Kimmy K2: An Open‑Source Thinking Model from Moonshot

Brian, Andy, and Jyunmi kicked off the show with a quick Veterans Day thank-you before diving into one of the most science-heavy shows in recent weeks. Topics ranged from AI-assisted dementia detection and brain decoding to new tools for developers and learners — including Time Magazine’s new AI archive and a deep dive into Google NotebookLM’s new mobile features.

Key Points Discussed

AI in Dementia Detection – A new study published in JAMA Network Open showed that embedding AI into electronic health records raised dementia diagnoses by 31% and follow-ups by 41%, proving AI can catch early warning signs in real-world clinics.

AI Brain Decoder – Scientists used a noninvasive brain scanner to let AI accurately describe what participants were seeing — even recalling or imagining actions like “a dog pushing a ball.” The group marveled at its potential for neurocommunication and ethical implications.

Lovable Hits 8 Million Users – The team discussed the rapid growth of Lovable and its no-code app-building platform, with Brian and Andy sharing personal experiences building and managing credits within the tool.

Time Magazine’s AI Agent – Time launched an AI trained on its 102-year archive, allowing users to query 750,000 stories in 13 languages. The hosts applauded the idea as “the new microfiche” and a model for how legacy media can use AI responsibly.

China’s Kimmi K2 Thinking Model – Andy explained how Moonshot Labs’ open-source reasoning model outperforms GPT-5 in long-form tasks while costing under $5M to train. It’s available via LMGateway.io, which lets developers access multiple AI models through one API.

Dr. Fei-Fei Li on Spatial Intelligence – Briefly previewed for a future episode, her new paper explores spatial reasoning as the next frontier of AI cognition.

Google NotebookLM’s Mobile App Update – Major new features include chat synchronization, flashcards, quizzes, selective source control, and a 6× memory boost for longer learning sessions.

Chrome Extensions for NotebookLM – Two standout add-ons:

NotebookLM to PDF – Saves chat threads as PDFs to add back as notebook sources.

YouTube to NotebookLM – Imports entire YouTube playlists or channels for instant research and study integration.

Tool of the Day – TLDR.wtf (Too Long, Don’t Watch) – A single-developer app that creates highlight reels of long YouTube videos by extracting the highest-signal moments based on transcript analysis.

Live Test on the Show – Brian tried TLDR on a past Daily AI Show episode in real time. It instantly generated timestamped highlight chapters, impressing the team with its speed and potential for content creators.

Timestamps & Topics

00:00:00 🇺🇸 Veterans Day intro

00:03:00 🧠 AI-assisted dementia detection study

00:06:07 🧩 Noninvasive brain decoder

00:11:00 💻 Lovable reaches 8M users

00:15:11 🗞️ Time Magazine’s AI archive

00:19:03 🇨🇳 Kimmi K2 Thinking open-source model

00:25:14 🧠 Fei-Fei Li’s spatial intelligence preview

00:26:29 📚 Google NotebookLM mobile app update

00:31:21 🧩 Chrome extensions for NotebookLM

00:37:41 🎥 TLDR.wtf highlight tool demo

00:45:54 🏁 Closing notes and live-stream mishap

The Daily AI Show Co-Hosts: Brian Maucere, Andy Halliday, and Jyunmi Hatcher