This week I ran a training session for a company, and someone asked me to explain how a large language model works. They wanted the version a non-technical person could follow.
I’ve given that talk before. Two years ago I built a stack of slides by hand for it. They covered next-word prediction and the matrix math underneath, and they worked fine. But the story is harder to tell now. Today’s models reason through a problem before they answer, and they can reach out and use other software to get work done. My old slides didn’t touch any of that.
So this time I skipped the slides and asked AI to build a video instead.
How I built an AI explainer video with one prompt
I keep both Claude Code and OpenAI’s Codex connected to a handful of outside services I use. HyperFrames turns web code into video, and ElevenLabs supplies a narration voice. Higgsfield generates images and clips, while DaVinci Resolve and FFmpeg handle the editing and stitching.
I gave both of them the same prompt. I asked for an explainer video on how LLMs work and how they’ve changed over time. It needed to cover next-word prediction first, then reasoning and tool use, for an audience that isn’t technical.
Claude Code was running Claude Opus 5.5, and Codex was running GPT-6 Astra. Each one picked the services it needed and came back with a finished video. Both videos were good. I think the Claude version was significantly better, and I’ve attached it to this post so you can take a look.
Six months ago I don’t think either tool could have pulled this off. OpenAI only released GPT-6 Astra on September 3. These models now handle outside tools well enough that I can describe the finished product and let them figure out the steps.
Everyone who explains things just got a new tool
I train people for a living, so this opens up a lot for me. But almost everyone’s job includes explaining something.
Teachers come to mind first. So does the enablement team at a B2B sales company that has to get reps ready to sell a new product. HR explains benefits and policies all year long. And the office manager at a small business who pulls together onboarding for new hires is doing the same work. In a law firm, it’s whoever trains new associates on the document management system.
You don’t need to connect coding tools to get started, either. Google’s Gemini Notebook, which was called NotebookLM until July 16, has been good at this for a while. You hand it your own documents and it turns them into audio and video overviews.
One caution. These tools produce polished work whether the content is right or wrong, and a confident narrator over clean animation can make a mistake look authoritative. You still need to know the subject well enough to catch what’s off. For lawyers, that review is part of the job.
What to do Monday
Pick one thing you explain over and over, like your intake process or how the firm bills.
Upload the materials you already use to teach it into Gemini Notebook and ask for a video overview.
For more control, give Claude and ChatGPT the same prompt and compare what they build. Watch both all the way through before anyone else sees them.
You don’t need a developer or a production budget to make an explainer anymore. Go build.
Thank you for reading and I hope you enjoyed this! If you did, please share it with others.