Podcast
AI with Kyle - Daily AI News and Updates
Hosted by Kyle Balmer · EN
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...
62episodes
Episodes
Newest firstAll episodes
183: How to Write With AI and Never Get Caught
May 500:23:02Tap to summarizeDoes your writing smell like AI? Grab the fix kit (Free Guide): https://aiwithkyle.com/mini/goblins Get AI-Ready with Kyle’s 5-Day Challenge: https://aiwithkyle.com/join Subscribe and turn on notifications to catch the next live stream: https://www.youtube.com/channel/UChlLglbHDASnoGkbjDeHnQg Summary: OpenAI banned the word ‘Goblin’ from their models, and the reason why is a reinforcement learning feedback loop that turbo-charged a "nerdy personality" setting into goblin obsession. This is an interesting insight into how AI writing ticks develop in the first place. Practically, what does this mean for you? I walk you through Wikipedia's living list of AI writing giveaways (yes, they built a 15,000-word guide to protect themselves from AI slop, and yes, I show you how we’ll be using it against them), how to compress it into system instructions and a skill file, and why running your output through both a pre-write and post-write filter is the closest thing to a reliable humaniser you're going to get. I also discuss AI detectors, spoiler: they're worse than useless, and the academic institutions using them to penalise students are in for a rough few years. —— Time Stamps —— 0:00 Intro: OpenAI Banned Goblins From ChatGPT 1:42 Why Goblins? The Reinforcement Learning Loop 4:33 What Are AI Tells? 5:39 Wall of Shame: AI's Most Overused Words 7:46 Wikipedia's Signs of AI Writing Guide 9:42 How to Stop AI From Sounding Like AI (Belt & Braces System) 13:02 Why AI Detectors Don't Work (And Never Will) 15:59 AI Tells Are a Moving Target 19:18 The Disclosure Paradox: Does Anyone Actually Care? 22:00 Free Resources & Wrap-Up — Useful Resources —— Find everything else at https://aiwithkyle.com/
Transcribe →182: GitHub for Beginners: Set Up, Connect AI Tools & Never Lose Work Again
Apr 3000:22:43Tap to summarizeNew to GitHub? Skip the confusion (Free Guide): https://aiwithkyle.com/mini/github Get AI-Ready with Kyle’s 5-Day Challenge: https://aiwithkyle.com/join Subscribe and turn on notifications to catch the next live stream: https://www.youtube.com/channel/UChlLglbHDASnoGkbjDeHnQg Summary: GitHub keeps coming up everywhere if you're building with AI, and nobody ever properly explains what it actually is. So here it is: a complete beginner's guide to GitHub, what it does, why it matters, and how to get it set up and connected to your AI tools of choice, whether that's Claude Code, Codex, Lovable, or Cursor. Think of it as Google Drive for your projects. One central folder that everything syncs to, across every tool and every device. I walk you through creating your first repo, the key terminology you'll actually encounter (commit, push, pull, branch), and how to connect it all up without touching the terminal once. I also show you how I use GitHub beyond code, as the backbone of my content system and second brain. One repo, set up properly, changes how you work with AI permanently. —— Time Stamps —— 0:00 What Is GitHub & Why Vibe Coders Need It 2:05 The Version Control Nightmare Without GitHub 4:05 Create Your First GitHub Repo (Step by Step) 6:12 GitHub Terms Explained: Repo, Commit, Push, Pull 8:42 Cloning, Branching & Merging Made Simple 11:01 Connect GitHub to Lovable, Cursor, Claude Code & Codex 13:12 GitHub Beyond Code: The Obsidian Second Brain Workflow 15:39 Install GitHub Desktop (No Terminal Required) 18:01 Best Practices: Folder Structure & When to Commit 19:57 Your Action Plan: Set Up GitHub Today 21:33 Wrap Up & Newsletter Resources — Useful Resources —— Find everything else at https://aiwithkyle.com/
Transcribe →181: Local LLM Beginner Guide: You Can Actually Run AI For Free (No More $20/Month AI Subscription)
Apr 2900:23:24Tap to summarizeWant to run AI privately on your own laptop? (Free Guide): https://aiwithkyle.com/mini/local-llm Get AI-Ready with Kyle’s 5-Day Challenge: https://aiwithkyle.com/join Subscribe and turn on notifications to catch the next live stream: https://www.youtube.com/channel/UChlLglbHDASnoGkbjDeHnQg Summary: Running a large language model on your own computer sounds like the kind of thing that requires a computer science degree and a spare weekend. It doesn't. In this video I walk you through exactly how to get a local AI model running on your machine today, from downloading LM Studio to picking the right model for your hardware. We cover how to figure out what your device can actually run (RAM is the main thing), what quantization means and why it's not as scary as it sounds, and why Gemma 4 is the sensible starting point for almost everyone. There are real limitations to local models and I'm not going to pretend otherwise. But for general chat, for working offline, for keeping your data off someone else's servers, they're genuinely good and getting better fast. By the end of this you'll have a working local LLM and a decent handle on where this whole space is heading. —— Time Stamps —— 0:00 Intro: How to Run AI on Your Device 2:11 What Is a Local Large Language Model? 3:53 4 Reasons to Run AI Locally (Privacy, Cost, Offline, Learning) 6:35 LM Studio vs Ollama: Best Tool to Start 8:49 Finding Local AI Models on Hugging Face 10:25 Best First Model for Beginners: Gemma 4 11:35 How Much RAM Do You Need? The Parameter Rule 15:36 Quantization Explained: Compressing AI Models 17:08 Picking Your Model Inside LM Studio 19:08 Edge AI: Run AI on Your iPhone or Android 21:06 Hobbyist vs Production: When Local AI Makes Sense 22:45 Outro & Free Resources — Useful Resources —— Find everything else at https://aiwithkyle.com/
Transcribe →180: Your Company Will Replace You With a $100/Month Bot. Plan Accordingly
Apr 2800:33:11Tap to summarizeGet AI-Ready with Kyle’s 5-Day Challenge: https://aiwithkyle.com/join Subscribe and turn on notifications to catch the next live stream: https://www.youtube.com/channel/UChlLglbHDASnoGkbjDeHnQg Summary: The ground has shifted. The nine-to-five used to be the low-risk option, the sensible, boring, reliable choice for anyone who didn't fancy betting everything on a business idea. That deal is off the table. AI is attacking your job at three levels simultaneously: the tasks inside your role, the company that employs you, and the industry you work in. Any one of those can collapse without warning, and most people are sitting there hoping it won't. I'd rather you got a bit annoyed at me now than blindsided later. The good news, and there is good news, is that entrepreneurship has never been more accessible or less intimidating than it is right now. I'm not talking about raising venture capital or moving to San Francisco. I'm talking about making your first quid from a stranger on the internet, building a small side income, and slowly rewiring your brain to understand that you can generate your own money. AI makes the operational parts of running a small business genuinely manageable for one or two people. The risk-reward on doing your own thing has never looked more reasonable by comparison. —— Time Stamps —— 0:00 Why Your Job Isn't Safe From AI Anymore 0:26 When the 9-to-5 Was Actually the Safe Bet 1:32 The Three Levels of AI Risk: Job, Company, Industry 5:19 Take Responsibility — The Brutal Data on AI Layoffs 9:09 The Risk-Reward of Work Has Completely Flipped 13:13 How Gen Z Are Building Their Own Way Out 17:14 Objection 1: "But Not Everyone Wants a Business" 19:56 Where to Start: Freelancing With AI as Your Team 22:15 Objection 2: "But My Job Is Creative" 25:02 Objection 3: "Just Go Into the Trades" 27:19 Why Businesses Can Be Beautifully Simple 31:13 The Bottom Line: Start Generating Your Own Income — Useful Resources —— Find everything else at https://aiwithkyle.com/
Transcribe →179: GPT-5.5 Is Out: Harnesses Explained, Codex, and Why the Model No Longer Matters
Apr 2700:35:05Tap to summarizeChatGPT Images 2.0 Business Guide (Free): https://aiwithkyle.com/mini/chatgpt-image-2 Get AI-Ready with Kyle’s 5-Day Challenge: https://aiwithkyle.com/join Subscribe and turn on notifications to catch the next live stream: https://www.youtube.com/channel/UChlLglbHDASnoGkbjDeHnQg Summary: GPT-5.5 dropped yesterday, which means I spent this morning doing what I always do on a Friday: working out what it actually means beyond the benchmark chest-beating. The short version is that the model itself matters less than you think. What OpenAI has done with 5.5 is build something that works exceptionally well inside a harness, and if you don't know what a harness is yet, that's exactly what this video covers. The concept is simple once you see it, and it changes how you think about every AI release from here on. We also get into Codex, why it's the most accessible entry point for anyone not spending serious money on Claude Code, and the quiet but very deliberate war OpenAI is picking with Anthropic over who actually gets to use these tools. There's also the small matter of Claude Code spending the entire month of March running at reduced intelligence, which Anthropic has now admitted was intentional. A lot happened this week. —— Time Stamps —— 0:00 Intro: GPT 5.5 Just Dropped 0:34 Why Non-Coders Should Care About Coding Models 2:23 GPT 5.5: Intelligence For Real Work 4:32 3 Ways to Access GPT 5.5 (ChatGPT, Codex & API) 7:54 How I Run Agents Overnight (64-Page PDF Example) 9:32 Why You Should Start With Codex (It's Free) 10:43 How AI Agents Actually Work 13:50 Why You Should Ignore Every AI Benchmark 15:28 Testing GPT 5.5 Live: Building a Slide Deck 17:54 Sam Altman's Tweet: An Attack on Anthropic? 23:14 The Harness Matters More Than the Model 26:52 Claude Code Got Dumber in March (Confirmed) 29:13 OpenAI's Push Into OpenClaw 32:08 Choosing Your Harness: A Practical Guide 34:24 Outro & Newsletter — Useful Resources —— Find everything else at https://aiwithkyle.com/
Transcribe →178: ChatGPT Images 2.0 vs Google's Nano Banana: 5 Tests, One Clear Winner — 65 chars
Apr 2400:36:20Tap to summarize(Free Guide) Stop Letting Claude Limits Own Your Workflow: https://aiwithkyle.com/mini/claude-opus-4-7 Get AI-Ready with Kyle’s 5-Day Challenge: https://aiwithkyle.com/join Subscribe and turn on notifications to catch the next live stream: https://www.youtube.com/channel/UChlLglbHDASnoGkbjDeHnQg Summary: I've been testing ChatGPT Images 2.0 against Google's Nano Banana (Gemini's image model) across five real business use cases: YouTube thumbnails, Facebook ads, explainer comic strips, event posters, and LinkedIn carousels. This isn't art generation or anything photorealistic. It's about what actually saves time and produces usable assets when you're running a business and need things to look good without hiring a designer every five minutes. The short version: ChatGPT won all five tests, which I genuinely wasn't expecting. What set it apart wasn't just image quality — it was consistency across sequential prompts, the ability to generate a functional QR code and embed it into a poster, and LinkedIn carousel slides that actually tell a story rather than vague abstract imagery. Imagen is still cheaper and faster via the API, so I'm not writing it off. But if you've been sleeping on ChatGPT's image generator, this is the video that'll change that. —— Time Stamps —— 0:00 Introduction — ChatGPT Images 2.0 vs Nano Banana 2 Pro 3:23 Where to Find the New ChatGPT Image Generator 4:17 Test 1 — YouTube Thumbnail Design Showdown 08:41 Test 2 — Facebook Ad Creative at Scale 14:59 Test 3 — Four-Panel Explainer Comic & Marvel Style 21:16 Test 4 — Event Poster With a Working QR Code 26:36 Test 5 — Turning a Newsletter Into a LinkedIn Carousel 31:51 Final Verdict — 5/5 for ChatGPT Images 2.0 — Useful Resources —— Find everything else at https://aiwithkyle.com/
Transcribe →177: Your AI Subscription Is About to Get a Lot More Expensive (And What To Do About It)
Apr 2300:19:24Tap to summarize(Free Guide) Stop Letting Claude Limits Own Your Workflow: https://aiwithkyle.com/mini/claude-opus-4-7 Get AI-Ready with Kyle’s 5-Day Challenge: https://aiwithkyle.com/join Subscribe and turn on notifications to catch the next live stream: https://www.youtube.com/channel/UChlLglbHDASnoGkbjDeHnQg Summary: Yesterday Anthropic quietly ran an AB test that removed Claude Code from the $20 Pro plan, apparently hoping nobody would notice. They noticed. What started as a pricing page edit turned into a useful window into where AI costs are actually heading, so I put together a presentation on what's going on: why inference is now the bottleneck, how Jevons Paradox means cheaper tokens still lead to higher bills, and why the all-you-can-eat subscription model is living on borrowed time. The practical bit covers what you can actually do about it right now: Codex as a serious alternative (genuinely good, and embarrassingly generous with limits), local models via LM Studio if you're happy spending on hardware, and a set of prompting habits that will meaningfully reduce your token consumption without slowing your output. If you're building anything with agentic tools, this is the context you need before prices stop being theoretical. —— Time Stamps —— 0:00 Intro: The AI Pricing Era Is Here 0:43 Anthropic Pulls Claude Code from Pro Plan 2:39 Codex Team Trolls Anthropic 4:14 What This Means for $20 Pro Users 5:41 The Broken Promise of Cheap AI 7:15 How Agents Changed the Economics 9:35 Inference: The New Bottleneck 11:45 China's Electricity Advantage 12:25 Solution 1: Switch to Codex 14:05 Solution 2: Run Local Models 15:25 Token Efficiency Tips 16:32 The Future: AI Employees, Not Subscriptions 18:53 Outro & Newsletter — Useful Resources —— Find everything else at https://aiwithkyle.com/
Transcribe →176: Claude Design Review: AI Websites, Figma Killer?? & Brutal Usage Limits
Apr 2100:21:06Tap to summarize(Free Guide) Protect Yourself from Opus 4.7’s Brutal Token Limits: https://aiwithkyle.com/mini/claude-opus-4-7 Get AI-Ready with Kyle’s 5-Day Challenge: https://aiwithkyle.com/join Subscribe and turn on notifications to catch the next live stream: https://www.youtube.com/channel/UChlLglbHDASnoGkbjDeHnQg Summary: Claude Design just dropped, and yes, people are already calling it the Figma killer. In this video I walk you through what the tool actually does, how to use it to design websites and app mockups from scratch or from an existing project, and why the "Figma is dead" crowd almost certainly don't use Figma. I cover the full workflow: connecting a GitHub repo, working with the built-in design system, adding inline comments, and iterating on designs the same way you would with a human designer. The honest bit: there are serious limitations. The usage limits are so aggressive I burned through my weekly allowance on one prompt while on the $200 plan, and every single output defaults to the same Anthropic oyster-cream colour palette that is going to become the new purple-gradient AI tell. I show you exactly how to fight that, including a skill pack that gives Claude Design nine alternative style directions so your work doesn't look like it was built by the same tool as everyone else's. If you're an entrepreneur who wants to get something decent out the door fast, this is worth your time. If you're a working designer, come back in a few weeks. —— Time Stamps —— 0:00 Intro: Claude Design and the Figma Killer Hype 0:45 How to Access Claude Design and Its Brutal Usage Limits 1:45 Shadow Drop, Figma Stock Crash, and What This Tool Really Is 3:45 Building a Duolingo-Style App Mockup from Scratch 6:47 Canvas, Comments, and Team Collaboration Features 7:57 Redesigning Your Website with GitHub Integration 10:58 How to Work With Claude Design: Chat, Comments, and Teams 12:05 The Cream Problem: Why Claude Design Outputs Look Identical 13:58 Beating AI Slop with Skill Packs and Exemplar References 16:03 Team Feedback Workflows and Export Options 17:44 Will Claude Design Kill Figma? The Honest Verdict 19:54 Final Thoughts: Usage Limits and Who Should Actually Use It —— More Useful Resources —— Find everything else at https://aiwithkyle.com/
Transcribe →175: Claude Opus 4.7 Breakdown: Community Backlash, Rate Limits, Overcoming Token Waste
Apr 2000:27:02Tap to summarizeFree step-by-step guide to avoiding token waste with Opus 4.7: https://aiwithkyle.com/mini/claude-opus-4-7 Get AI-Ready with Kyle’s 5-Day Challenge: https://aiwithkyle.com/join Subscribe and turn on notifications to catch the next live stream: https://www.youtube.com/channel/UChlLglbHDASnoGkbjDeHnQg Summary: Claude Opus 4.7 dropped yesterday and I've been poking around with it all morning. On paper, it looks great. Anthropic's benchmarks even include their own unreleased Mythos model, which is a first, and Opus 4.7 holds up well against everything else out there. But the community is not having it. Reddit, Hacker News, the usual corners of the internet are all saying the same thing: this is basically the original 4.6 in a new coat of paint, conveniently arriving after weeks of 4.6 getting suspiciously dumber and hitting limits faster. The new tokenizer uses up to 35% more input tokens for the same content, which means "same price per token" is a bit of a magic trick when everything costs more tokens. They've also replaced extended thinking with adaptive reasoning, which auto-routes your queries to different effort levels. Sound familiar? It's exactly what OpenAI did with GPT-5, and everyone hated that too. The bigger picture here is one most people don't want to hear. Anthropic doesn't really want individual users. Their revenue is surging because of enterprise and API sales, not people on the $20 a month plan. They're reportedly hitting a $30 billion run rate, possibly overtaking OpenAI, and nearly all of that comes from businesses paying thousands a day in API costs. So when we complain about limits and sneaky cost increases, it's worth remembering we're not the customer they're building for. Intelligence is getting more expensive, not cheaper, at least for now. The days of unlimited access to frontier models for a flat monthly fee are properly behind us. —— Time Stamps —— 0:00 Claude Opus 4.7 Launch: Community Reaction & Early Verdict 05:36 Adaptive Reasoning Removed: A Step Backward for Power Users 08:31 The Hidden Price Hike: New Tokenizer Costs Explained 11:21 MRCR Benchmark Drop: What Anthropic Didn't Tell You 13:52 Usage Limits & Safeguards: Getting Worse, Not Better 16:16 Anthropic's Real Priority: Enterprise Over Individual Users 18:40 The Brutal Truth: AI Is Getting More Expensive 21:13 How to Survive the Token Crunch: Practical Usage Tips —— More Useful Resources —— Find everything else at https://aiwithkyle.com/
Transcribe →174: Complete Beginner’s Guide to Anthropic’s Claude Code App
Apr 1700:12:14Tap to summarizeGet AI-Ready with Kyle’s 5-Day Challenge: https://aiwithkyle.com/join Kyle’s AI Canon — Curated list of AI learning essentials: https://aiwithkyle.com/ai-canon Subscribe and turn on notifications to catch the next live stream: https://www.youtube.com/channel/UChlLglbHDASnoGkbjDeHnQg Summary: Anthropic dropped a new Claude desktop app yesterday, and the headline feature for me is the preview pane now baked into Claude Code. That's the bit that used to scare people off. You'd hear about Claude Code, try the terminal install guide, see a wall of curl commands, and quietly back out of the room. Now it works more like Lovable or Bolt, where you chat on one side and watch your project come to life on the other. So I thought I'd do a proper beginner's walkthrough for anyone who's been hovering near Claude Code without actually pressing the button. I take you from zero: downloading the app, starting a new local session, creating a folder, talking to Claude about what you want to build, all the way through to the preview pane and the new element selector that lets you point at a bit of the page and say "this looks terrible, fix it." —— Time Stamps —— 0:00 Claude Code for Beginners: New Desktop App Makes It Easy 1:27 Inside the Claude Desktop App: Chat, Cowork & Code Explained 2:35 Setting Up Your First Claude Code Project from Scratch 4:33 Talking to Claude Code: Voice Input & Describing Your Build 6:34 Claude Code's New Preview Pane: See Your App in Real Time 7:35 Iterating with Claude Code: Giving Feedback & Fixing Errors 9:48 Full Step-by-Step Recap: How to Use Claude Code Today 11:31 Closing Thoughts & Where to Find the Community —— More Useful Resources —— Find everything else at https://aiwithkyle.com/
Transcribe →