Bullish on Meta's AI implementation and broader AI app ecosystem

Alex Wang is truly talented. When he joined FB, he figured out how to implement large models and actually built them. Pure researchers who don't know how to play with them can't do a good job with large models. Large models are just tools; what matters is how you plan to use them. Large models + Guardrails can be used for games and 3D. Once three.js animations came out, it should have become clear that Roblox-style game design could be handed over to large models. Primitives don't need to be drawn from scratch every time; I previously did some of this for kids to play with. Large models + HTML5 animations can create videos, which HF and Opus 5.5 have both demonstrated. Modeldock had multimedia built-in from the start, and demand in this area has always been strong. Large models + voice-controlled devices can create handheld conversational AI companions, which are especially important for the elderly and children. Don't underestimate the barrier of typing; this threshold filters out more than half of the global population. Before Muse Charm came out, I was researching this device. Options included making a phone app (too big and heavy) or an EPS32 device (too light and limited). But I always thought ChatGPT would come out with such a device first. I'm so bullish on this, but I sold my position too early.