Keith AI

I paid $600 for Descript, then challenged Kimi K3 to build a video editor I would actually use. The first result worked, but it still looked like AI slop. The difference was teaching the AI my taste: camera movement, emphasis, sound, and specific feedback. Watch the full video: https://youtu.be/SxB3kSepJxE

I paid $600 for Descript. So I challenged Kimi to help me build my own AI video editor. Could it actually make videos I’d want to watch? Building the app was only the beginning. The real challenge was teaching it my taste: when to zoom, which words to emphasize, how to explain ideas with animation, and when to use sound. I show the full process—including the disappointing first attempts, the references I used, and the specific feedback that improved the results. Then I compare Kimi with OpenAI on the same 60-second video editing task and discuss cost, effort, and human intervention. This video is sponsored by Kimi. 🎁 Extra 10% bonus for new users’ first purchase, until Oct 31: https://platform.kimi.ai?track_id=track-77d9c1c36fe642fb97a2cc953cf65f1f 🚀 Join my AI community for the workflow and taste files: https://www.skool.com/keith-ai-3958 📖 My book, The 38X AI Company: https://rumjahn.com/book CHAPTERS 00:00 Can Kimi replace my $600 video editor? 00:31 Eight things the editor needs to get right 03:03 Why “make it more engaging” failed 04:06 Learning from YouTube retention data 04:59 The plan for this experiment 05:51 Building the editor with Kimi 08:59 Testing the working app 10:01 From generic output to better editing 10:23 The editing playbook and taste loop 12:42 Rules for visuals, sound, and transitions 14:06 Making a 60-second vertical video 15:35 Teaching Kimi a creator-inspired style 17:21 Reviewing Kimi’s first attempt 18:23 Fixing framing, zooms, and text 19:22 The improved Kimi result 20:30 Same video, same instructions: OpenAI 21:40 Which result looks better? 22:03 Costs, tokens, and local rendering 22:59 My verdict: would I replace Descript? 24:56 Which AI model would I choose? 25:46 Download the workflow and taste files The biggest lesson: “make it better” isn’t enough. Explain what better looks like, review the result, and save the corrections. #Kimi #AIVideoEditing #HyperFrames

Running AI locally solves access. The next step is teaching it your standards. 📘 Read The 38X AI Company: https://rumjahn.com/book Buzz gives you six built-in local models, but LM Studio opens the door to thousands more. Here’s how to find a model that fits your computer, start the local server, enable authentication, and prepare it for use inside Buzz. Run vision, tool-use, and reasoning models locally on your own machine. #LocalAI #LMStudio #BuzzAI

Build AI agents that learn your standards—not just follow prompts. 📘 Read The 38X AI Company: https://rumjahn.com/book Should you self host your own buzz relay and AI LLM model? 00:00 Start of tutorial

Build AI agents that learn your standards—not just follow prompts. 📘 Read The 38X AI Company: https://rumjahn.com/book This is a video about Shared compute 00:00 Shared Compute Mesh LLM 00:40 Mesh LLM Shared Compute