Updated: Discover AI, an app with 51 hands-on labs for learning how AI works

Two weeks ago I posted Discover AI, an app with 51 hands-on labs for learning how AI works, from math fundamentals, a single neuron up to agents, fine-tuning and serving. Thanks to everyone who tried it, commented, filed issues or sent a PR. Thank you!! A lot of what's below came out of that thread. GitHub: https://github.com/Fazmin/AILearningGuide new/updated Runs in the browser. All the labs and lessons now work on the web. The desktop app is still there if you want the local AI guide. https://fazmin.github.io/AILearningGuide/#/ Installers for Mac, Windows and Linux. You don't have to build from source any more. Get the .dmg, .exe/.msi, or .AppImage/.deb/.rpm from the releases page. They aren't signed yet, so your OS will warn you the first time you open one. "See it in the lab" cards. This is the main thing in v0.2.0. Lesson sections now have cards that set the lab up to match what you're reading. If the attention lesson says to edit a key and watch every bar move, one click takes you to that step with the key editor ready. There are over 600 of these across all 51 labs, and a "Back to my lab" button puts your own settings back. Simple reading mode. This is a third level below Plain. It uses short sentences and everyday words, with a real-world comparison for every abstract idea. Every Simple lesson is checked to a grade 8 reading level. Someone asked for middle school and high school levels, and this is a first step toward that. Better Learn panel. There's a contents bar for jumping between sections, plus bookmarks and read-aloud. It also remembers where you stopped reading in each lab, and you can copy a link that opens a lab exactly as you left it. AI guide settings. You can now set the maximum reply length and edit the guide's system prompt (thanks boosnie for the PR). The guide can run fully local: one click downloads Qwen3.5 2B and llama.cpp. You can also point it at any OpenAI-compatible endpoint, like your own llama.cpp server or LM Studio. Fixes. The startup crash on Linux Wayland is fixed (thanks MarinX for the report and the help). The web version also shows a loading screen now instead of a blank page while it downloads. Anyways, If you teach this or work in a specialized area of AI, I'd love help with a new lab, a better visualization, or a tweak that makes the cause and effect in an existing lab clearer. I'd also like to hear which labs are confusing and what's missing. As before, the lesson text was written with help from Claude and GPT Astra, with Grok providing a second pass on the references. Every reference was hand-checked. If you find any mistakes, please let me know, open an issue, or submit a PR. And feel free to share. Thank you!!