The 10 Trillion Parameter AI Model With 300 IQ

TL;DR

The video explores the implications of scaling AI models to 10 trillion parameters (two orders of magnitude beyond current frontier models), arguing it could replicate the leap from GPT‑2 to GPT‑3.5 that ignited the 2023 AI boom.

Key insights

• 10‑trillion‑parameter models represent a two‑order‑of‑magnitude increase from today’s ~500B‑parameter models (Llama 3, GPT‑4o, Anthropic), mirroring the jump from ~1B (GPT‑2) to ~170B (GPT‑3.5) that t
• Current state‑of‑the‑art models (with O1 chain‑of‑thought) already rival normal human intelligence; AGI is arguably here for 98% of knowledge worker tasks at 90–98% accuracy.
• Applications of future ASI (200‑300 IQ) may take decades to percolate into everyday life, analogous to Fourier transforms (discovered in the 1800s, only widely applied ~150 years later via color TV, t