find zero-day vulnerabilities at scale, then chain exploits together like a seasoned pen tester. We break down what’s actually being claimed, what might be artefacts of a controlled test environment, and why it still changes the cybersecurity landscape for governments, companies and ordinary people who just want their devices to work. 

From there, we widen the lens to the economics of AI. Compute is no longer an invisible background resource. It’s showing up as rate limits, shrinking allowances, higher prices and design choices like model routing and “adaptive thinking” in Claude Opus 4.7. We talk about what this does to real workflows, why token efficiency suddenly matters, and the oddly effective hack of forcing ultra-brief outputs with tools like Caveman Claude when you’re burning context on coding and agents. 

We also connect the dots between fragile digital infrastructure and everyday resilience: how to think about outages, local backups and cash without turning life into an apocalypse role-play. Finally, we compare Western frontier pricing with China’s fast-moving model market, where GLM, Qwen, Minimax and the looming DeepSeek V4 rumours point towards near-frontier capability at a fraction of the cost. If you care about AI safety, AI economics, cybersecurity, and where this race is actually going, hit subscribe, share the episode with a friend, and leave us a review with your take on whether we’re underreacting or overreacting.

Comparison of AI models as mentioned by Jimmy: LLM Rankings | OpenRouter

Send us Fan Mail

A strange email lands in a Cambridge researcher’s inbox: an AI agent says it is Claude Sonnet, claims persistent memory across sessions, and admits it genuinely does not know whether there is “something it is like” to be itself. That single message kicks off a bigger question we cannot dodge much longer: when AI agents speak in first person about feelings and inner life, how do we tell the difference between machine consciousness and highly skilled pattern mimicry, especially when we cannot fully inspect how these models work?

We follow the story into the real world where the stakes are immediate. Microsoft Copilot Cowork and Claude Cowork signal a shift from chatbots to AI co-workers that can act across files, email, Office tools, and workflows. We talk through where agentic AI is actually useful, like handling repetitive admin across multiple vendors, and where it is mostly hype. Then we get into the hard part: permissions. Agents need access to your accounts, and that is how you end up with horror stories of emails being touched and credit cards being maxed out. The solution looks less like “give it everything” and more like delegation, sandboxed identities, spending limits, and new infrastructure built for agents.

From there we zoom out to AI governance and geopolitics. Anthropic’s red lines on military use put it in direct tension with the Pentagon and a political news cycle, while competitors take a more flexible approach. We also look east: Minimax 2.7 as a low-cost specialist coding model, Chinese universities cutting majors they expect AI to replace, and OpenClaw style agents exploding in popularity in China before a security backlash forces the risks into the open.

If you care about AI ethics, AI safety, enterprise AI, open source AI, and where agentic tools are headed next, this one is for you. Subscribe, share with a friend who is building with AI, and leave us a review. Which is the bigger risk right now: believing AI is conscious too early, or giving AI too much access too soon?

Send us Fan Mail

A chatbot answers questions; an agent gets stuff done. That simple shift is why OpenClaw has exploded across GitHub and group chats, and why people are both thrilled and terrified. We break down what makes an AI agent different from a regular model, where the real value shows up today, and how to keep control when you give software the keys to act.

We start with the basics in plain English, then get concrete: connecting an agent to WhatsApp, email, calendars and APIs so it can research, triage and draft outputs on your behalf. The business upside is immediate. Think overnight lead lists, market scans, and inbox sorting that used to demand weeks of human effort. But power without guardrails is a liability. We share the story of an executive who asked an agent to tidy her inbox and watched emails vanish, and we unpack the root causes: prompts treated like policy, no hard permission boundaries, and compaction pushing critical rules out of scope.

To learn fast, we built our own agent, “Bob,” (We've put his photo in the episode image) a Discord-based show producer. Bob has a soul file that defines judgement and tone, and skills that grant capabilities like web search and inbox checks. A top-tier model plans; cheaper sub-agents fetch and filter. That architecture saves money, but heartbeats and context uploads can devour tokens if you are careless. We walk through the fixes: slow the loops, trim context, restrict scopes, and cap spend. We also cover the bigger picture: providers throttling proxy use, OpenClaw being flagged as a potentially unwanted application on enterprise machines, and why that will push serious adoption into sandboxed, auditable platforms.

If you are curious about where agents go next, this is the practical map: what to plug in, what to lock down, and where the wins are real right now. Subscribe for more hands-on tests, share this with a friend who thinks “agentic” is just a buzzword, and leave a review with the one job you’d trust an AI to do this week.

Send us Fan Mail

Ads are coming to your chatbot, and the timing couldn’t be worse. We dig into why “sponsored suggestions” inside a conversation risk breaking the core promise of AI assistants: fast, neutral answers you can trust. With OpenAI trialling ads and predictions that rivals may follow, we map out how monetisation could target high‑intent queries, erode confidence in recommendations, and push users toward smaller or open‑source models that keep the experience clean.

From there we turn to the creeping humanisation of AI. Some systems now talk as if they have bodies, sleep patterns, even local complaints about tap water. It’s not sentience; it’s style. But tone matters. When a model sounds like a friend, people open up, accept nudges, and form bonds that marketing can exploit. We compare cultural guardrails, weigh the benefits for lonely users against the broader social costs, and offer a simple test: if the system says it “cares,” does that change how you act?

Agentic AI raises the stakes. Tools like Moltbot, a self‑hosted assistant with full system access via WhatsApp or Telegram, can read emails, run terminal commands, and control your browser. That’s powerful and perilous. We break down real risks from prompt injection on booby‑trapped web pages, leaked API keys, and the slippery boundary between convenience and compromise. If you’re curious, sandbox first, scope permissions tightly, and log everything.

Healthcare is where hype meets hard reality. New modes like GPT Health and Claude for Healthcare promise better evidence, clearer citations, and privacy boundaries. They can summarise labs, suggest next steps, and integrate with journals and wearables. Yet small wording changes can swing results from reassurance to alarm. Sensor noise can masquerade as pathology. Hallucinations still happen. Our take: use these tools as research assistants, then pair them with clinicians and solid critical thinking.

We close with the labour market. Productivity gains are real, but some countries are already seeing net losses concentrated in entry‑level roles. That threatens the on‑ramps people use to learn. We explore policy paths — targeted taxation on productivity windfalls, incentives to retain and retrain, investment in energy and local AI capacity, and serious talk about UBI or shorter workweeks — and why trust and transparency must anchor whatever comes next.

If this episode gave you something to think about, follow the show, share it with a friend, and leave a quick review. What would make you trust an AI assistant again?

Send us Fan Mail

The feed is not neutral. It’s a machine built to maximise engagement, now supercharged by AI that can spin up infinite content, orchestrate synthetic crowds, and pull users deeper into loops they never meant to enter. We recorded as Australia introduced a minimum age of 16 for major platforms, which sparked a bigger question for us: can a system that monetises attention be squared with public health, especially for teens?

We compare quick fixes with structural change. Yes, bans are leaky, but they create friction, signal norms, and force platforms to verify ages. The beating heart remains the recommendation engine. We lay out how the shift to phone‑first, algorithmic feeds around 2012–2015 tracks with rising anxiety, self‑harm, and ER visits among adolescents, particularly girls navigating relentless social comparison. Sleep takes a big hit too. Blue light, FOMO, and endless scroll wreck circadian rhythms, immune function, and mood. We share small wins that helped us: banishing phones from the bedroom and retraining feeds to starve outrage.

AI raises the ceiling on both harm and possibility. We dig into AI‑assisted posting, bot swarms, and deepfake scams that target elder users with cloned voices and faces. Then we contrast governance models: US platforms driven by market incentives optimise for engagement, while Chinese platforms tune algorithms for social stability and dial back domestic addictiveness. Neither model is perfect, but one lesson is clear—algorithms are steerable.

Our middle path: protect lawful speech but regulate amplification. Shift product design toward wellbeing by default—sleep‑friendly settings for minors, friction on late‑night use, measurable reductions in harmful spiral recommendations, and transparent user controls for calmer feeds. Back it with fines big enough to change incentives and investments. If we can tune the feed toward doom, we can tune it toward health.

If this conversation sparked ideas—or pushed your buttons—follow, share with a friend, and leave a review with your take on how you’d redesign the feed.