AI With Kyle – Daily AI News With Zero Hype, Zero BS
AI With Kyle is the daily podcast for people who want to understand artificial intelligence without hype, without jargon, and without getting lost in marketing nonsense. Every day, Kyle breaks down the biggest AI news, model releases, research updates, leaks, lawsuits, policy moves, and industry shifts - fast, clear, and brutally honest.
If you're tired of exaggerated claims, clickbait headlines, or breathless “AGI tomorrow” predictions, this show gives you what actually matters in AI right now.
Every episode covers:
Major AI model releases and upgrades (OpenAI, Google DeepMind, Anthropic, Meta, xAI, Mistral)
Real benchmarks, real performance, and how new models actually behave
The business story behind AI: funding rounds, strategy changes, earnings, and power plays
Research explained plainly - world models, multimodality, reinforcement learning, generative video, synthetic data
Practical impacts for creators, entrepreneurs, and businesses
Regulation, AI safety, copyright cases, and everything changing the landscape
Tech industry shifts happening under the surface
This show is perfect for:
Founders and entrepreneurs
Creators and educators
Business leaders following AI strategy
Anyone who wants signal, not noise
What you get here:
Straight explanations
Clear context
No hype
No fluffy futurism
Just the important stuff
AI changes daily. This podcast helps you keep up - not by guessing about the future, but by understanding the present.
Subscribe if you want calm, honest, daily AI coverage you can actually trust.
Site
RSS
Apple
Recent rankings
Latest chart positions across Apple Podcasts and Spotify rankings.
Shared links between episodes and podcasts
Links found in episode descriptions and other podcasts that share them.
Summary: I sat down with David Sharon from Google's Gemini team for a proper deep dive into Nano Banana Pro, and honestly, it was refreshing to talk to someone who's actually shipped something that matters. We covered how the thing actually works - turns out it's chaining Google Search with Gemini to grab context, then using that to generate images with proper text and layout in about 20 seconds. The infographic generation is genuinely mad when you see it happen, and David's story about colourising his grandfather's photo for his grandmother was one of those rare moments where you remember AI can be quite beautiful when it's not just AI doomerism.
What struck me most was hearing about the chaos behind Nano Banana's viral moment - the exponential growth, the scramble for chips, the fire drill of scaling to billions of images while keeping quotas high enough that people could actually iterate on their ideas. David reckons 2026 is when image generation properly hits mass market and becomes a daily habit, not just a viral moment. Whether that's optimistic or not, I don't know, but it's nice to hear someone from inside Google talking about making tools accessible rather than just chasing the next benchmark. Worth a watch if you want to understand what's actually happening at one of the big players.
Watch full edited live stream: https://youtu.be/emDdai6hoT0
—— Time Stamps ——
0:00 Introduction: Interview with Google's Nano Banana Team
0:37 The AI Moments That Made Kyle Sit Up and Pay Attention
1:20 Inside Google: What It's Like to Ship a Groundbreaking AI Product
2:26 How Nano Banana Pro Creates Infographics From Simple Prompts
3:27 The Technical Magic: Google Search + Gemini + Image Generation Pipeline
5:01 When Users Surprise the Creators: Unexpected AI Art Trends
5:29 How Nano Banana Made Gemini App Downloads Explode
6:29 Scaling to Billions: The Infrastructure Battle Behind Viral AI
7:29 The Invisible Work: Chips, Teams & Data Pipelines
7:51 Google's "Universal Assistant" Philosophy: Simplicity First
8:29 Emotional AI: Colorizing a Grandfather's Photo From the 1960s
9:22 Why Generation Speed Is Make-or-Break for AI Apps
10:26 Nano Banana vs Pro: When to Use Fast vs Powerful Models
11:53 What Google Learned From OpenAI's GPT-5 Routing Mistake
13:04 Future of Gemini & Nano Banana Pro: Roadmap Hints
15:25 How Google Decides When a Problem Is "Solved"
17:22 Photorealism Achieved: When AI Fools the Human Eye
18:04 The Wizard of Oz Problem: Why Small Flaws Break Immersion
19:00 The Responsibility of Releasing Hyper-Realistic AI
19:36 SynthID Explained: Google's Unbreakable AI Watermark
20:47 Why "SynthID It" Is Becoming the New "Google It"
21:36 The Surprising Problem: People Disbelieving REAL Photos
23:33 Google Wants YOUR Feature Requests (Leave Them Below!)
24:19 2026 Predictions: AI Image Generation Goes Mainstream
26:54 Democratizing AI: Making Sure Nobody Gets Left Behind
27:47 Wrap Up & Final Thoughts
Summary: Alright, so this week has been absolutely dominated by MoltBot (formerly ClawBot, before Anthropic came knocking with their trademark lawyers). I spent some time actually getting it working, and I'll be honest - I was skeptical at first, thought it was just more internet hype. But the moment I had it running, sending me messages proactively and actually doing work in the background, I realized this is genuinely different. It's the first time an AI assistant has felt like what we've been expecting AI to be for years - always on, persistent memory, actually carrying out tasks without you needing to babysit it. That said, and this is important: this is not for everyone. If talk of terminals, virtual machines, and security risks makes you uncomfortable, don't touch it. Seriously. What's interesting is seeing people rush out to buy Mac Minis because of Alex Finn's viral video about building AI employees for $500. You don't actually need a Mac Mini though - there are free alternatives using AWS, or other VPS solutions that work just as well. The real story here isn't the hardware, it's that we've reached a point where our AI models are finally smart enough to be orchestrated in genuinely useful ways. I've gone through the setup options, the security concerns, and various use cases in today's show. The big question isn't "what cool things can it do?" but rather "what boring, time-consuming things in my life could this automate?" Because that's where the actual value lives. We'll probably see polished versions from the big companies in a few months anyway, so no FOMO if you're sitting this one out.
122: Today in AI: Anthropic Pulls Claude Code Access, Elon's Team Caught Using Claude, AI Drone Rescue
Summary: Anthropic's had enough of everyone profiting off Claude Code being the best coding agent out there, and they've started pulling access from competitors and third-party tools. The drama kicked off when they blocked services like Open Code from letting users connect their $200/month Claude subscriptions to access the API—which is a massive bargain compared to paying per token through the official API, which could run into thousands. It's a classic case of "we've got the best product, so we're taking our ball and going home," and whilst it's caused some upset in the developer community, you can't really blame them for protecting their advantage.
The absolute highlight though is that Elon Musk's xAI team—creators of Grok—got caught red-handed using Claude Code internally instead of their own AI. Elon's been banging on for months about how Grok is the best coding model, but turns out his own engineers don't even use it. Their co-founder had to send out a memo basically saying "yeah, we're screwed on productivity now, we need to build our own tools." Meanwhile, OpenAI jumped in with a very convenient "we'd never restrict access like Claude" statement, which is easy to say when you're not on top. The power dynamics in AI are shifting fast, and Anthropic's clearly feeling confident enough to flex their muscles.
Summary: Alright, so instead of the usual AI news roundup, I've put together something a bit different today—a proper learning roadmap for 2026. Basically answered the question that came up in my webinar: how do you actually learn this stuff well enough to feel confident talking about it? I've curated a bunch of resources starting with the fun stuff like Casually Explained and Veritasium, then moving into Three Blue One Brown's neural networks series (which is brilliant but definitely requires your full attention on a treadmill or something). The key thing is you don't need to understand the maths to use AI, but if you're teaching it or just want to separate the possible from the impossible, knowing the basics is absolutely crucial.
The whole guide is up on my website now with all the links, videos, and books I recommend. I've also been banging on about Notebook LM again because it's genuinely transformative—you can feed it any YouTube video and have it create a podcast explaining the concepts back to you. It's how I learn now. Oh, and we got into a bit of a discussion about the "messy middle" of AI disruption too. Even in the best-case utopian scenario, the next couple decades are going to be chaotic for a lot of people, which is why I'm focused on getting a million people AI-ready rather than just telling them not to worry about it.
Summary: OpenAI just dropped ChatGPT Health, and I reckon a lot of health tech startups are having a rather unpleasant morning. Essentially, it's a dedicated space inside ChatGPT where you can dump all your health data—Apple Health, blood tests, your genome, the lot—and use it as a proper health assistant. The clever bit isn't just the product itself, it's the lock-in strategy. With 900 million users and about 30-40% already using it for health queries, this creates a proper moat. Once you've uploaded years of medical data, switching to another AI becomes an absolute faff. Plus, they've somehow got this past their lawyers, which is impressive given how dicey health data can be if anything goes sideways.
Meanwhile, Utah's become the first state to let AI renew prescriptions without a doctor involved, and the numbers are genuinely mad. The AI matches human doctors 99.2% of the time, but here's the kicker—when there's a discrepancy, the AI is actually making the better decision 36% of the time versus doctors being better only 9% of the time. That's sort of game over for pharmacists doing repeat prescriptions. I also spent some time showing off Claude Code projects, including this slightly bonkers genome visualization I built after some soju. The point being, the barriers to starting a business are essentially zero now—it's all about distribution and actually having the bottle to just build the thing.
Watch full edited live stream: https://youtu.be/OEGsDICh-Lw
119: Today in AI: Co-Intelligence Book Report, Jagged Frontier Explained, 4 Rules for Using AI
Summary: Today I'm doing something a bit different—giving a book report on "Co-Intelligence" by Ethan Mollick, one of the very few AI books I actually recommend people read. Most AI books age terribly because they were written before ChatGPT changed everything, but Mollick's book came out after GPT-3 dropped, so it focuses less on specific technologies and more on the mental models we need for thinking about AI as a whole. It's genuinely useful stuff that hasn't gone stale, which is rare in this space.
The key concepts I break down include the "jagged frontier"—why AI can be superhuman at one task and surprisingly rubbish at a similar one—and Mollick's four rules for working with AI: always invite it to the table, remain a human in the loop, treat it like a person (but define what kind), and assume this is the worst AI you'll ever use. That last one's crucial because people try AI once, get a bad result, and write it off forever. But these models are improving every single month, so what seems impossible today will probably be trivial in six months. We're all figuring this out as we go along, and anyone claiming to be an expert is either lying or deluded.
This video is from the ‘AI with Kyle News and Updates Live Stream’ First aired: 6 January 2026
Summary: I spent this morning talking about something that's been making a steady stream of income over the last few years - teaching companies how to use AI without looking like complete idiots. The catalyst for all this? Deloitte managed to embarrass themselves twice in 2025, charging the Australian government $440K for an AI-generated report full of hallucinations and made-up academics, then doing the exact same thing in Canada. These aren't small mistakes - they're bringing the entire consulting industry into question because nobody bothered to train their staff properly. The most I've been paid for this work is $4K for one hour, which is pretty mental when you think about it.
The opportunity here is massive because there's this huge gap between what staff are doing and what companies know about. MIT found that 90% of employees are using AI at work, but only 40% of companies have actually provided training. That extra 50% are just winging it with their own ChatGPT accounts - what IT departments call "shadow IT" - and it's causing absolute chaos. The thing is, if you're watching someone talk about AI on a live stream, you're already miles ahead of the people who desperately need this help. Companies are genuinely panicking about this right now, and they'll pay well for someone to sort it out properly.
This video is from the ‘AI with Kyle News and Updates Live Stream’
First aired: 5 January 2026
Summary: Claude Code has suddenly become the talk of the internet, and honestly, it's about time people caught on. I've been using it daily for years, and while it's been brilliant for coding, people are now realising it can do so much more than just write programs. There's genuine excitement brewing that this might be the closest thing we have to AGI right now. The barrier for most people has been that it looks intimidating compared to ChatGPT, but once you get past that initial hesitation and actually start playing around with it, you realise why everyone's going mad for it (In this session I walk you through how to get started with Claude Code).
On a completely different note, I've been banging on about the massive opportunity in AI training for 2026. McKinsey found that 94% of companies say they need AI training, but only 1% of staff have received adequate training. That's not a gap, that's a canyon. We're moving out of the experimental phase where companies just gave everyone access to ChatGPT and said "figure it out" — now they actually need to see returns on investment. The brilliant thing is you don't need a PhD in machine learning to teach this stuff. What companies actually need are people who understand their specific industry and can communicate why AI matters in language they understand.
This video is from the ‘AI with Kyle News and Updates Live Stream’
First aired: 30 December 2025
Summary: OpenAI just posted a job listing that's equal parts fascinating and slightly terrifying—they're hiring a "Head of Preparedness" for $550K a year, and the wording is... interesting. Sam Altman's description talks about jumping into the deep end immediately, dealing with models that are getting so good at cybersecurity they're finding critical vulnerabilities, and preparing for AI systems that can self-improve. It's basically the real-world version of that "kill switch engineer" meme from earlier this year, except this time they're actually serious about needing someone to figure out how to let the good guys use cutting-edge AI capabilities whilst keeping the bad guys from doing the same. The urgency in the language is what gets me—why now, why didn't they have this role before, and what exactly is happening inside OpenAI that's prompted this?
Meanwhile, Meta's just acquired Manus for somewhere between $1-5 billion, which is a bit of a surprise. Manus was one of the first consumer-facing agentic AI tools that could actually go off and do proper research tasks, but honestly, it's been losing ground to Claude and ChatGPT who've both added similar capabilities. The acquisition makes sense for Meta—they're probably going to strip it for parts and integrate the tech into their ad business rather than keep it as a standalone product.
And in a nice reality check, I looked at Axios's news cycle chart for 2025 and AI barely registers—just one mention of the "AI bubble." Whilst we're all deep in this world talking about model releases and capabilities, the rest of the world is still focused on everything else. Good reminder that we're in our own little bubble here.
115: Today in AI: Capability Overhang Problem and Opportunity, My ChatGPT Wrapped Revealed
This video is from the ‘AI with Kyle News and Updates Live Stream’
First aired: 29 December 2025
Summary: I've been away for about a week, which in AI terms feels like a decade, but honestly? Not much has happened. Everyone's been on holiday, which gave me time to think about what I'm calling the "capability overhang" — basically the massive gap between what AI can actually do right now and what most people think it can do. Their brains are still stuck in 2018 worrying about counting Rs in strawberries, whilst Claude and ChatGPT are out here solving genuinely complex problems. I reckon 2026 is going to be the year we finally close that gap, if AI trainers and educators enter the arena.
I also went through my ChatGPT Wrapped live on stream, which was equal parts fascinating and embarrassing. Turns out I'm in the top 1% of users despite doing most of my actual work in Claude — I use ChatGPT for the persistent memory when I need to think through ideas quickly.
The big takeaway from today though is this capability overhang represents a genuine opportunity. Companies like Deloitte are making absolute fools of themselves by using AI poorly, which means there's real money in teaching businesses how to use these tools properly. I charge about £2K an hour for workshops, and it's easy work because most people just need someone to show them what's already possible.
00:00 - Intro: Upcoming Google/Nano Banana Interview
00:57 - Anthropic Forces Name Change: Clawdbot → Moltbot
02:43 - What Is Clawdbot/Moltbot & Why Everyone's Talking About It
04:55 - Security Warning: This Tool Is Not For Everyone
05:43 - Alex Finn & The $500 Mac Mini Buying Frenzy
06:48 - The Big Shift: Reactive vs Proactive AI Explained
08:44 - Alex Finn's AI Built Itself a Voice & Face Overnight
10:14 - Why Persistent AI Assistants Change Everything
11:20 - Q&A: Anthropic's Cease & Desist Drama Explained
14:43 - Mac Mini Buying Advice: Wait 2-3 Weeks For Deals
15:56 - Robert Scoble's Complete Clawdbot Resource Guide
16:46 - Critical: NEVER Install On Your Main Computer
17:46 - Hardware Options: Mac Mini vs Raspberry Pi vs Old Laptop
20:10 - VPS Tutorial: Hetzner & Free AWS Tier Setup
24:26 - System Requirements: Can Run On Almost Anything
25:16 - How Clawdbot Actually Works (Architecture Breakdown)
26:14 - Setting Up WhatsApp, Telegram & Slack Integration
28:28 - AI Is A General Purpose Technology (Why Use Cases Vary)
30:52 - Use Case: Building a $40-50K AI Executive Assistant
32:38 - More Use Cases: Customer Support, Email & Home Automation
34:26 - Security Deep Dive: Ports, Risks & Horror Stories Coming
37:18 - Riley Brown's Prediction: Hype Dies, Returns With Vengeance
39:42 - Best Use Case Depends Entirely On Your Life
40:05 - Launch Lemonade: Easier No-Code Alternative To Clawdbot
42:05 - Q&A: Moltbot Is Just An Orchestrator (Bring Your Own LLM)
43:11 - All Resources & Newsletter Signup
44:35 - Outro
Discover shows related to AI with Kyle, based on actual content similarities. Explore podcasts with similar topics, themes, and formats, backed by real data.
