MoE episode · Oct 19, 2026 · 59 min

Why I Canceled My AI Subscriptions

Not because AI got worse — because I got tired of RENTING my intelligence and decided to figure out how to OWN it. This is the full breakdown: why the industry chased bigger and bigger models, what Mixture of Experts (MoE) actually is, how GLM sharding on my own little GPU fleet gets me serious intelligence without the subscription treadmill, and why your phone might already have the RAM to run this stuff. I even pull up two videos on screen that explain MoE better than I can — so you can chew on it with me.

In this episode

What you'll see

  • The honest part: why my own AI setup has been a disaster
  • The danger is centralization — data centers AND power
  • AI avatars and voice cloning: what I've already built with this stuff
  • Renting intelligence vs owning it — the thing that pushed me to cancel
  • Why the industry stopped chasing bigger models in 2023
  • What is MoE? Mixture of Experts, scribed in plain language
  • Token burn: how AI labs keep making money off you
  • GPT-4 was MoE — widely reported, never officially confirmed
  • How to verify a model when you can't read code: watch the benchmarks
  • Your phone and your laptop can run more than you think
  • Two MoE explainer videos, spaced out with my own take in between
Every claim sourced

Sources cited in this episode

  1. Kernel Cast — MoE vs Dense Models: How AI Saves Compute (YouTube Short, shown on screen) — the dense-vs-MoE explainer my episode cuts to
  2. Maarten Grootendorst — A Visual Guide to Mixture of Experts (MoE) in LLMs (video, shown on screen) — the router / gate-network walkthrough; read the companion article here
  3. Mistral AI — Mixtral of Experts (arXiv) — the open MoE model that made the approach mainstream
  4. GPT-4 as MoE — widely reported, never officially confirmed by OpenAI; discussed in the episode as reporting, not fact
  5. + the full chapter list and timestamps in the video description
How I cite: the two explainer videos I show on screen are credited and linked above — go subscribe to both creators. Primary sources linked wherever they exist; anything I'm not reading from a source is clearly flagged as my opinion. This episode is commentary and criticism, not financial, medical, or legal advice.
Community

Compare notes with me all week

I build all of this in the open, with a crew of AI agents. If you're directing agents too — or wondering where to start — the Discord is where we trade what actually works.