Why Tiny Corp Won't Build Chips or Data Centers | George Hotz

In his 14 October 2024 stream, George Hotz explains where a future tiny chip would fit: only after tinygrad abstracts training jobs from the GPU and reveals which workloads matter. He argues that a software company should focus on its comparative advantage and make a plan for others to build chips and data centers, that the best realistic distribution of AI compute is a free and diverse market, and that restricting jobs to non-Turing-complete computation makes them statically analyzable and easier to optimize. He also explains why he avoids the seductive goal of building AGI himself. These are his October 2024 plans and views. Source:    • George Hotz | Just Chatting | how to actua...   Chapters 0:00 Where a tiny chip fits 0:40 Why franchise instead of building 1:43 Why software looks unsexy 2:36 Make a plan for others to make chips 3:47 A free market for compute 6:42 Where you can change the future 7:52 Why not build AGI himself 8:29 Why limited computation is easier to optimize 10:21 Avoiding seductive traps Clip from George Hotz Original:    • George Hotz | Just Chatting | how to actua...   Twitch:   / georgehotz   #georgehotz #programming #ai #tinygrad Source: George Hotz stream    • George Hotz | Just Chatting | how to actua...   Streamed: 2024-10-14